{ "generated": "2026-09-28", "generated_by": "analysis/run_analysis.py (Claude Fable 5.1 session, 2026-09-28)", "title": "alzhub.org definitive Alzheimer's correlation analysis: county, state, country and study level", "method_summary": "Pairwise-complete Pearson r with Fisher 95% CI and Student-t p, Spearman rho, and partial correlations by OLS residualisation on the named confounder(s) (scripts/pystats.py, identical maths to the audited baseline). Benjamini-Hochberg FDR is applied within each level and used as the raw significance gate. A result 'survives' if BH q<0.05 and the partial keeps at least half the raw |r| with p<0.05; it 'collapses' if the partial drops below half the raw |r|. Blanks are never zero-filled, imputed or interpolated; suppressed NORC cells stay blank. Twin predictors (|r|>=0.95) are listed per frame. Every variable carries unit, year, source and a measured/modelled tag. Every place-level result carries the ecological-fallacy caveat; every claims-based outcome carries the under-diagnosis caveat.", "baseline": { "file": "data/correlation_results.json (unchanged)", "n_results": 43, "n_analysed": 43, "levels": { "state": 27, "county": 6, "country": 10 } }, "what_the_baseline_establishes": "The 43-result baseline (scripts/run_correlations.py) tests one exposure against one outcome per card on three cross-sections (51 states, ~3,100 counties, ~190 countries) plus a BRFSS join, with the same pairwise-complete Pearson/Spearman/partial machinery. It already shows: (1) two raw counts correlate through population and collapse when population is controlled (S11: r=0.944 to -0.046); (2) the state education gradient in Alzheimer's deaths halves under age+income control (S4: -0.58 to -0.26) and rurality does the same (S9: 0.54 to 0.27); (3) country health spending vs dementia DALYs collapses under GDP (I3: 0.76 to 0.28); (4) two sign flips against the literature (state hearing loss S2, country diabetes I2, plus country air pollution I5) that it correctly reads as measurement rather than risk; (5) at county level the six NORC-based cards find only weak associations, the largest being pct_black vs diagnosed prevalence (D1: r=0.30, partial 0.27 on income). What it does not do: it uses the unadjusted state death rate, the modelled Alzheimer's Association prevalence, the all-age NORC denominator and a country death rate whose own metadata says it is crude despite its column name; it never tests diagnosis intensity, coding practice, race-stratified rates, the exposure time series, the study library, genetics or drug data; and it applies no multiple-comparison control.", "levels": { "county": { "n_coefficients": 129, "n_min": 0, "n_max": 3143, "n_unit": "counties", "classification_counts": { "survives": 25, "collapses": 20, "collapses_extended": 21, "collapses_strict": 6, "negligible": 18, "null": 12, "sign_reverses": 1, "consistency_check": 9, "artefact_documented": 1, "artefact_collapses": 3, "artefact_test_survives": 2, "twin": 8, "uninterpretable": 2, "cannot_compute": 1 } }, "state": { "n_coefficients": 76, "n_min": 49, "n_max": 51, "n_unit": "states (incl. DC)", "classification_counts": { "artefact_documented": 4, "consistency_check": 9, "twin": 2, "uninterpretable": 3, "survives": 9, "null": 36, "collapses_strict": 4, "attenuated": 3, "collapses": 3, "artefact_collapses": 3 } }, "country": { "n_coefficients": 32, "n_min": 51, "n_max": 192, "n_unit": "countries", "classification_counts": { "artefact_documented": 5, "twin": 1, "survives": 1, "collapses_extended": 3, "collapses": 12, "survives_sign_flipped": 3, "null": 3, "sign_reverses": 2, "uninterpretable": 2 } }, "study": { "n_coefficients": 15, "n_min": 2103, "n_max": 10073, "n_unit": "studies", "classification_counts": { "artefact_documented": 2, "survives": 3, "raw_only": 2, "negligible": 4, "null": 1, "uninterpretable": 1, "twin": 1, "consistency_check": 1 } }, "genetics": { "n_coefficients": 4, "n_min": 73, "n_max": 107, "n_unit": "curated variant records", "classification_counts": { "null": 1, "raw_only": 2, "consistency_check": 1 } }, "drugs": { "n_coefficients": 1, "n_min": 9, "n_max": 9, "n_unit": "drug-program pairs", "classification_counts": { "uninterpretable": 1 } } }, "n_coefficients_total": 257, "findings": { "counts": { "survives": 38, "survives_sign_flipped": 3, "collapsed_or_artefact": 81, "uninterpretable": 9, "null_or_negligible": 75, "twins": 12, "checks_and_documented": 39 }, "survivors": [ { "id": "KP-1", "level": "country", "pair": "gdp_per_capita_usd ~ dementia_deaths_per_100k_age_std", "n": 191, "r": 0.8176, "rho": 0.6366, "q": 2.73e-46, "partial_final_r": 0.7013, "controls": [ "pct_65plus", "life_expectancy" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "SP-24", "level": "state", "pair": "medicare_advantage_penetration_pct ~ norc_state65_prevalence_rate_2020", "n": 51, "r": 0.5105, "rho": 0.4325, "q": 0.00052, "partial_final_r": 0.48, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "measured/administrative (derived ratio); blank for Connecticut's 8 legacy counties and small-cell-suppressed counties", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-21", "level": "state", "pair": "nursing_home_beds_per_1000 ~ norc_state65_prevalence_rate_2020", "n": 51, "r": 0.4771, "rho": 0.495, "q": 0.00132, "partial_final_r": 0.4127, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "measured/administrative", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CD-4", "level": "county", "pair": "norc_share_highly_likely_2020 ~ norc65_prevalence_rate_2020", "n": 2579, "r": -0.3223, "rho": -0.2256, "q": 6.45e-63, "partial_final_r": -0.4117, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "x_tag": "measured/administrative (derived ratio)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-1", "level": "state", "pair": "pct_race_black ~ norc_state65_prevalence_rate_2020", "n": 51, "r": 0.6508, "rho": 0.7472, "q": 1.48e-06, "partial_final_r": 0.4073, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd" ], "x_tag": "survey estimate (ACS 1-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-20", "level": "state", "pair": "medicare_percapita_ffs_payment_usd ~ norc_state65_prevalence_rate_2020", "n": 51, "r": 0.4274, "rho": 0.5251, "q": 0.00495, "partial_final_r": 0.3915, "controls": [ "median_household_income", "pct_85plus", "region_south", "pct_race_black" ], "x_tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-22", "level": "state", "pair": "pct_physical_inactivity ~ norc_state65_prevalence_rate_2020", "n": 50, "r": 0.5906, "rho": 0.5962, "q": 3.44e-05, "partial_final_r": 0.3913, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate (BRFSS direct)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-23", "level": "state", "pair": "cms_nh_beds_per_1000_pop ~ norc_state65_prevalence_rate_2020", "n": 50, "r": 0.4172, "rho": 0.3984, "q": 0.00654, "partial_final_r": 0.3662, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "measured/administrative (derived ratio); blank for Connecticut (0 matching CMS rows under either FIPS scheme) and 316 other counties with no Care Compare facility", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-18", "level": "county", "pair": "pct_65plus ~ norc65_prevalence_rate_2020", "n": 2580, "r": -0.2181, "rho": -0.2163, "q": 6.45e-29, "partial_final_r": -0.3478, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS via CHR)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SM-14", "level": "state", "pair": "pct_hearing_loss ~ ts_alz_age_adjusted_death_rate_2022", "n": 51, "r": 0.4325, "rho": 0.3627, "q": 0.00447, "partial_final_r": 0.3471, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "survey estimate (self-report)", "y_tag": "measured/administrative (vital registration)" }, { "id": "CX-19", "level": "county", "pair": "median_age ~ norc65_prevalence_rate_2020", "n": 2572, "r": -0.2188, "rho": -0.2112, "q": 5.34e-29, "partial_final_r": -0.3204, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CH-6", "level": "county", "pair": "pm25_avg_ugm3 ~ norc_highlylikely_prevalence_rate_2020", "n": 3066, "r": 0.4019, "rho": 0.4004, "q": 1.57e-118, "partial_final_r": 0.2946, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (CDC/EPA modelled PM2.5 via CHR)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "ST-3", "level": "state", "pair": "pct_bachelors_plus ~ ts_age_adj_rate_slope_2010_2019", "n": 51, "r": -0.4178, "rho": -0.354, "q": 0.00598, "partial_final_r": -0.2872, "controls": [ "median_household_income" ], "x_tag": "survey estimate (ACS 1-year)", "y_tag": "measured/administrative (derived trend)" }, { "id": "CX-36", "level": "county", "pair": "exp_pm25_modelled_mean_2016_2019 ~ norc65_prevalence_rate_2020", "n": 2571, "r": 0.3991, "rho": 0.4076, "q": 3.33e-98, "partial_final_r": 0.2839, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (statistical downscaler)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SM-4", "level": "state", "pair": "median_household_income ~ ts_alz_age_adjusted_death_rate_2022", "n": 51, "r": -0.4778, "rho": -0.3883, "q": 0.00132, "partial_final_r": -0.2792, "controls": [ "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "survey estimate (ACS 1-year)", "y_tag": "measured/administrative (vital registration)" }, { "id": "CC-5", "level": "county", "pair": "region_south ~ norc65_prevalence_rate_2020", "n": 2580, "r": 0.4257, "rho": 0.461, "q": 3.03e-113, "partial_final_r": 0.2599, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "pct_black" ], "x_tag": "classification (not a measurement)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "LS-9", "level": "study", "pair": "journal_tier ~ log10_rcr", "n": 6818, "r": -0.2603, "rho": -0.2295, "q": 1.48e-105, "partial_final_r": -0.2579, "controls": [ "year" ], "x_tag": "curated label (site rubric)", "y_tag": "modelled metric (iCite normalisation)" }, { "id": "CI-3", "level": "county", "pair": "median_household_income ~ norc65_incidence_rate_2020", "n": 1936, "r": -0.355, "rho": -0.3926, "q": 3.95e-58, "partial_final_r": -0.2346, "controls": [ "pct_85plus", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-56", "level": "county", "pair": "cms_nh_beds_per_1000_pop ~ norc65_prevalence_rate_2020", "n": 2495, "r": 0.2258, "rho": 0.26, "q": 5.92e-30, "partial_final_r": 0.2312, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured/administrative (derived ratio); blank for Connecticut (0 matching CMS rows under either FIPS scheme) and 316 other counties with no Care Compare facility", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-35", "level": "county", "pair": "pm25_avg_ugm3 ~ norc65_prevalence_rate_2020", "n": 2579, "r": 0.3529, "rho": 0.3643, "q": 6.65e-76, "partial_final_r": 0.2307, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (CDC/EPA modelled PM2.5 via CHR)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-30", "level": "county", "pair": "pct_physical_inactivity ~ norc65_prevalence_rate_2020", "n": 2392, "r": 0.4513, "rho": 0.4436, "q": 1.79e-119, "partial_final_r": 0.2274, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (PLACES small-area estimate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-1", "level": "county", "pair": "median_household_income ~ norc65_prevalence_rate_2020", "n": 2580, "r": -0.329, "rho": -0.3779, "q": 1.18e-65, "partial_final_r": -0.2254, "controls": [ "pct_85plus", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-48", "level": "county", "pair": "exp_fullservice_restaurants_per_1000pop_2020 ~ norc65_prevalence_rate_2020", "n": 2494, "r": -0.2992, "rho": -0.3121, "q": 2.46e-52, "partial_final_r": -0.2232, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured/administrative (USDA establishment counts)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CH-8", "level": "county", "pair": "pct_hispanic ~ norc_highlylikely_prevalence_rate_2020", "n": 3070, "r": 0.1722, "rho": 0.1159, "q": 1.23e-21, "partial_final_r": 0.2041, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "LS-11", "level": "study", "pair": "is_review_or_meta ~ log10_rcr", "n": 2103, "r": 0.1861, "rho": 0.2028, "q": 1.46e-17, "partial_final_r": 0.1906, "controls": [ "year" ], "x_tag": "curated label", "y_tag": "modelled metric (iCite normalisation)" }, { "id": "CI-6", "level": "county", "pair": "pm25_avg_ugm3 ~ norc65_incidence_rate_2020", "n": 1936, "r": 0.2817, "rho": 0.2889, "q": 2.68e-36, "partial_final_r": 0.1858, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (CDC/EPA modelled PM2.5 via CHR)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CP-4", "level": "county", "pair": "median_household_income ~ norc_pooled_prevalence_rate_2020_2022", "n": 3125, "r": -0.2727, "rho": -0.3181, "q": 5.34e-54, "partial_final_r": -0.1821, "controls": [ "pct_85plus", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (rounded counts)" }, { "id": "CD-7", "level": "county", "pair": "medicare_advantage_penetration_pct ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.2869, "rho": 0.2867, "q": 1.54e-49, "partial_final_r": 0.1677, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "x_tag": "measured/administrative (derived ratio); blank for Connecticut's 8 legacy counties and small-cell-suppressed counties", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CC-1", "level": "county", "pair": "pct_black ~ norc_white_prevalence_rate_2020", "n": 3085, "r": 0.3202, "rho": 0.4202, "q": 7.2e-74, "partial_final_r": 0.1644, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-20", "level": "county", "pair": "pct_veterans ~ norc65_prevalence_rate_2020", "n": 2572, "r": -0.1731, "rho": -0.1863, "q": 1.51e-18, "partial_final_r": -0.1636, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CD-5", "level": "county", "pair": "norc_ffs_avg_total_cost_usd_2020 ~ norc65_prevalence_rate_2020", "n": 2564, "r": 0.2651, "rho": 0.313, "q": 3.98e-42, "partial_final_r": 0.1633, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "x_tag": "measured/administrative", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-46", "level": "county", "pair": "exp_grocery_stores_per_1000pop_2020 ~ norc65_prevalence_rate_2020", "n": 2078, "r": -0.1382, "rho": -0.1253, "q": 3.42e-10, "partial_final_r": -0.1517, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured/administrative (USDA establishment counts)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CF-4", "level": "county", "pair": "pct_bachelors_plus ~ norc_dementia_mortality_rate_2020", "n": 2997, "r": -0.1765, "rho": -0.1593, "q": 3.62e-22, "partial_final_r": -0.1388, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CF-7", "level": "county", "pair": "pct_smoking ~ norc_dementia_mortality_rate_2020", "n": 2810, "r": 0.1565, "rho": 0.1717, "q": 1.11e-16, "partial_final_r": 0.1335, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (PLACES small-area estimate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CD-3", "level": "county", "pair": "norc_ma_share_of_bene_2020 ~ norc65_prevalence_rate_2020", "n": 2579, "r": 0.148, "rho": 0.166, "q": 6.27e-14, "partial_final_r": 0.1279, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "x_tag": "measured/administrative (derived ratio)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "LS-3", "level": "study", "pair": "log10_n_participants ~ log10_rcr", "n": 2870, "r": 0.1894, "rho": 0.1958, "q": 2.91e-24, "partial_final_r": 0.1238, "controls": [ "year", "journal_tier", "is_review_or_meta" ], "x_tag": "curated/extracted metadata", "y_tag": "modelled metric (iCite normalisation)" }, { "id": "CF-2", "level": "county", "pair": "median_household_income ~ norc_dementia_mortality_rate_2020", "n": 3005, "r": -0.1113, "rho": -0.0936, "q": 1.28e-09, "partial_final_r": -0.1157, "controls": [ "pct_85plus", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CC-3", "level": "county", "pair": "pct_hispanic ~ norc_hispanic_prevalence_rate_2020", "n": 1296, "r": 0.1288, "rho": -0.0322, "q": 4.11e-06, "partial_final_r": 0.1126, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" } ], "survivors_sign_flipped": [ { "id": "KP-7", "level": "country", "pair": "pct_diabetes ~ dementia_deaths_per_100k_age_std", "n": 190, "r": -0.4251, "rho": -0.3159, "q": 2.01e-09, "partial_final_r": -0.2329, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "modelled (NCD-RisC)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "KP-9", "level": "country", "pair": "pct_hypertension ~ dementia_deaths_per_100k_age_std", "n": 190, "r": -0.3186, "rho": -0.1207, "q": 1.1e-05, "partial_final_r": -0.3611, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "modelled (NCD-RisC)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "KP-14", "level": "country", "pair": "pct_air_pollution_exposure ~ dementia_deaths_per_100k_age_std", "n": 186, "r": -0.6411, "rho": -0.4898, "q": 2.5e-22, "partial_final_r": -0.4289, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" } ], "collapsed": [ { "id": "SA-3", "level": "state", "pair": "unpaid_caregivers_count ~ norc_state_prev_cnt_2020", "n": 51, "r": 0.9712, "rho": 0.9762, "q": 1.38e-30, "partial_final_r": 0.2611, "controls": [ "pop_total" ], "x_tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CA-2", "level": "county", "pair": "norc_prev_cnt_2020 ~ exp_primary_care_physicians_count_2023", "n": 3104, "r": 0.9446, "rho": 0.9039, "q": 0.0, "partial_final_r": 0.1363, "controls": [ "population" ], "x_tag": "measured/administrative (claims-based DIAGNOSED cases)", "y_tag": "measured/administrative (AHRF headcount)" }, { "id": "CA-3", "level": "county", "pair": "norc_prev_cnt_2020 ~ exp_hospital_beds_count_2023", "n": 3104, "r": 0.9257, "rho": 0.7488, "q": 0.0, "partial_final_r": 0.2468, "controls": [ "population" ], "x_tag": "measured/administrative (claims-based DIAGNOSED cases)", "y_tag": "measured/administrative (AHRF headcount)" }, { "id": "SA-1", "level": "state", "pair": "geriatricians_count ~ alz_count_65plus", "n": 51, "r": 0.9256, "rho": 0.9341, "q": 4.88e-21, "partial_final_r": 0.439, "controls": [ "pop_total" ], "x_tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)", "y_tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, { "id": "CA-7", "level": "county", "pair": "cms_nh_certified_beds_total ~ norc65_prev_cnt_2020", "n": 2495, "r": 0.9238, "rho": 0.868, "q": 0.0, "partial_final_r": 0.3902, "controls": [ "population" ], "x_tag": "measured/administrative (CMS certification count)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SA-2", "level": "state", "pair": "medicaid_dementia_spend_total_usd ~ ts_dementia_deaths_total_2017", "n": 51, "r": 0.884, "rho": 0.941, "q": 1.28e-16, "partial_final_r": 0.2554, "controls": [ "pop_total" ], "x_tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)", "y_tag": "measured/administrative (count)" }, { "id": "KP-2", "level": "country", "pair": "health_spend_per_capita_usd ~ dementia_deaths_per_100k_age_std", "n": 191, "r": 0.7396, "rho": 0.6589, "q": 1.52e-33, "partial_final_r": 0.0489, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "SP-19", "level": "state", "pair": "gini ~ norc_state65_prevalence_rate_2020", "n": 51, "r": 0.6858, "rho": 0.7033, "q": 1.93e-07, "partial_final_r": 0.2908, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate (ACS)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "KP-23", "level": "country", "pair": "dementia_plan_exists ~ dementia_deaths_per_100k_age_std", "n": 51, "r": 0.6565, "rho": 0.6475, "q": 2.91e-07, "partial_final_r": 0.164, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "measured/administrative", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "KP-3", "level": "country", "pair": "physicians_per_1000 ~ dementia_deaths_per_100k_age_std", "n": 192, "r": 0.6204, "rho": 0.6527, "q": 2.83e-21, "partial_final_r": 0.0159, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "KP-6", "level": "country", "pair": "education_years_mean ~ dementia_deaths_per_100k_age_std", "n": 142, "r": 0.5745, "rho": 0.6234, "q": 1.98e-13, "partial_final_r": -0.1621, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "SM-1", "level": "state", "pair": "pct_bachelors_plus ~ ts_alz_age_adjusted_death_rate_2022", "n": 51, "r": -0.5598, "rho": -0.4425, "q": 9.78e-05, "partial_final_r": -0.2014, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "survey estimate (ACS 1-year)", "y_tag": "measured/administrative (vital registration)" }, { "id": "SP-3", "level": "state", "pair": "pct_below_poverty ~ norc_state65_prevalence_rate_2020", "n": 51, "r": 0.5584, "rho": 0.5642, "q": 9.78e-05, "partial_final_r": 0.1503, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate (ACS 1-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-7", "level": "state", "pair": "pct_diabetes ~ norc_state65_prevalence_rate_2020", "n": 50, "r": 0.5572, "rho": 0.5348, "q": 0.000118, "partial_final_r": 0.1742, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate (BRFSS direct)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "KP-17", "level": "country", "pair": "hospital_beds_per_1000 ~ dementia_deaths_per_100k_age_std", "n": 191, "r": 0.5364, "rho": 0.5384, "q": 3.58e-15, "partial_final_r": -0.0465, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "SP-9", "level": "state", "pair": "pct_hypertension ~ norc_state65_prevalence_rate_2020", "n": 49, "r": 0.5348, "rho": 0.4418, "q": 0.000318, "partial_final_r": 0.1482, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate (BRFSS direct)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SM-8", "level": "state", "pair": "primary_care_physicians_per_100k ~ ts_alz_age_adjusted_death_rate_2022", "n": 51, "r": -0.5047, "rho": -0.4283, "q": 0.000604, "partial_final_r": -0.179, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "survey estimate / administrative compilation", "y_tag": "measured/administrative (vital registration)" }, { "id": "SM-11", "level": "state", "pair": "pct_obesity ~ ts_alz_age_adjusted_death_rate_2022", "n": 50, "r": 0.5035, "rho": 0.4221, "q": 0.000698, "partial_final_r": 0.297, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "survey estimate (BRFSS direct)", "y_tag": "measured/administrative (vital registration)" }, { "id": "KP-16", "level": "country", "pair": "uhc_service_coverage_index ~ dementia_deaths_per_100k_age_std", "n": 192, "r": 0.4709, "rho": 0.6543, "q": 1.3e-11, "partial_final_r": -0.1334, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "modelled/derived", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "KP-4", "level": "country", "pair": "pct_urban ~ dementia_deaths_per_100k_age_std", "n": 192, "r": 0.4319, "rho": 0.4377, "q": 8.77e-10, "partial_final_r": 0.012, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "CX-33", "level": "county", "pair": "pct_fair_poor_health ~ norc65_prevalence_rate_2020", "n": 2392, "r": 0.4275, "rho": 0.44, "q": 4.17e-106, "partial_final_r": 0.1396, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (PLACES small-area estimate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-18", "level": "state", "pair": "pm25_avg_ugm3 ~ norc_state65_prevalence_rate_2020", "n": 51, "r": 0.423, "rho": 0.3919, "q": 0.00537, "partial_final_r": 0.0978, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate / administrative compilation", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-6", "level": "county", "pair": "exp_svi_theme1_socioeconomic_percentile ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.4162, "rho": 0.4448, "q": 1.69e-107, "partial_final_r": 0.1087, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "derived index (from ACS survey estimates)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-3", "level": "county", "pair": "pct_below_poverty ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.4119, "rho": 0.4206, "q": 4.11e-105, "partial_final_r": 0.104, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-7", "level": "county", "pair": "exp_svi_overall_percentile ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.4115, "rho": 0.4392, "q": 6.26e-105, "partial_final_r": 0.1499, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "derived index (from ACS survey estimates)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CI-5", "level": "county", "pair": "pct_diabetes ~ norc65_incidence_rate_2020", "n": 1787, "r": 0.4034, "rho": 0.4176, "q": 2.56e-70, "partial_final_r": 0.0493, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (PLACES small-area estimate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SM-2", "level": "state", "pair": "pct_rural ~ ts_alz_age_adjusted_death_rate_2022", "n": 51, "r": 0.3986, "rho": 0.3667, "q": 0.00924, "partial_final_r": 0.1912, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "survey estimate (ACS)", "y_tag": "measured/administrative (vital registration)" }, { "id": "KP-20", "level": "country", "pair": "fertility_rate ~ dementia_deaths_per_100k_age_std", "n": 192, "r": -0.3921, "rho": -0.6785, "q": 3.6e-08, "partial_final_r": 0.2364, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "CI-7", "level": "county", "pair": "pct_below_poverty ~ norc65_incidence_rate_2020", "n": 1928, "r": 0.3886, "rho": 0.3942, "q": 5.8e-70, "partial_final_r": 0.0806, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-34", "level": "county", "pair": "pct_65plus_with_disability ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.3876, "rho": 0.4071, "q": 2.89e-92, "partial_final_r": 0.1915, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CI-9", "level": "county", "pair": "exp_svi_overall_percentile ~ norc65_incidence_rate_2020", "n": 1928, "r": 0.3823, "rho": 0.3978, "q": 1.41e-67, "partial_final_r": 0.1263, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "derived index (from ACS survey estimates)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-26", "level": "county", "pair": "pct_diabetes ~ norc65_prevalence_rate_2020", "n": 2392, "r": 0.3784, "rho": 0.3891, "q": 1.2e-81, "partial_final_r": -0.0114, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (PLACES small-area estimate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CP-1", "level": "county", "pair": "pct_black ~ norc_pooled_prevalence_rate_2020_2022", "n": 3117, "r": 0.3695, "rho": 0.4957, "q": 1.01e-100, "partial_final_r": 0.1124, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (rounded counts)" }, { "id": "KP-15", "level": "country", "pair": "pm25_avg_ugm3 ~ dementia_deaths_per_100k_age_std", "n": 192, "r": -0.3678, "rho": -0.4806, "q": 2.81e-07, "partial_final_r": 0.0411, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "modelled/derived", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "KP-5", "level": "country", "pair": "literacy_rate ~ dementia_deaths_per_100k_age_std", "n": 162, "r": 0.3674, "rho": 0.4707, "q": 2.48e-06, "partial_final_r": -0.1747, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "SM-13", "level": "state", "pair": "pct_hypertension ~ ts_alz_age_adjusted_death_rate_2022", "n": 49, "r": 0.3618, "rho": 0.2217, "q": 0.0244, "partial_final_r": 0.1148, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "survey estimate (BRFSS direct)", "y_tag": "measured/administrative (vital registration)" }, { "id": "CX-11", "level": "county", "pair": "pct_black ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.3596, "rho": 0.4471, "q": 9.45e-79, "partial_final_r": 0.1177, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-28", "level": "county", "pair": "pct_hypertension ~ norc65_prevalence_rate_2020", "n": 2392, "r": 0.3575, "rho": 0.3642, "q": 1.85e-72, "partial_final_r": -0.0111, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (PLACES small-area estimate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-27", "level": "county", "pair": "pct_obesity ~ norc65_prevalence_rate_2020", "n": 2392, "r": 0.3461, "rho": 0.3431, "q": 1.01e-67, "partial_final_r": 0.144, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (PLACES small-area estimate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "KP-21", "level": "country", "pair": "pct_safe_water_access ~ dementia_deaths_per_100k_age_std", "n": 191, "r": 0.3363, "rho": 0.5957, "q": 3.05e-06, "partial_final_r": -0.133, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "CH-1", "level": "county", "pair": "pct_black ~ norc_highlylikely_prevalence_rate_2020", "n": 3070, "r": 0.3339, "rho": 0.4533, "q": 3.43e-80, "partial_final_r": 0.1289, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "KP-19", "level": "country", "pair": "gini ~ dementia_deaths_per_100k_age_std", "n": 169, "r": -0.3329, "rho": -0.3613, "q": 1.32e-05, "partial_final_r": -0.0326, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "CX-29", "level": "county", "pair": "pct_smoking ~ norc65_prevalence_rate_2020", "n": 2392, "r": 0.3253, "rho": 0.3451, "q": 1.35e-59, "partial_final_r": 0.1086, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (PLACES small-area estimate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-25", "level": "county", "pair": "exp_svi_theme2_household_percentile ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.323, "rho": 0.3373, "q": 5.15e-63, "partial_final_r": 0.1657, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "derived index (from ACS survey estimates)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "KP-18", "level": "country", "pair": "out_of_pocket_pct_health ~ dementia_deaths_per_100k_age_std", "n": 191, "r": -0.314, "rho": -0.3164, "q": 1.32e-05, "partial_final_r": -0.0136, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "CH-9", "level": "county", "pair": "exp_svi_overall_percentile ~ norc_highlylikely_prevalence_rate_2020", "n": 3070, "r": 0.3005, "rho": 0.3168, "q": 1.46e-64, "partial_final_r": 0.0772, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "derived index (from ACS survey estimates)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CI-2", "level": "county", "pair": "pct_bachelors_plus ~ norc65_incidence_rate_2020", "n": 1928, "r": -0.3003, "rho": -0.3226, "q": 4.16e-41, "partial_final_r": -0.0881, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-5", "level": "county", "pair": "pct_uninsured ~ norc65_prevalence_rate_2020", "n": 2580, "r": 0.3002, "rho": 0.3322, "q": 1.92e-54, "partial_final_r": 0.1157, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "administrative compilation (CHR pass-through)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CD-6", "level": "county", "pair": "pct_uninsured ~ norc65_prevalence_rate_2020", "n": 2580, "r": 0.3002, "rho": 0.3322, "q": 1.92e-54, "partial_final_r": 0.0551, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "x_tag": "administrative compilation (CHR pass-through)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CI-1", "level": "county", "pair": "pct_black ~ norc65_incidence_rate_2020", "n": 1928, "r": 0.2995, "rho": 0.3565, "q": 7.03e-41, "partial_final_r": 0.0951, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-45", "level": "county", "pair": "exp_dollar_stores_per_1000pop_2020 ~ norc65_prevalence_rate_2020", "n": 2580, "r": 0.2983, "rho": 0.3061, "q": 9.37e-54, "partial_final_r": 0.054, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured/administrative (USDA establishment counts)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "KP-12", "level": "country", "pair": "pct_alcohol_use_disorder ~ dementia_deaths_per_100k_age_std", "n": 188, "r": 0.2942, "rho": 0.3063, "q": 5.37e-05, "partial_final_r": -0.0962, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "measured/administrative", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "CX-9", "level": "county", "pair": "pct_broadband ~ norc65_prevalence_rate_2020", "n": 2580, "r": -0.2683, "rho": -0.2554, "q": 2.07e-43, "partial_final_r": -0.0312, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CH-5", "level": "county", "pair": "pct_diabetes ~ norc_highlylikely_prevalence_rate_2020", "n": 2883, "r": 0.2678, "rho": 0.2579, "q": 3.77e-48, "partial_final_r": -0.0576, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (PLACES small-area estimate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-13", "level": "county", "pair": "pct_white_nh ~ norc65_prevalence_rate_2020", "n": 2572, "r": -0.2549, "rho": -0.2605, "q": 4.42e-39, "partial_final_r": -0.072, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-32", "level": "county", "pair": "pct_binge_drinking ~ norc65_prevalence_rate_2020", "n": 2392, "r": -0.2435, "rho": -0.2609, "q": 2.67e-33, "partial_final_r": 0.1003, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (PLACES small-area estimate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-2", "level": "county", "pair": "pct_bachelors_plus ~ norc65_prevalence_rate_2020", "n": 2572, "r": -0.2378, "rho": -0.2759, "q": 4.59e-34, "partial_final_r": -0.0859, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-21", "level": "county", "pair": "exp_svi_theme3_minority_status_percentile ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.231, "rho": 0.2651, "q": 3.47e-32, "partial_final_r": 0.0633, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "derived index (from ACS survey estimates)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CD-1", "level": "county", "pair": "pct_checkup_annual ~ norc65_prevalence_rate_2020", "n": 2392, "r": 0.2292, "rho": 0.2272, "q": 1.27e-29, "partial_final_r": 0.1116, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "x_tag": "modelled (PLACES small-area estimate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-4", "level": "county", "pair": "unemployment_rate ~ norc65_prevalence_rate_2020", "n": 2580, "r": 0.2273, "rho": 0.2342, "q": 2.82e-31, "partial_final_r": -0.0059, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-50", "level": "county", "pair": "exp_snap_authorized_stores_per_1000pop_2023 ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.2265, "rho": 0.2481, "q": 5.45e-31, "partial_final_r": -0.0779, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured/administrative (USDA establishment counts)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-8", "level": "county", "pair": "exp_snap_benefits_per_capita_2022 ~ norc65_prevalence_rate_2020", "n": 2533, "r": 0.2263, "rho": 0.2597, "q": 1.71e-30, "partial_final_r": -0.0995, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured/administrative (USDA establishment counts)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-10", "level": "county", "pair": "pct_no_vehicle ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.225, "rho": 0.2723, "q": 1.34e-30, "partial_final_r": 0.0696, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "KP-22", "level": "country", "pair": "measles_immunization_pct ~ dementia_deaths_per_100k_age_std", "n": 192, "r": 0.2209, "rho": 0.3203, "q": 0.00248, "partial_final_r": -0.0401, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "CP-2", "level": "county", "pair": "pct_bachelors_plus ~ norc_pooled_prevalence_rate_2020_2022", "n": 3117, "r": -0.2163, "rho": -0.2545, "q": 5.6e-34, "partial_final_r": -0.1159, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (rounded counts)" }, { "id": "CH-7", "level": "county", "pair": "pct_below_poverty ~ norc_highlylikely_prevalence_rate_2020", "n": 3070, "r": 0.2103, "rho": 0.2298, "q": 1.02e-31, "partial_final_r": 0.0005, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CH-10", "level": "county", "pair": "pct_checkup_annual ~ norc_highlylikely_prevalence_rate_2020", "n": 2883, "r": 0.1918, "rho": 0.1781, "q": 4.68e-25, "partial_final_r": -0.0537, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020" ], "x_tag": "modelled (PLACES small-area estimate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "KP-10", "level": "country", "pair": "pct_smoking ~ dementia_deaths_per_100k_age_std", "n": 165, "r": 0.1864, "rho": 0.4003, "q": 0.0189, "partial_final_r": -0.1836, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "modelled/derived", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "CI-10", "level": "county", "pair": "pct_checkup_annual ~ norc65_incidence_rate_2020", "n": 1787, "r": 0.1853, "rho": 0.1858, "q": 4.41e-15, "partial_final_r": -0.006, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020" ], "x_tag": "modelled (PLACES small-area estimate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-22", "level": "county", "pair": "pct_living_alone_65plus ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.1812, "rho": 0.1842, "q": 3.33e-20, "partial_final_r": 0.1186, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CA-6", "level": "county", "pair": "pct_living_alone_65plus ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.1812, "rho": 0.1842, "q": 3.33e-20, "partial_final_r": 0.1251, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-37", "level": "county", "pair": "exp_pm25_monitor_mean_2016_2019 ~ norc65_prevalence_rate_2020", "n": 559, "r": 0.1752, "rho": 0.2018, "q": 3.78e-05, "partial_final_r": 0.0527, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured (physical monitor)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CF-1", "level": "county", "pair": "pct_85plus ~ norc_dementia_mortality_rate_2020", "n": 2997, "r": 0.1704, "rho": 0.1659, "q": 9.94e-21, "partial_final_r": 0.0627, "controls": [ "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CH-3", "level": "county", "pair": "median_household_income ~ norc_highlylikely_prevalence_rate_2020", "n": 3078, "r": -0.1627, "rho": -0.1986, "q": 1.71e-19, "partial_final_r": -0.0453, "controls": [ "pct_85plus", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "KP-13", "level": "country", "pair": "mean_bmi ~ dementia_deaths_per_100k_age_std", "n": 188, "r": 0.1587, "rho": 0.2609, "q": 0.0328, "partial_final_r": -0.1146, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "modelled (NCD-RisC)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "CX-44", "level": "county", "pair": "exp_pct_low_income_low_access_2019 ~ norc65_prevalence_rate_2020", "n": 2580, "r": 0.1516, "rho": 0.1558, "q": 1.49e-14, "partial_final_r": 0.0548, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured/administrative (USDA establishment counts)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-49", "level": "county", "pair": "exp_convenience_stores_per_1000pop_2020 ~ norc65_prevalence_rate_2020", "n": 2542, "r": 0.139, "rho": 0.1643, "q": 2.89e-12, "partial_final_r": -0.114, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured/administrative (USDA establishment counts)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CP-3", "level": "county", "pair": "pct_85plus ~ norc_pooled_prevalence_rate_2020_2022", "n": 3117, "r": -0.1239, "rho": -0.1504, "q": 5.7e-12, "partial_final_r": 0.0058, "controls": [ "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (rounded counts)" }, { "id": "CH-2", "level": "county", "pair": "pct_bachelors_plus ~ norc_highlylikely_prevalence_rate_2020", "n": 3070, "r": -0.1231, "rho": -0.121, "q": 1.1e-11, "partial_final_r": -0.0949, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CF-5", "level": "county", "pair": "pm25_avg_ugm3 ~ norc_dementia_mortality_rate_2020", "n": 2997, "r": -0.1223, "rho": -0.0921, "q": 2.67e-11, "partial_final_r": 0.0118, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (CDC/EPA modelled PM2.5 via CHR)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-51", "level": "county", "pair": "hospital_beds_per_1000 ~ norc65_prevalence_rate_2020", "n": 2146, "r": 0.1108, "rho": 0.2337, "q": 3.46e-07, "partial_final_r": 0.0305, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured/administrative", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" } ], "uninterpretable": [ { "id": "CU-1", "level": "county", "pair": "norc65_prevalence_rate_2020 ~ norc_dementia_mortality_rate_2020", "n": 2576, "r": -0.2225, "rho": -0.1953, "q": 5.39e-30, "partial_final_r": null, "controls": [], "x_tag": "measured/administrative (claims-based DIAGNOSED cases)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CU-2", "level": "county", "pair": "pct_65plus_with_disability ~ norc_dementia_mortality_rate_2020", "n": 2997, "r": -0.0035, "rho": 0.0036, "q": 0.863, "partial_final_r": null, "controls": [], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SU-1", "level": "state", "pair": "pct_race_black ~ alz_prevalence_65plus_pct", "n": 51, "r": 0.7668, "rho": 0.7869, "q": 4.53e-10, "partial_final_r": 0.8023, "controls": [ "pct_85plus" ], "x_tag": "survey estimate (ACS 1-year)", "y_tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, { "id": "SU-2", "level": "state", "pair": "pct_race_hispanic ~ alz_prevalence_65plus_pct", "n": 51, "r": 0.2385, "rho": 0.339, "q": 0.171, "partial_final_r": 0.2389, "controls": [ "pct_85plus" ], "x_tag": "survey estimate (BRFSS direct)", "y_tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, { "id": "SU-3", "level": "state", "pair": "pct_85plus ~ alz_prevalence_65plus_pct", "n": 51, "r": -0.005, "rho": 0.0655, "q": 0.982, "partial_final_r": null, "controls": [], "x_tag": "survey estimate (ACS 1-year)", "y_tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, { "id": "KU-1", "level": "country", "pair": "life_expectancy ~ dementia_deaths_per_100k_age_std", "n": 192, "r": 0.563, "rho": 0.6755, "q": 5.89e-17, "partial_final_r": 0.0727, "controls": [ "pct_65plus" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "KU-2", "level": "country", "pair": "life_expectancy ~ dementia_daly_per_100k", "n": 192, "r": 0.6395, "rho": 0.7617, "q": 8.02e-23, "partial_final_r": 0.0601, "controls": [ "pct_65plus" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "LU-1", "level": "study", "pair": "evidence_ordinal ~ site_score", "n": 10073, "r": 0.8227, "rho": 0.7935, "q": 0.0, "partial_final_r": null, "controls": [], "x_tag": "curated label (editorial judgement)", "y_tag": "derived index (site rubric; built FROM evidence/design/journal/year/size fields)" }, { "id": "LD-1", "level": "drugs", "pair": "total_claims ~ faers_reports_generic_name", "n": 9, "r": 0.8475, "rho": 0.8488, "q": 0.00391, "partial_final_r": null, "controls": [], "x_tag": "measured/administrative", "y_tag": "measured count (spontaneous reports, no denominator)" } ], "null_or_negligible": [ { "id": "CX-12", "level": "county", "pair": "pct_hispanic ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.0478, "rho": -0.0251, "q": 0.0172, "partial_final_r": 0.1323, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-14", "level": "county", "pair": "pct_asian ~ norc65_prevalence_rate_2020", "n": 2572, "r": -0.0651, "rho": -0.0782, "q": 0.00112, "partial_final_r": 0.0671, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-15", "level": "county", "pair": "pct_foreign_born ~ norc65_prevalence_rate_2020", "n": 2572, "r": -0.0013, "rho": -0.0503, "q": 0.956, "partial_final_r": 0.1303, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-16", "level": "county", "pair": "pct_female ~ norc65_prevalence_rate_2020", "n": 2580, "r": 0.1289, "rho": 0.2809, "q": 7.04e-11, "partial_final_r": 0.0662, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-17", "level": "county", "pair": "pct_85plus ~ norc65_prevalence_rate_2020", "n": 2572, "r": -0.1316, "rho": -0.1501, "q": 2.93e-11, "partial_final_r": -0.0667, "controls": [ "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-23", "level": "county", "pair": "pct_single_person_households ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.0719, "rho": 0.0678, "q": 0.000308, "partial_final_r": -0.0961, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-24", "level": "county", "pair": "social_association_index ~ norc65_prevalence_rate_2020", "n": 2580, "r": -0.0304, "rho": -0.0214, "q": 0.134, "partial_final_r": 0.0531, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "administrative compilation (CHR pass-through)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-31", "level": "county", "pair": "pct_depression ~ norc65_prevalence_rate_2020", "n": 2392, "r": -0.0086, "rho": 0.0041, "q": 0.703, "partial_final_r": -0.076, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled (PLACES small-area estimate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-38", "level": "county", "pair": "exp_ozone_monitor_mean_2016_2019 ~ norc65_prevalence_rate_2020", "n": 720, "r": -0.1726, "rho": -0.1363, "q": 4.04e-06, "partial_final_r": -0.0863, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured (physical monitor)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-39", "level": "county", "pair": "drinking_water_violations ~ norc65_prevalence_rate_2020", "n": 2540, "r": -0.0269, "rho": -0.0155, "q": 0.188, "partial_final_r": 0.0127, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "administrative compilation (CHR pass-through)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-40", "level": "county", "pair": "population_density_per_sqmi ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.0599, "rho": 0.14, "q": 0.00276, "partial_final_r": 0.0986, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled/derived", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-41", "level": "county", "pair": "county_type_metro ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.0009, "rho": 0.0295, "q": 0.964, "partial_final_r": 0.0675, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled/derived", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-42", "level": "county", "pair": "exp_pct_seniors_low_access_2019 ~ norc65_prevalence_rate_2020", "n": 2580, "r": 0.0092, "rho": 0.0182, "q": 0.68, "partial_final_r": -0.1794, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured/administrative (USDA establishment counts)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-43", "level": "county", "pair": "exp_pct_low_access_to_store_2019 ~ norc65_prevalence_rate_2020", "n": 2580, "r": -0.0061, "rho": 0.0028, "q": 0.774, "partial_final_r": 0.0475, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured/administrative (USDA establishment counts)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-47", "level": "county", "pair": "exp_fastfood_restaurants_per_1000pop_2020 ~ norc65_prevalence_rate_2020", "n": 2465, "r": 0.1053, "rho": 0.1423, "q": 2.14e-07, "partial_final_r": 0.0851, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured/administrative (USDA establishment counts)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-52", "level": "county", "pair": "exp_hospital_beds_per_1000pop_2023 ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.1236, "rho": 0.1579, "q": 4.4e-10, "partial_final_r": 0.0873, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "modelled/derived", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-53", "level": "county", "pair": "mental_health_providers_per_100k ~ norc65_prevalence_rate_2020", "n": 2552, "r": -0.098, "rho": -0.1385, "q": 9.05e-07, "partial_final_r": -0.0441, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "administrative compilation (CHR pass-through)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CD-2", "level": "county", "pair": "primary_care_physicians_per_100k ~ norc65_prevalence_rate_2020", "n": 2529, "r": -0.0854, "rho": -0.1077, "q": 2.1e-05, "partial_final_r": 0.08, "controls": [ "pct_85plus", "median_household_income", "region_south" ], "x_tag": "measured/administrative (AHRF headcount)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CH-4", "level": "county", "pair": "pct_85plus ~ norc_highlylikely_prevalence_rate_2020", "n": 3070, "r": -0.0077, "rho": -0.0654, "q": 0.703, "partial_final_r": 0.1502, "controls": [ "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CI-4", "level": "county", "pair": "pct_85plus ~ norc65_incidence_rate_2020", "n": 1928, "r": -0.0874, "rho": -0.0986, "q": 0.000145, "partial_final_r": -0.0239, "controls": [ "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CI-8", "level": "county", "pair": "pct_hispanic ~ norc65_incidence_rate_2020", "n": 1928, "r": 0.0481, "rho": -0.0587, "q": 0.0388, "partial_final_r": 0.1363, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CH-11", "level": "county", "pair": "norc_ma_share_of_bene_2020 ~ norc_highlylikely_prevalence_rate_2020", "n": 3057, "r": 0.1615, "rho": 0.1439, "q": 4.15e-19, "partial_final_r": 0.0992, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "pct_checkup_annual" ], "x_tag": "measured/administrative (derived ratio)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CI-11", "level": "county", "pair": "norc_ma_share_of_bene_2020 ~ norc65_incidence_rate_2020", "n": 1935, "r": 0.0945, "rho": 0.098, "q": 3.78e-05, "partial_final_r": 0.0882, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "pct_checkup_annual" ], "x_tag": "measured/administrative (derived ratio)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CF-3", "level": "county", "pair": "pct_black ~ norc_dementia_mortality_rate_2020", "n": 2997, "r": -0.0852, "rho": -0.1099, "q": 3.87e-06, "partial_final_r": -0.0195, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CF-6", "level": "county", "pair": "hospital_beds_per_1000 ~ norc_dementia_mortality_rate_2020", "n": 2419, "r": 0.1084, "rho": 0.1572, "q": 1.2e-07, "partial_final_r": 0.0951, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured/administrative", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CC-2", "level": "county", "pair": "pct_black ~ norc_black_prevalence_rate_2020", "n": 1539, "r": 0.0407, "rho": 0.1026, "q": 0.122, "partial_final_r": -0.0276, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "x_tag": "survey estimate (ACS 5-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-54", "level": "county", "pair": "pct_low_access_pop_1mi ~ norc65_prevalence_rate_2020", "n": 2580, "r": 0.0294, "rho": 0.0345, "q": 0.147, "partial_final_r": -0.1528, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured/administrative (USDA tract-distance flag, population-weighted county aggregate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CX-55", "level": "county", "pair": "pct_seniors_low_access_1mi ~ norc65_prevalence_rate_2020", "n": 2580, "r": -0.0537, "rho": -0.0461, "q": 0.00729, "partial_final_r": -0.2167, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "x_tag": "measured/administrative (USDA tract-distance flag, population-weighted county aggregate)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CD-8", "level": "county", "pair": "naics62_health_socasst_emplvl_per_1000pop ~ norc65_prevalence_rate_2020", "n": 2572, "r": 0.0336, "rho": 0.0119, "q": 0.098, "partial_final_r": 0.156, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "x_tag": "measured/administrative (derived ratio; inherits the suppression floor of the employment count)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "CD-9", "level": "county", "pair": "naics623_nursing_residential_care_emplvl_per_1000pop ~ norc65_prevalence_rate_2020", "n": 2547, "r": -0.0067, "rho": 0.007, "q": 0.761, "partial_final_r": 0.128, "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "x_tag": "measured/administrative (derived ratio; inherits the suppression floor of the employment count)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-2", "level": "state", "pair": "pct_race_hispanic ~ norc_state65_prevalence_rate_2020", "n": 51, "r": 0.0438, "rho": 0.0675, "q": 0.902, "partial_final_r": 0.0437, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate (BRFSS direct)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-4", "level": "state", "pair": "median_household_income ~ norc_state65_prevalence_rate_2020", "n": 51, "r": -0.2695, "rho": -0.3062, "q": 0.116, "partial_final_r": -0.3706, "controls": [ "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate (ACS 1-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-5", "level": "state", "pair": "pct_bachelors_plus ~ norc_state65_prevalence_rate_2020", "n": 51, "r": -0.0875, "rho": -0.1583, "q": 0.685, "partial_final_r": 0.0663, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate (ACS 1-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-6", "level": "state", "pair": "pct_rural ~ norc_state65_prevalence_rate_2020", "n": 51, "r": -0.1491, "rho": -0.1798, "q": 0.433, "partial_final_r": -0.1908, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate (ACS)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-8", "level": "state", "pair": "pct_obesity ~ norc_state65_prevalence_rate_2020", "n": 50, "r": 0.2289, "rho": 0.2622, "q": 0.19, "partial_final_r": -0.0498, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate (BRFSS direct)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-10", "level": "state", "pair": "pct_smoking ~ norc_state65_prevalence_rate_2020", "n": 50, "r": 0.1804, "rho": 0.1538, "q": 0.332, "partial_final_r": -0.0167, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate (BRFSS direct)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-11", "level": "state", "pair": "pct_hearing_loss ~ norc_state65_prevalence_rate_2020", "n": 51, "r": -0.2561, "rho": -0.2329, "q": 0.139, "partial_final_r": 0.0552, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate (self-report)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-12", "level": "state", "pair": "pct_depression ~ norc_state65_prevalence_rate_2020", "n": 51, "r": -0.0709, "rho": -0.0318, "q": 0.774, "partial_final_r": -0.0111, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate / administrative compilation", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-13", "level": "state", "pair": "pct_uninsured ~ norc_state65_prevalence_rate_2020", "n": 51, "r": 0.1152, "rho": 0.0523, "q": 0.542, "partial_final_r": -0.0708, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate (BRFSS direct)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-14", "level": "state", "pair": "primary_care_physicians_per_100k ~ norc_state65_prevalence_rate_2020", "n": 51, "r": -0.0647, "rho": -0.1192, "q": 0.787, "partial_final_r": -0.112, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate / administrative compilation", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-15", "level": "state", "pair": "neurologists_per_100k ~ norc_state65_prevalence_rate_2020", "n": 51, "r": 0.1317, "rho": 0.105, "q": 0.487, "partial_final_r": 0.1127, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "measured/administrative", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-16", "level": "state", "pair": "pct_65plus ~ norc_state65_prevalence_rate_2020", "n": 51, "r": -0.1227, "rho": -0.1205, "q": 0.521, "partial_final_r": -0.325, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate (ACS 1-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-17", "level": "state", "pair": "pct_85plus ~ norc_state65_prevalence_rate_2020", "n": 51, "r": -0.0244, "rho": -0.1073, "q": 0.939, "partial_final_r": 0.2856, "controls": [ "median_household_income", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "survey estimate (ACS 1-year)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SM-3", "level": "state", "pair": "pm25_avg_ugm3 ~ ts_alz_age_adjusted_death_rate_2022", "n": 51, "r": 0.0667, "rho": 0.1033, "q": 0.787, "partial_final_r": -0.0243, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "survey estimate / administrative compilation", "y_tag": "measured/administrative (vital registration)" }, { "id": "SM-5", "level": "state", "pair": "pct_race_black ~ ts_alz_age_adjusted_death_rate_2022", "n": 51, "r": 0.0032, "rho": 0.0138, "q": 0.982, "partial_final_r": -0.1722, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "x_tag": "survey estimate (ACS 1-year)", "y_tag": "measured/administrative (vital registration)" }, { "id": "SM-6", "level": "state", "pair": "pct_race_hispanic ~ ts_alz_age_adjusted_death_rate_2022", "n": 51, "r": -0.1444, "rho": -0.1814, "q": 0.447, "partial_final_r": -0.1993, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "survey estimate (BRFSS direct)", "y_tag": "measured/administrative (vital registration)" }, { "id": "SM-7", "level": "state", "pair": "gini ~ ts_alz_age_adjusted_death_rate_2022", "n": 51, "r": -0.2346, "rho": -0.1439, "q": 0.177, "partial_final_r": -0.351, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "survey estimate (ACS)", "y_tag": "measured/administrative (vital registration)" }, { "id": "SM-9", "level": "state", "pair": "geriatricians_count ~ ts_alz_age_adjusted_death_rate_2022", "n": 51, "r": -0.2685, "rho": -0.2173, "q": 0.116, "partial_final_r": -0.2169, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)", "y_tag": "measured/administrative (vital registration)" }, { "id": "SM-10", "level": "state", "pair": "pct_diabetes ~ ts_alz_age_adjusted_death_rate_2022", "n": 50, "r": 0.2349, "rho": 0.1977, "q": 0.179, "partial_final_r": -0.069, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "survey estimate (BRFSS direct)", "y_tag": "measured/administrative (vital registration)" }, { "id": "SM-12", "level": "state", "pair": "pct_smoking ~ ts_alz_age_adjusted_death_rate_2022", "n": 50, "r": 0.2562, "rho": 0.2729, "q": 0.141, "partial_final_r": -0.1115, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "survey estimate (BRFSS direct)", "y_tag": "measured/administrative (vital registration)" }, { "id": "SM-15", "level": "state", "pair": "nursing_home_beds_per_1000 ~ ts_alz_age_adjusted_death_rate_2022", "n": 51, "r": 0.0071, "rho": -0.0056, "q": 0.982, "partial_final_r": -0.0837, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "measured/administrative", "y_tag": "measured/administrative (vital registration)" }, { "id": "SM-16", "level": "state", "pair": "norc_state65_prevalence_rate_2020 ~ ts_alz_age_adjusted_death_rate_2022", "n": 51, "r": -0.0266, "rho": -0.0444, "q": 0.939, "partial_final_r": -0.0711, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "measured/administrative (claims-based DIAGNOSED cases)", "y_tag": "measured/administrative (vital registration)" }, { "id": "SM-17", "level": "state", "pair": "pct_65plus ~ ts_alz_age_adjusted_death_rate_2022", "n": 51, "r": -0.2152, "rho": -0.3092, "q": 0.218, "partial_final_r": -0.2684, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "survey estimate (ACS 1-year)", "y_tag": "measured/administrative (vital registration)" }, { "id": "SM-18", "level": "state", "pair": "region_south ~ ts_alz_age_adjusted_death_rate_2022", "n": 51, "r": 0.1849, "rho": 0.2247, "q": 0.321, "partial_final_r": 0.1508, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "pct_race_black" ], "x_tag": "classification (not a measurement)", "y_tag": "measured/administrative (vital registration)" }, { "id": "ST-1", "level": "state", "pair": "ts_age_adj_rate_2010 ~ ts_age_adj_rate_slope_2010_2019", "n": 51, "r": -0.0208, "rho": 0.0152, "q": 0.947, "partial_final_r": null, "controls": [], "x_tag": "measured/administrative (vital registration)", "y_tag": "measured/administrative (derived trend)" }, { "id": "ST-4", "level": "state", "pair": "pct_race_black ~ ts_age_adj_rate_change_2019_2020_pct", "n": 51, "r": 0.2995, "rho": 0.3096, "q": 0.071, "partial_final_r": 0.2526, "controls": [ "median_household_income", "region_south" ], "x_tag": "survey estimate (ACS 1-year)", "y_tag": "measured/administrative (derived change)" }, { "id": "ST-5", "level": "state", "pair": "norc_state65_prevalence_rate_2020 ~ ts_age_adj_rate_change_2019_2020_pct", "n": 51, "r": 0.2404, "rho": 0.2996, "q": 0.169, "partial_final_r": 0.2175, "controls": [ "median_household_income" ], "x_tag": "measured/administrative (claims-based DIAGNOSED cases)", "y_tag": "measured/administrative (derived change)" }, { "id": "ST-6", "level": "state", "pair": "nursing_home_beds_per_1000 ~ ts_age_adj_rate_change_2019_2020_pct", "n": 51, "r": -0.0372, "rho": 0.0595, "q": 0.93, "partial_final_r": -0.0815, "controls": [ "median_household_income" ], "x_tag": "measured/administrative", "y_tag": "measured/administrative (derived change)" }, { "id": "ST-7", "level": "state", "pair": "pct_rural ~ ts_age_adj_rate_change_2019_2020_pct", "n": 51, "r": -0.3067, "rho": -0.26, "q": 0.0639, "partial_final_r": -0.4803, "controls": [ "median_household_income" ], "x_tag": "survey estimate (ACS)", "y_tag": "measured/administrative (derived change)" }, { "id": "SS-1", "level": "state", "pair": "pct_depression ~ ts_scd_reported__65plus_pct_latest", "n": 51, "r": 0.1707, "rho": 0.1632, "q": 0.351, "partial_final_r": 0.2987, "controls": [ "median_household_income", "region_south" ], "x_tag": "survey estimate / administrative compilation", "y_tag": "survey estimate (self-report)" }, { "id": "SS-2", "level": "state", "pair": "pct_bachelors_plus ~ ts_scd_reported__65plus_pct_latest", "n": 51, "r": 0.1572, "rho": 0.2868, "q": 0.402, "partial_final_r": -0.0325, "controls": [ "median_household_income" ], "x_tag": "survey estimate (ACS 1-year)", "y_tag": "survey estimate (self-report)" }, { "id": "SS-3", "level": "state", "pair": "pct_hearing_loss ~ ts_scd_reported__65plus_pct_latest", "n": 51, "r": -0.034, "rho": -0.0934, "q": 0.936, "partial_final_r": 0.1016, "controls": [ "median_household_income" ], "x_tag": "survey estimate (self-report)", "y_tag": "survey estimate (self-report)" }, { "id": "SM-19", "level": "state", "pair": "cms_nh_beds_per_1000_pop ~ ts_alz_age_adjusted_death_rate_2022", "n": 50, "r": -0.0108, "rho": -0.0481, "q": 0.979, "partial_final_r": -0.1266, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "measured/administrative (derived ratio); blank for Connecticut (0 matching CMS rows under either FIPS scheme) and 316 other counties with no Care Compare facility", "y_tag": "measured/administrative (vital registration)" }, { "id": "SM-20", "level": "state", "pair": "medicare_advantage_penetration_pct ~ ts_alz_age_adjusted_death_rate_2022", "n": 51, "r": 0.1361, "rho": 0.1603, "q": 0.48, "partial_final_r": 0.0598, "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "x_tag": "measured/administrative (derived ratio); blank for Connecticut's 8 legacy counties and small-cell-suppressed counties", "y_tag": "measured/administrative (vital registration)" }, { "id": "SP-25", "level": "state", "pair": "naics62_health_socasst_emplvl_per_1000pop ~ norc_state65_prevalence_rate_2020", "n": 51, "r": -0.0126, "rho": -0.0689, "q": 0.979, "partial_final_r": 0.0144, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "measured/administrative (derived ratio; inherits the suppression floor of the employment count)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "SP-26", "level": "state", "pair": "naics623_nursing_residential_care_emplvl_per_1000pop ~ norc_state65_prevalence_rate_2020", "n": 51, "r": 0.1312, "rho": 0.1959, "q": 0.487, "partial_final_r": 0.2887, "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "x_tag": "measured/administrative (derived ratio; inherits the suppression floor of the employment count)", "y_tag": "measured/administrative (claims-based DIAGNOSED cases)" }, { "id": "KP-8", "level": "country", "pair": "pct_obesity ~ dementia_deaths_per_100k_age_std", "n": 190, "r": -0.0176, "rho": 0.1224, "q": 0.83, "partial_final_r": -0.1298, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "modelled (NCD-RisC)", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "KP-11", "level": "country", "pair": "pct_physical_inactivity ~ dementia_deaths_per_100k_age_std", "n": 191, "r": -0.0156, "rho": 0.1383, "q": 0.83, "partial_final_r": -0.182, "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "x_tag": "modelled/derived", "y_tag": "modelled (WHO GHE cause-of-death ensemble)" }, { "id": "KT-2", "level": "country", "pair": "gdp_per_capita_usd ~ ts_crude_rate_change_2000_2021_pct", "n": 142, "r": 0.067, "rho": 0.559, "q": 0.458, "partial_final_r": -0.1077, "controls": [ "ts_pop65_change_2000_2021_points" ], "x_tag": "administrative compilation (World Bank)", "y_tag": "modelled numerator (derived change)" }, { "id": "LS-5", "level": "study", "pair": "evidence_ordinal ~ log10_rcr", "n": 6818, "r": -0.0493, "rho": -0.0283, "q": 5.74e-05, "partial_final_r": -0.1189, "controls": [ "year", "journal_tier", "is_review_or_meta" ], "x_tag": "curated label (editorial judgement)", "y_tag": "modelled metric (iCite normalisation)" }, { "id": "LS-7", "level": "study", "pair": "design_ordinal ~ log10_rcr", "n": 2103, "r": 0.0792, "rho": 0.0887, "q": 0.000299, "partial_final_r": 0.0832, "controls": [ "year" ], "x_tag": "curated label (editorial judgement)", "y_tag": "modelled metric (iCite normalisation)" }, { "id": "LS-8", "level": "study", "pair": "open_access ~ log10_rcr", "n": 6663, "r": 0.0001, "rho": -0.0059, "q": 0.995, "partial_final_r": -0.0268, "controls": [ "year" ], "x_tag": "measured/administrative (PubMed flag)", "y_tag": "modelled metric (iCite normalisation)" }, { "id": "LS-10", "level": "study", "pair": "has_effect_size ~ log10_rcr", "n": 6828, "r": 0.0441, "rho": 0.0344, "q": 0.000299, "partial_final_r": 0.0369, "controls": [ "year" ], "x_tag": "curated metadata", "y_tag": "modelled metric (iCite normalisation)" }, { "id": "LU-2", "level": "study", "pair": "site_score ~ log10_rcr", "n": 6818, "r": 0.0842, "rho": 0.1015, "q": 4.99e-12, "partial_final_r": -0.1112, "controls": [ "year", "journal_tier", "is_review_or_meta" ], "x_tag": "derived index (site rubric; built FROM evidence/design/journal/year/size fields)", "y_tag": "modelled metric (iCite normalisation)" }, { "id": "LG-1", "level": "genetics", "pair": "log10_allele_frequency ~ abs_log_or", "n": 90, "r": -0.0917, "rho": -0.4257, "q": 0.39, "partial_final_r": null, "controls": [], "x_tag": "measured (population genetics database)", "y_tag": "curated from published studies" } ], "headline_artefacts": [ { "title": "State 'Alzheimer's' death rates are mostly a coding-practice league table", "ids": [ "SK-1", "SK-2", "SK-3", "SM-1", "SM-8", "ST-2" ], "numbers": { "share_G30_vs_age_adj_rate_2017_r": 0.8183, "share_G30_2017_vs_rate_2022_r": 0.7892, "share_F03_vs_rate_r": -0.7143, "education_vs_age_adj_rate_raw_r": -0.5598, "education_partial_on_income_and_G30_share": -0.2323, "pcp_supply_raw_r": -0.5047, "pcp_partial": -0.2024 }, "text": "The fraction of a state's dementia deaths that certifiers code as G30 (Alzheimer's) rather than F03 (unspecified dementia) explains most of the between-state variation in the age-adjusted Alzheimer's death rate, and predicts the 2010-2019 trend. The baseline's state education and physician-supply gradients (S4, B5) fall by more than half once income and the G30 share are controlled." }, { "title": "There is no age-standardised country outcome, and the GDP-dementia correlation is age structure plus certification", "ids": [ "KA-1", "KA-2", "KA-3", "KT-3", "KP-1", "KP-2", "KP-3", "KP-17" ], "numbers": { "labelled_age_std_vs_crude_r": 0.9994, "pct65_vs_death_rate_r": 0.7179, "gdp_raw_r": 0.8176, "gdp_partial_pct65": 0.6878, "health_spend_raw_r": 0.7396, "health_spend_partial_pct65_gdp": 0.029, "physicians_raw_r": 0.6204, "physicians_partial": 0.1333 }, "text": "The timeseries column labelled age-standardised is numerically the crude rate. Age structure alone correlates 0.72 with the country death rate. Health spending, physician density, hospital beds, urbanisation, literacy, UHC coverage and immunisation all collapse under age structure (and GDP). GDP itself survives a linear pct_65plus control, which cannot be read as anything more than 'rich countries are older AND certify dementia deaths': no file here can separate those." }, { "title": "County diagnosed prevalence tracks detection, not just disease", "ids": [ "CD-3", "CD-4", "CD-5", "CD-1", "CX-18", "CX-17", "CA-1" ], "numbers": { "MA_share_partial": 0.1279, "share_highly_likely_partial": -0.4117, "ffs_cost_partial": 0.1633, "checkup_raw_r": 0.2292, "checkup_partial_ext": 0.1116, "pct65_strict_partial": -0.3478, "pct85_raw_r": -0.1316, "all_age_rate_vs_share_bene_65plus_r": -0.2902 }, "text": "Counties with a higher Medicare Advantage share and higher per-beneficiary spending have higher diagnosed prevalence; counties with high prevalence have a LOWER share of 'highly likely' cases, i.e. the excess sits in the softer tiers. Counties with older populations (more 65+ and 85+) have LOWER diagnosed prevalence among the 65+, the opposite of biology. The all-age rate the baseline used is also diluted by under-65 disability beneficiaries." }, { "title": "The county race gradient is a place effect, not (only) composition, and it collapses under place controls", "ids": [ "CX-11", "CC-1", "CC-2", "CC-4", "CC-5", "CH-1", "CP-1" ], "numbers": { "pct_black_raw_r": 0.3596, "pct_black_ext_partial": 0.1596, "pct_black_strict_partial": 0.1177, "pct_black_vs_WHITE_stratum_rate_r": 0.3202, "pct_black_vs_BLACK_stratum_rate_r": 0.0407, "white_vs_black_stratum_rate_r": 0.495, "region_south_strict": 0.2599 }, "text": "The baseline's D1 (pct_black ~ prevalence, partial 0.27 on income) is real but is not about Black beneficiaries: White beneficiaries in high-pct_black counties also have higher diagnosed prevalence (CC-1), while prevalence among Black beneficiaries does not rise with pct_black at all (CC-2). Adding region (South), density and coding-intensity markers takes the pct_black partial to 0.12. The candidate third thing is Southern place: stroke-belt vascular risk and Southern coding practice, which these files cannot separate." }, { "title": "Modelled PLACES risk factors collapse; PM2.5 is the one environmental exposure that survives, with caveats", "ids": [ "CX-26", "CX-28", "CX-30", "CX-35", "CX-36", "CX-37", "CH-6", "CI-6", "CM-1", "SP-18", "KP-15" ], "numbers": { "diabetes_raw_r": 0.3784, "diabetes_partial": 0.0843, "hypertension_raw_r": 0.3575, "hypertension_partial": 0.0736, "inactivity_raw_r": 0.4513, "inactivity_strict": 0.2274, "pm25_modelled_lag_raw_r": 0.3991, "pm25_modelled_lag_strict": 0.2839, "pm25_monitor_lag_raw_r": 0.1752, "pm25_monitor_lag_strict": 0.0527, "pm25_monitor_vs_modelled_r": 0.8733, "state_pm25_strict": 0.0978 }, "text": "County diabetes, hypertension, obesity, smoking and fair/poor health (all PLACES small-area MODEL outputs that share ACS inputs with the controls) collapse under age+income+physician+region. Physical inactivity keeps a partial of about 0.23. Modelled 2016-2019 PM2.5 keeps about 0.28 under the strict set, but the MEASURED monitor-based PM2.5 (n=559, urban-selected counties) falls to 0.05 under the same controls, and at state level PM2.5 collapses to 0.10 once race and claims intensity are controlled. The modelled PM2.5 survivor is therefore not confirmed by the measured one." }, { "title": "Raw counts correlate through population, at every level", "ids": [ "CA-2", "CA-3", "SA-1", "SA-2", "SA-3" ], "numbers": { "county_prev_count_vs_pcp_count_r": 0.9446, "partial_on_population": 0.1363, "state_geriatricians_vs_alz_count_r": 0.9256, "state_partial": 0.439, "state_caregivers_vs_norc_count_r": 0.9712, "state_caregivers_partial": 0.2611 }, "text": "Five count-versus-count pairs with raw r between 0.88 and 0.97 fall to 0.14-0.44 once population is controlled. None is a relationship." }, { "title": "Measured and modelled versions of the same outcome disagree", "ids": [ "SK-5", "SK-6", "SK-7", "SK-8", "SU-1", "SU-2", "SU-3" ], "numbers": { "norc_measured_vs_AA_modelled_prevalence_r": 0.6447, "norc_prevalence_vs_age_adj_mortality_r": -0.0266, "AA_prevalence_vs_mortality_r": -0.3873, "norc_vs_self_reported_SCD_r": -0.1787, "AA_prevalence_vs_pct_black_r": 0.7668 }, "text": "State diagnosed prevalence (claims) and the Alzheimer's Association modelled prevalence agree at r=0.65; neither agrees with the age-adjusted death rate (r=-0.03 and -0.39) nor with self-reported cognitive decline. The AA estimate correlates 0.77 with pct_race_black because race is an input to that model (baseline B2/B4/S2/S3/S10/S12/A3 all used it as the outcome)." } ], "cannot_answer": [ "Individual-level risk. Every coefficient here is a property of places (or of library records); none can say whether a person with the exposure is more or less likely to develop dementia.", "True disease prevalence. The only county and state outcome is DIAGNOSED dementia in Medicare claims (NORC). Under-diagnosis and lower true disease produce the same number; CD-1 to CD-6 and CH-11 show diagnosed prevalence tracks Medicare Advantage share, check-up rates and per-beneficiary claims spending, so part of the county map is detection intensity.", "County Alzheimer's mortality. panel_county.csv's dementia_mortality_per_100k, dementia_deaths_total and alz_deaths_total are NOT SOURCED (CDC WONDER forbids county grouping through the API and blocks automated requests; see data/county_outcomes_README.md). The only county 'mortality' column is NORC's all-cause deaths among diagnosed cases, which is a case-fatality rate and is uninterpretable against prevalence (CU-1).", "Age-standardised country comparisons. Both country outcome columns are crude despite their names: panel_country_meta.json says so for dementia_deaths_per_100k_age_std, and KA-1/KA-2/KT-3 show the timeseries 'age-standardised' series is numerically identical (r=0.999) to deaths divided by population. A linear partial on pct_65plus is not standardisation. Every country result is therefore about age structure plus death-certification completeness unless proven otherwise.", "Country prevalence and incidence (NOT POPULATED in panel_country.csv; GBD-only, login-gated).", "County time trends in dementia outcomes. NORC gives one year (2020) plus a rounded 2020-2022 pooled figure. The exposure panel is a 2016-2025 time series, but it has nothing to be correlated against over time; only lagged (2016-2019) exposure vs 2020 outcome is possible, and it was done (CX-36 to CX-38).", "State crude death rates before 2020 (no state population before 2020 in panel_state_timeseries.csv; Census API is key-gated) and BRFSS cognitive measures after 2022 (not published upstream).", "Race-stratified county rates for most counties. Complementary suppression leaves the Black stratum unsuppressed in 1,547 counties and the Hispanic stratum in 1,304; those are the counties with the largest minority populations, so CC-2 and CC-3 are selected samples.", "Lead exposure and all-physician supply at county level (NOT SOURCED columns); water-system violations at state level (blank for all 51). The panel's original county nursing_home_beds_per_1000 column is STILL entirely unsourced (0 of 3,143 rows, see CM-15); Phase 4 added a working replacement, cms_nh_beds_per_1000_pop (CMS Care Compare certified beds), populated for 2,826 of 3,143 counties (blank for Connecticut's 8 legacy counties and 309 further counties with no Care Compare facility).", "Anything about drugs. FAERS has no exposure denominator and Part D covers 9 drug-program-year rows (LD-1); a correlation on n=9 with cumulative report counts cannot be interpreted.", "Genetics against geography. genetics.json is 135 curated published effect sizes (mixed odds-ratio definitions), not a genotype dataset; it supports meta-research checks (LG-1 to LG-4) and nothing spatial.", "Whether a null place-level correlation means no effect. Ecological nulls (e.g. depression, social association, food access at county level) are compatible with real individual-level effects hidden by aggregation." ] }, "twins_by_frame": { "county": [ { "a": "dollar_stores_per_1000pop_2020", "b": "exp_dollar_stores_per_1000pop_2020", "r": 1.0, "n": 3143 }, { "a": "exp_fastfood_restaurants_per_1000pop_2020", "b": "fastfood_restaurants_per_1000pop_2020", "r": 1.0, "n": 2672 }, { "a": "exp_fullservice_restaurants_per_1000pop_2020", "b": "fullservice_restaurants_per_1000pop_2020", "r": 1.0, "n": 2861 }, { "a": "exp_grocery_stores_per_1000pop_2020", "b": "grocery_stores_per_1000pop_2020", "r": 1.0, "n": 2234 }, { "a": "exp_pct_65plus_2020", "b": "pct_65plus", "r": 0.9917, "n": 3135 }, { "a": "exp_pct_seniors_low_access_2019", "b": "pct_low_access_pop_1mi", "r": 0.9756, "n": 3139 }, { "a": "exp_primary_care_physicians_count_2023", "b": "population", "r": 0.9643, "n": 3143 }, { "a": "exp_primary_care_physicians_per_1000pop_2023", "b": "primary_care_physicians_per_100k", "r": 0.951, "n": 2978 }, { "a": "norc65_prev_cnt_2020", "b": "norc_prev_cnt_2020", "r": 0.9998, "n": 2579 }, { "a": "norc65_prev_cnt_2020", "b": "population", "r": 0.9682, "n": 2580 }, { "a": "norc65_prevalence_rate_2020", "b": "norc_dementia_prevalence_rate_2020", "r": 0.9752, "n": 2579 }, { "a": "norc65_prevalence_rate_2020", "b": "norc_pooled_prevalence_rate_2020_2022", "r": 0.9712, "n": 2580 }, { "a": "norc65_prevalence_rate_2020", "b": "norc_white_prevalence_rate_2020", "r": 0.9583, "n": 2579 }, { "a": "norc_dementia_prevalence_rate_2020", "b": "norc_white_prevalence_rate_2020", "r": 0.9736, "n": 3092 }, { "a": "norc_prev_cnt_2020", "b": "population", "r": 0.9706, "n": 3104 } ], "state": [ { "a": "alz_count_65plus", "b": "norc_state_prev_cnt_2020", "r": 0.9912, "n": 51 }, { "a": "alz_count_65plus", "b": "pop_total", "r": 0.9785, "n": 51 }, { "a": "alz_count_65plus", "b": "ts_dementia_deaths_total_2017", "r": 0.9729, "n": 51 }, { "a": "alz_count_65plus", "b": "unpaid_caregivers_count", "r": 0.9726, "n": 51 }, { "a": "cms_nh_beds_per_1000_pop", "b": "nursing_home_beds_per_1000", "r": 0.9534, "n": 50 }, { "a": "geriatricians_count", "b": "medicaid_dementia_spend_total_usd", "r": 0.9561, "n": 51 }, { "a": "norc_state65_prevalence_rate_2020", "b": "norc_state_prevalence_rate_2020", "r": 0.9837, "n": 51 }, { "a": "norc_state_prev_cnt_2020", "b": "pop_total", "r": 0.9725, "n": 51 }, { "a": "norc_state_prev_cnt_2020", "b": "ts_dementia_deaths_total_2017", "r": 0.9746, "n": 51 }, { "a": "norc_state_prev_cnt_2020", "b": "unpaid_caregivers_count", "r": 0.9712, "n": 51 }, { "a": "pop_total", "b": "ts_dementia_deaths_total_2017", "r": 0.9746, "n": 51 }, { "a": "pop_total", "b": "unpaid_caregivers_count", "r": 0.9897, "n": 51 }, { "a": "ts_dementia_deaths_total_2017", "b": "unpaid_caregivers_count", "r": 0.9712, "n": 51 } ], "country": [ { "a": "dementia_daly_per_100k", "b": "dementia_deaths_per_100k_age_std", "r": 0.977, "n": 192 }, { "a": "dementia_daly_per_100k", "b": "ts_gbd_deaths_crude_per_100k_2021", "r": 0.9776, "n": 143 }, { "a": "dementia_daly_per_100k", "b": "ts_gbd_deaths_per_100k_age_std_2021", "r": 0.9748, "n": 143 }, { "a": "dementia_deaths_per_100k_age_std", "b": "ts_gbd_deaths_crude_per_100k_2021", "r": 0.9994, "n": 143 }, { "a": "dementia_deaths_per_100k_age_std", "b": "ts_gbd_deaths_per_100k_age_std_2021", "r": 1.0, "n": 143 }, { "a": "ts_age_std_rate_change_2000_2021_pct", "b": "ts_crude_rate_change_2000_2021_pct", "r": 0.9984, "n": 143 }, { "a": "ts_gbd_deaths_crude_per_100k_2021", "b": "ts_gbd_deaths_per_100k_age_std_2021", "r": 0.9994, "n": 143 } ], "study": [ { "a": "log10_icite_citations", "b": "log10_openalex_citations", "r": 0.9748, "n": 6899 }, { "a": "year", "b": "years_since_pub", "r": -1.0, "n": 14775 } ], "genetics": [] }, "assembly": { "built_from": { "county": [ "data/panel_county.csv", "data/norc/NORC_Dementia_DataHub_County_2020.csv", "data/norc/DDH_POOL_2020_2022_publishable.csv", "data/panel_exposure_county.csv" ], "state": [ "data/panel_state.csv", "data/norc/NORC_Dementia_DataHub_Nat_State_2020.csv", "data/panel_state_timeseries.csv" ], "country": [ "data/panel_country.csv", "data/panel_country_timeseries.csv" ], "study": [ "data/research.json", "data/citations_gaps.json", "data/research_score.json", "data/most_cited.json (top-40 lists only; no per-record RCR)" ], "genetics": [ "data/genetics.json" ], "drugs": [ "data/partd_ad_drugs.csv", "data/faers_ad_drugs.json" ] }, "per_entity_fill": { "panel_exposure_county.svi_overall_percentile": { "rows_filled": 3144, "rows_total": 29341, "entities_filled": 3144, "entities_total": 3240, "year_used": "2022" }, "panel_exposure_county.svi_theme1_socioeconomic_percentile": { "rows_filled": 3144, "rows_total": 29341, "entities_filled": 3144, "entities_total": 3240, "year_used": "2022" }, "panel_exposure_county.svi_theme2_household_percentile": { "rows_filled": 3144, "rows_total": 29341, "entities_filled": 3144, "entities_total": 3240, "year_used": "2022" }, "panel_exposure_county.svi_theme3_minority_status_percentile": { "rows_filled": 3144, "rows_total": 29341, "entities_filled": 3144, "entities_total": 3240, "year_used": "2022" }, "panel_exposure_county.svi_theme4_housing_transport_percentile": { "rows_filled": 3144, "rows_total": 29341, "entities_filled": 3144, "entities_total": 3240, "year_used": "2022" }, "panel_exposure_county.primary_care_physicians_count_2023": { "rows_filled": 3225, "rows_total": 29341, "entities_filled": 3225, "entities_total": 3240, "year_used": "2023" }, "panel_exposure_county.primary_care_physicians_per_1000pop_2023": { "rows_filled": 3214, "rows_total": 29341, "entities_filled": 3214, "entities_total": 3240, "year_used": "2023" }, "panel_exposure_county.hospitals_count_2023": { "rows_filled": 3226, "rows_total": 29341, "entities_filled": 3226, "entities_total": 3240, "year_used": "2023" }, "panel_exposure_county.hospital_beds_count_2023": { "rows_filled": 3226, "rows_total": 29341, "entities_filled": 3226, "entities_total": 3240, "year_used": "2023" }, "panel_exposure_county.hospital_beds_per_1000pop_2023": { "rows_filled": 3214, "rows_total": 29341, "entities_filled": 3214, "entities_total": 3240, "year_used": "2023" }, "panel_exposure_county.pct_low_access_to_store_2019": { "rows_filled": 3141, "rows_total": 29341, "entities_filled": 3141, "entities_total": 3240, "year_used": "2019" }, "panel_exposure_county.pct_low_income_low_access_2019": { "rows_filled": 3141, "rows_total": 29341, "entities_filled": 3141, "entities_total": 3240, "year_used": "2019" }, "panel_exposure_county.pct_seniors_low_access_2019": { "rows_filled": 3139, "rows_total": 29341, "entities_filled": 3139, "entities_total": 3240, "year_used": "2019" }, "panel_exposure_county.grocery_stores_per_1000pop_2020": { "rows_filled": 2234, "rows_total": 29341, "entities_filled": 2234, "entities_total": 3240, "year_used": "2020" }, "panel_exposure_county.convenience_stores_per_1000pop_2020": { "rows_filled": 2859, "rows_total": 29341, "entities_filled": 2859, "entities_total": 3240, "year_used": "2020" }, "panel_exposure_county.dollar_stores_per_1000pop_2020": { "rows_filled": 3143, "rows_total": 29341, "entities_filled": 3143, "entities_total": 3240, "year_used": "2020" }, "panel_exposure_county.fastfood_restaurants_per_1000pop_2020": { "rows_filled": 2672, "rows_total": 29341, "entities_filled": 2672, "entities_total": 3240, "year_used": "2020" }, "panel_exposure_county.fullservice_restaurants_per_1000pop_2020": { "rows_filled": 2861, "rows_total": 29341, "entities_filled": 2861, "entities_total": 3240, "year_used": "2020" }, "panel_exposure_county.snap_authorized_stores_per_1000pop_2023": { "rows_filled": 3114, "rows_total": 29341, "entities_filled": 3114, "entities_total": 3240, "year_used": "2023" }, "panel_exposure_county.wic_authorized_stores_per_1000pop_2022": { "rows_filled": 3028, "rows_total": 29341, "entities_filled": 3028, "entities_total": 3240, "year_used": "2022" }, "panel_exposure_county.snap_benefits_per_capita_2022": { "rows_filled": 3090, "rows_total": 29341, "entities_filled": 3090, "entities_total": 3240, "year_used": "2022" }, "panel_exposure_county.pm25_annual_mean_ugm3": { "rows_filled": 6313, "rows_total": 29341, "entities_filled": 714, "entities_total": 3240, "year_used": "2020" }, "panel_exposure_county.pm25_monitor_count": { "rows_filled": 6313, "rows_total": 29341, "entities_filled": 714, "entities_total": 3240, "year_used": "2020" }, "panel_exposure_county.ozone_annual_mean_ppm": { "rows_filled": 7744, "rows_total": 29341, "entities_filled": 835, "entities_total": 3240, "year_used": "2020" }, "panel_exposure_county.pm25_annual_mean_popweighted_ugm3": { "rows_filled": 21592, "rows_total": 29341, "entities_filled": 3109, "entities_total": 3240, "year_used": "2020" }, "panel_exposure_county.total_population": { "rows_filled": 15720, "rows_total": 29341, "entities_filled": 3144, "entities_total": 3240, "year_used": "2020" }, "panel_exposure_county.pct_65plus": { "rows_filled": 15720, "rows_total": 29341, "entities_filled": 3144, "entities_total": 3240, "year_used": "2020" }, "panel_exposure_county.pct_85plus": { "rows_filled": 15720, "rows_total": 29341, "entities_filled": 3144, "entities_total": 3240, "year_used": "2020" }, "panel_exposure_county.pct_black_alone": { "rows_filled": 15720, "rows_total": 29341, "entities_filled": 3144, "entities_total": 3240, "year_used": "2020" }, "panel_exposure_county.pct_hispanic": { "rows_filled": 15720, "rows_total": 29341, "entities_filled": 3144, "entities_total": 3240, "year_used": "2020" }, "panel_exposure_county.pct_white_nonhispanic": { "rows_filled": 15720, "rows_total": 29341, "entities_filled": 3144, "entities_total": 3240, "year_used": "2020" } }, "joins": { "county_norc_raw_65plus": { "matched": 3129, "of_panel_counties": 3143 }, "county_norc_pooled": { "matched": 3132, "of_panel_counties": 3143 }, "norc_state_code_check": [ { "code": "06", "state": "CA", "state_row_bene_cnt": 6206224.0, "sum_of_county_total_rows": 6205879.0, "ratio": 1.0000555924470973 }, { "code": "48", "state": "TX", "state_row_bene_cnt": 4230413.0, "sum_of_county_total_rows": 4230365.0, "ratio": 1.0000113465386558 }, { "code": "12", "state": "FL", "state_row_bene_cnt": 4639032.0, "sum_of_county_total_rows": 4638893.0, "ratio": 1.0000299640453014 } ], "state_norc": { "matched": 51, "of_states": 51 }, "country_timeseries": { "matched": 143, "of_countries": 217 }, "study_rcr": { "research_records": 14779, "matched_to_citations_gaps_rcr": 6838, "citations_gaps_records_with_rcr": 7024 } }, "notes": [ "NORC raw 65+ stratum matched 3129 of 3143 panel counties (the rest have no raw row or complementary suppression)", "NORC pooled 2020-2022 county totals matched 3132 counties; counts in that file are rounded by NORC, so rates are approximate", "exp_pm25_monitor_mean_2016_2019: filled for 585 counties (requires all four years 2016-2019 present)", "exp_pm25_modelled_mean_2016_2019: filled for 3107 counties (requires all four years 2016-2019 present)", "exp_ozone_monitor_mean_2016_2019: filled for 757 counties (requires all four years 2016-2019 present)", "Phase 5: derived naics62/naics623 employment-per-1000-population from Phase 4's BLS QCEW join and panel population (raw employment levels are population-scaled counts; a suppressed ownership cell makes the numerator a floor, so the ratio is too)", "NORC FIPS_STATE code check (state row vs sum of county totals): CA(06) 6206224 vs 6205879; TX(48) 4230413 vs 4230365; FL(12) 4639032 vs 4638893", "NORC state totals matched 51 of 51 states", "Phase 5: derived naics62/naics623 employment-per-1000-population at state level from Phase 4's BLS QCEW state rollup and pop_total", "country time series (IHME GBD via OWID) matched 143 of 217 World Bank economies; Americas and Oceania are absent upstream", "RCR joined for 6838 of 14779 research.json records (by pmid then doi) from citations_gaps.json", "drug-level frame has 9 drug-program rows and 7 FAERS drugs; FAERS counts are all-time cumulative, Part D/B is a single year" ] }, "results": [ { "id": "CX-1", "level": "county", "kind": "exposure", "x": { "name": "median_household_income", "unit": "US dollars (nominal, ACS inflation-adjusted to survey-period dollars)", "year": "2020-2024 (ACS) or CHR vintage for 8 CT counties", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 3143, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.329, "p": 3.53e-66, "ci95": [ -0.363, -0.2941 ] }, "spearman": { "rho": -0.3779, "p": 2.31e-88 }, "partial_primary": { "controls": [ "pct_85plus" ], "r": -0.3519, "p": 7.23e-76, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "primary_care_physicians_per_100k", "region_south" ], "r": -0.2582, "p": 1.11e-39, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.2254, "p": 2.01e-28, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "negative", "why": "higher income / education is protective at the individual level" }, "confounders_named": [ "diagnosis intensity (insurance, physician supply)", "region", "race composition" ], "note": "", "fdr_q_within_level": 1.18e-65, "classification": "survives", "verdict": "Raw moderate negative correlation r=-0.33 (95% CI -0.36 to -0.29), Spearman rho=-0.38, n=2580 counties, p=3.53e-66, BH q=1.2e-65. SURVIVES the confounder test: partial r=-0.352 (p=7.23e-76) controlling for pct_85plus; extended partial r=-0.2582; strict partial r=-0.2254 controlling for pct_85plus+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: diagnosis intensity (insurance, physician supply), region, race composition. Provenance: median_household_income is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-2", "level": "county", "kind": "exposure", "x": { "name": "pct_bachelors_plus", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.2378, "p": 2.11e-34, "ci95": [ -0.274, -0.201 ] }, "spearman": { "rho": -0.2759, "p": 3.59e-46 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.0186, "p": 0.346, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0432, "p": 0.0302, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0859, "p": 3.1e-05, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "negative", "why": "education is one of the largest modifiable dementia risk factors (Lancet Commission)" }, "confounders_named": [ "income", "urbanicity", "diagnosis intensity" ], "note": "", "fdr_q_within_level": 4.59e-34, "classification": "collapses", "verdict": "Raw weak negative correlation r=-0.24 (95% CI -0.27 to -0.20), Spearman rho=-0.28, n=2572 counties, p=2.11e-34, BH q=4.6e-34. COLLAPSES: partial r=0.019 is under half the raw r=-0.238 once pct_85plus+median_household_income is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_bachelors_plus. Provenance: pct_bachelors_plus is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-3", "level": "county", "kind": "exposure", "x": { "name": "pct_below_poverty", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.4119, "p": 6.96e-106, "ci95": [ 0.3793, 0.4435 ] }, "spearman": { "rho": 0.4206, "p": 8.53e-111 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.2197, "p": 1.73e-29, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1566, "p": 2.64e-15, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.104, "p": 4.43e-07, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region (South)", "race composition", "Medicaid dual-eligibility raising claims contact" ], "note": "", "fdr_q_within_level": 4.11e-105, "classification": "collapses_extended", "verdict": "Raw moderate positive correlation r=0.41 (95% CI 0.38 to 0.44), Spearman rho=0.42, n=2572 counties, p=6.96e-106, BH q=4.1e-105. COLLAPSES under the extended controls: survives the primary control (partial r=0.220) but falls to 0.157 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. Provenance: pct_below_poverty is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-4", "level": "county", "kind": "exposure", "x": { "name": "unemployment_rate", "unit": "percent (0-100)", "year": "2020-2024 (ACS) or CHR vintage for 8 CT counties", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 3143, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.2273, "p": 1.41e-31, "ci95": [ 0.1903, 0.2635 ] }, "spearman": { "rho": 0.2342, "p": 1.74e-33 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.0983, "p": 5.9e-07, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0627, "p": 0.00164, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0059, "p": 0.775, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region", "income" ], "note": "", "fdr_q_within_level": 2.82e-31, "classification": "collapses", "verdict": "Raw weak positive correlation r=0.23 (95% CI 0.19 to 0.26), Spearman rho=0.23, n=2580 counties, p=1.41e-31, BH q=2.8e-31. COLLAPSES: partial r=0.098 is under half the raw r=0.227 once pct_85plus+median_household_income is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with unemployment_rate. Provenance: unemployment_rate is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-5", "level": "county", "kind": "exposure", "x": { "name": "pct_uninsured", "unit": "percent (0-100)", "year": "~2021 (Small Area Health Insurance Estimates vintage embedded in CHR 2024)", "source": "County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release via data/panel_county.csv", "tag": "administrative compilation (CHR pass-through)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 3142, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.3002, "p": 6.95e-55, "ci95": [ 0.2647, 0.3349 ] }, "spearman": { "rho": 0.3322, "p": 1.7e-67 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1759, "p": 2.54e-19, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0551, "p": 0.00565, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1157, "p": 1.89e-08, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region (South, non-expansion states)", "poverty" ], "note": "pct_uninsured is a working-age measure; Medicare beneficiaries are insured by definition, so it can only act as a place marker.", "fdr_q_within_level": 1.92e-54, "classification": "collapses_extended", "verdict": "Raw moderate positive correlation r=0.30 (95% CI 0.26 to 0.33), Spearman rho=0.33, n=2580 counties, p=6.95e-55, BH q=1.9e-54. COLLAPSES under the extended controls: survives the primary control (partial r=0.176) but falls to 0.055 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. pct_uninsured is a working-age measure; Medicare beneficiaries are insured by definition, so it can only act as a place marker. Provenance: pct_uninsured is administrative compilation (CHR pass-through); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-6", "level": "county", "kind": "exposure", "x": { "name": "exp_svi_theme1_socioeconomic_percentile", "unit": "percentile (0-1)", "year": "2022 (single cross-section; placed only on each county's 2022 row)", "source": "CDC/ATSDR Social Vulnerability Index, 2022 database, County via data/panel_exposure_county.csv", "tag": "derived index (from ACS survey estimates)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.4162, "p": 2.59e-108, "ci95": [ 0.3837, 0.4477 ] }, "spearman": { "rho": 0.4448, "p": 3.48e-125 }, "partial_primary": { "controls": [ "pct_85plus" ], "r": 0.402, "p": 1.73e-100, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.14, "p": 1.69e-12, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1087, "p": 1.29e-07, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "income (SVI theme 1 is built from income/poverty/education)", "region" ], "note": "", "fdr_q_within_level": 1.69e-107, "classification": "collapses_extended", "verdict": "Raw moderate positive correlation r=0.42 (95% CI 0.38 to 0.45), Spearman rho=0.44, n=2572 counties, p=2.59e-108, BH q=1.7e-107. COLLAPSES under the extended controls: survives the primary control (partial r=0.402) but falls to 0.140 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. Provenance: exp_svi_theme1_socioeconomic_percentile is derived index (from ACS survey estimates); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-7", "level": "county", "kind": "exposure", "x": { "name": "exp_svi_overall_percentile", "unit": "percentile (0-1)", "year": "2022 (single cross-section; placed only on each county's 2022 row)", "source": "CDC/ATSDR Social Vulnerability Index, 2022 database, County via data/panel_exposure_county.csv", "tag": "derived index (from ACS survey estimates)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.4115, "p": 1.11e-105, "ci95": [ 0.3789, 0.4431 ] }, "spearman": { "rho": 0.4392, "p": 8.67e-122 }, "partial_primary": { "controls": [ "pct_85plus" ], "r": 0.3974, "p": 4.81e-98, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1681, "p": 1.94e-17, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1499, "p": 2.88e-13, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "income", "race composition (theme 3 is minority status)", "region" ], "note": "", "fdr_q_within_level": 6.26e-105, "classification": "collapses_extended", "verdict": "Raw moderate positive correlation r=0.41 (95% CI 0.38 to 0.44), Spearman rho=0.44, n=2572 counties, p=1.11e-105, BH q=6.3e-105. COLLAPSES under the extended controls: survives the primary control (partial r=0.397) but falls to 0.168 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. Provenance: exp_svi_overall_percentile is derived index (from ACS survey estimates); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-8", "level": "county", "kind": "exposure", "x": { "name": "exp_snap_benefits_per_capita_2022", "unit": "US dollars per capita", "year": "2022 (single cross-section; placed only on each county's 2022 row)", "source": "USDA ERS Food Environment Atlas 2025 release, variable PC_SNAPBEN22 via data/panel_exposure_county.csv", "tag": "measured/administrative (USDA establishment counts)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2533, "n_unit": "counties", "x_filled": 3090, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.2263, "p": 8.95e-31, "ci95": [ 0.189, 0.2629 ] }, "spearman": { "rho": 0.2597, "p": 2.63e-40 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.0495, "p": 0.0129, "n": 2525 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0017, "p": 0.934, "n": 2479 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0995, "p": 1.71e-06, "n": 2305 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "poverty", "region", "Medicaid dual eligibility" ], "note": "", "fdr_q_within_level": 1.71e-30, "classification": "collapses", "verdict": "Raw weak positive correlation r=0.23 (95% CI 0.19 to 0.26), Spearman rho=0.26, n=2533 counties, p=8.95e-31, BH q=1.7e-30. COLLAPSES: partial r=0.050 is under half the raw r=0.226 once pct_85plus+median_household_income is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with exp_snap_benefits_per_capita_2022. Provenance: exp_snap_benefits_per_capita_2022 is measured/administrative (USDA establishment counts); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-9", "level": "county", "kind": "exposure", "x": { "name": "pct_broadband", "unit": "percent (0-100)", "year": "2020-2024 (ACS) or CHR vintage for 8 CT counties", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 3143, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.2683, "p": 8.51e-44, "ci95": [ -0.3038, -0.2322 ] }, "spearman": { "rho": -0.2554, "p": 1.07e-39 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": -0.077, "p": 9.32e-05, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0453, "p": 0.0229, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0312, "p": 0.131, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "income", "urbanicity", "age structure" ], "note": "", "fdr_q_within_level": 2.07e-43, "classification": "collapses", "verdict": "Raw weak negative correlation r=-0.27 (95% CI -0.30 to -0.23), Spearman rho=-0.26, n=2580 counties, p=8.51e-44, BH q=2.1e-43. COLLAPSES: partial r=-0.077 is under half the raw r=-0.268 once pct_85plus+median_household_income is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_broadband. Provenance: pct_broadband is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-10", "level": "county", "kind": "exposure", "x": { "name": "pct_no_vehicle", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.225, "p": 6.92e-31, "ci95": [ 0.188, 0.2614 ] }, "spearman": { "rho": 0.2723, "p": 5.62e-45 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1554, "p": 2.29e-15, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1732, "p": 1.95e-18, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0696, "p": 0.000736, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "urban density", "poverty" ], "note": "", "fdr_q_within_level": 1.34e-30, "classification": "collapses_strict", "verdict": "Raw weak positive correlation r=0.23 (95% CI 0.19 to 0.26), Spearman rho=0.27, n=2572 counties, p=6.92e-31, BH q=1.3e-30. COLLAPSES under place / diagnosis-intensity controls: survives age/income/region controls (extended partial r=0.1732) but falls to 0.070 once place and diagnosis-intensity markers are added (pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual). What looked like an exposure effect is carried by where the county is and how intensively it codes diagnoses. Provenance: pct_no_vehicle is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-11", "level": "county", "kind": "exposure", "x": { "name": "pct_black", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.3596, "p": 2.21e-79, "ci95": [ 0.3255, 0.3928 ] }, "spearman": { "rho": 0.4471, "p": 1.16e-126 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.2664, "p": 5.01e-43, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1596, "p": 7.59e-16, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1177, "p": 1.07e-08, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region (South / stroke belt)", "poverty", "differential diagnosis and coding by race" ], "note": "", "fdr_q_within_level": 9.45e-79, "classification": "collapses_extended", "verdict": "Raw moderate positive correlation r=0.36 (95% CI 0.33 to 0.39), Spearman rho=0.45, n=2572 counties, p=2.21e-79, BH q=9.5e-79. COLLAPSES under the extended controls: survives the primary control (partial r=0.266) but falls to 0.160 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. Provenance: pct_black is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-12", "level": "county", "kind": "exposure", "x": { "name": "pct_hispanic", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.0478, "p": 0.0153, "ci95": [ 0.0092, 0.0863 ] }, "spearman": { "rho": -0.0251, "p": 0.203 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.07, "p": 0.00038, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0651, "p": 0.00107, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1323, "p": 1.24e-10, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region (Southwest, Florida)", "under-diagnosis in Spanish-speaking populations", "age structure" ], "note": "", "fdr_q_within_level": 0.0172, "classification": "negligible", "verdict": "Raw negligible positive correlation r=0.05 (95% CI 0.01 to 0.09), Spearman rho=-0.03, n=2572 counties, p=0.0153, BH q=0.017. NEGLIGIBLE: raw r=0.048 is statistically non-zero at n=2572 but below 0.10 in size; too small to interpret. Not a finding worth reporting as an association. Provenance: pct_hispanic is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-13", "level": "county", "kind": "exposure", "x": { "name": "pct_white_nh", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.2549, "p": 1.96e-39, "ci95": [ -0.2907, -0.2184 ] }, "spearman": { "rho": -0.2605, "p": 3.53e-41 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": -0.2156, "p": 2e-28, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.128, "p": 1.11e-10, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.072, "p": 0.000485, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region", "income" ], "note": "", "fdr_q_within_level": 4.42e-39, "classification": "collapses_strict", "verdict": "Raw weak negative correlation r=-0.25 (95% CI -0.29 to -0.22), Spearman rho=-0.26, n=2572 counties, p=1.96e-39, BH q=4.4e-39. COLLAPSES under place / diagnosis-intensity controls: survives age/income/region controls (extended partial r=-0.128) but falls to -0.072 once place and diagnosis-intensity markers are added (pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual). What looked like an exposure effect is carried by where the county is and how intensively it codes diagnoses. Provenance: pct_white_nh is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-14", "level": "county", "kind": "exposure", "x": { "name": "pct_asian", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.0651, "p": 0.000956, "ci95": [ -0.1035, -0.0265 ] }, "spearman": { "rho": -0.0782, "p": 7.2e-05 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1256, "p": 1.66e-10, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1125, "p": 1.5e-08, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0671, "p": 0.00115, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "urban density", "income" ], "note": "", "fdr_q_within_level": 0.00112, "classification": "negligible", "verdict": "Raw negligible negative correlation r=-0.07 (95% CI -0.10 to -0.03), Spearman rho=-0.08, n=2572 counties, p=0.000956, BH q=0.0011. NEGLIGIBLE: raw r=-0.065 is statistically non-zero at n=2572 but below 0.10 in size; too small to interpret. Not a finding worth reporting as an association. Provenance: pct_asian is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-15", "level": "county", "kind": "exposure", "x": { "name": "pct_foreign_born", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.0013, "p": 0.948, "ci95": [ -0.0399, 0.0374 ] }, "spearman": { "rho": -0.0503, "p": 0.0107 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1646, "p": 4.41e-17, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1399, "p": 1.75e-12, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1303, "p": 2.37e-10, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "urban density", "language barriers to diagnosis" ], "note": "", "fdr_q_within_level": 0.956, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.00 (95% CI -0.04 to 0.04), Spearman rho=-0.05, n=2572 counties, p=0.948, BH q=0.96. NULL: raw association not distinguishable from zero (BH q=0.96 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_foreign_born is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-16", "level": "county", "kind": "exposure", "x": { "name": "pct_female", "unit": "percent (0-100)", "year": "2020-2024 (ACS) or CHR vintage for the 8 CT counties missing ACS", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 3143, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.1289, "p": 5.05e-11, "ci95": [ 0.0907, 0.1666 ] }, "spearman": { "rho": 0.2809, "p": 5.26e-48 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1536, "p": 4.87e-15, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1034, "p": 1.96e-07, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0662, "p": 0.00133, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "women have higher age-specific dementia prevalence" }, "confounders_named": [ "age structure (women live longer)" ], "note": "", "fdr_q_within_level": 7.04e-11, "classification": "negligible", "verdict": "Raw weak positive correlation r=0.13 (95% CI 0.09 to 0.17), Spearman rho=0.28, n=2580 counties, p=5.05e-11, BH q=7e-11. NEGLIGIBLE: survives every control but the fully-controlled partial r=0.066 is below 0.10: statistically non-zero at this n, practically nothing. Not a finding worth reporting as an association. Provenance: pct_female is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-17", "level": "county", "kind": "exposure", "x": { "name": "pct_85plus", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.1316, "p": 2.08e-11, "ci95": [ -0.1694, -0.0934 ] }, "spearman": { "rho": -0.1501, "p": 1.98e-14 }, "partial_primary": { "controls": [ "median_household_income" ], "r": -0.1854, "p": 2.57e-21, "n": 2572 }, "partial_extended": { "controls": [ "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0825, "p": 3.37e-05, "n": 2521 }, "partial_strict": { "controls": [ "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0667, "p": 0.00123, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": true, "literature_expectation": { "direction": "positive", "why": "prevalence roughly doubles every five years after 65" }, "confounders_named": [ "retirement migration (healthy movers)", "diagnosis intensity", "share of very old in institutions" ], "note": "Within the 65+ denominator, more 85+ residents should mean MORE diagnosed dementia.", "fdr_q_within_level": 2.93e-11, "classification": "negligible", "verdict": "Raw weak negative correlation r=-0.13 (95% CI -0.17 to -0.09), Spearman rho=-0.15, n=2572 counties, p=2.08e-11, BH q=2.9e-11. NEGLIGIBLE: survives every control but the fully-controlled partial r=-0.067 is below 0.10: statistically non-zero at this n, practically nothing. Not a finding worth reporting as an association. SIGN FLIP versus individual-level literature (prevalence roughly doubles every five years after 65): read as a statement about measurement or place, not about risk. Within the 65+ denominator, more 85+ residents should mean MORE diagnosed dementia. Provenance: pct_85plus is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-18", "level": "county", "kind": "exposure", "x": { "name": "pct_65plus", "unit": "percent (0-100)", "year": "~2018-2022 (ACS vintage embedded in CHR 2024)", "source": "County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release via data/panel_county.csv", "tag": "survey estimate (ACS via CHR)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 3143, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.2181, "p": 3.64e-29, "ci95": [ -0.2546, -0.1811 ] }, "spearman": { "rho": -0.2163, "p": 1.07e-28 }, "partial_primary": { "controls": [ "median_household_income" ], "r": -0.3206, "p": 9.71e-63, "n": 2580 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.3043, "p": 3.68e-55, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.3478, "p": 1.04e-67, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "retirement migration", "rurality" ], "note": "", "fdr_q_within_level": 6.45e-29, "classification": "survives", "verdict": "Raw weak negative correlation r=-0.22 (95% CI -0.25 to -0.18), Spearman rho=-0.22, n=2580 counties, p=3.64e-29, BH q=6.5e-29. SURVIVES the confounder test: partial r=-0.321 (p=9.71e-63) controlling for median_household_income; extended partial r=-0.3043; strict partial r=-0.3478 controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: retirement migration, rurality. Provenance: pct_65plus is survey estimate (ACS via CHR); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-19", "level": "county", "kind": "exposure", "x": { "name": "median_age", "unit": "years", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.2188, "p": 2.97e-29, "ci95": [ -0.2553, -0.1817 ] }, "spearman": { "rho": -0.2112, "p": 2.48e-27 }, "partial_primary": { "controls": [ "median_household_income" ], "r": -0.2713, "p": 1.25e-44, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.2579, "p": 1.33e-39, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.3204, "p": 3.72e-57, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "retirement migration", "rurality" ], "note": "", "fdr_q_within_level": 5.34e-29, "classification": "survives", "verdict": "Raw weak negative correlation r=-0.22 (95% CI -0.26 to -0.18), Spearman rho=-0.21, n=2572 counties, p=2.97e-29, BH q=5.3e-29. SURVIVES the confounder test: partial r=-0.271 (p=1.25e-44) controlling for median_household_income; extended partial r=-0.2579; strict partial r=-0.3204 controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: retirement migration, rurality. Provenance: median_age is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-20", "level": "county", "kind": "exposure", "x": { "name": "pct_veterans", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.1731, "p": 9.6e-19, "ci95": [ -0.2103, -0.1353 ] }, "spearman": { "rho": -0.1863, "p": 1.61e-21 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": -0.1607, "p": 2.45e-16, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.1913, "p": 3.35e-22, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.1636, "p": 1.51e-15, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "VA care substituting for Medicare claims", "rurality" ], "note": "", "fdr_q_within_level": 1.51e-18, "classification": "survives", "verdict": "Raw weak negative correlation r=-0.17 (95% CI -0.21 to -0.14), Spearman rho=-0.19, n=2572 counties, p=9.6e-19, BH q=1.5e-18. SURVIVES the confounder test: partial r=-0.161 (p=2.45e-16) controlling for pct_85plus+median_household_income; extended partial r=-0.1913; strict partial r=-0.1636 controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: VA care substituting for Medicare claims, rurality. Provenance: pct_veterans is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-21", "level": "county", "kind": "exposure", "x": { "name": "exp_svi_theme3_minority_status_percentile", "unit": "percentile (0-1)", "year": "2022 (single cross-section; placed only on each county's 2022 row)", "source": "CDC/ATSDR Social Vulnerability Index, 2022 database, County via data/panel_exposure_county.csv", "tag": "derived index (from ACS survey estimates)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.231, "p": 1.68e-32, "ci95": [ 0.1941, 0.2673 ] }, "spearman": { "rho": 0.2651, "p": 1.29e-42 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.211, "p": 2.85e-27, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1059, "p": 9.95e-08, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0633, "p": 0.00214, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region", "poverty" ], "note": "", "fdr_q_within_level": 3.47e-32, "classification": "collapses_extended", "verdict": "Raw weak positive correlation r=0.23 (95% CI 0.19 to 0.27), Spearman rho=0.27, n=2572 counties, p=1.68e-32, BH q=3.5e-32. COLLAPSES under the extended controls: survives the primary control (partial r=0.211) but falls to 0.106 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. Provenance: exp_svi_theme3_minority_status_percentile is derived index (from ACS survey estimates); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-22", "level": "county", "kind": "exposure", "x": { "name": "pct_living_alone_65plus", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.1812, "p": 2.04e-20, "ci95": [ 0.1435, 0.2183 ] }, "spearman": { "rho": 0.1842, "p": 4.52e-21 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.0782, "p": 7.2e-05, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1251, "p": 2.95e-10, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1186, "p": 8.33e-09, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "social isolation is a Lancet Commission risk factor" }, "confounders_named": [ "age structure", "institutional population is excluded from 'living alone'" ], "note": "", "fdr_q_within_level": 3.33e-20, "classification": "collapses", "verdict": "Raw weak positive correlation r=0.18 (95% CI 0.14 to 0.22), Spearman rho=0.18, n=2572 counties, p=2.04e-20, BH q=3.3e-20. COLLAPSES: partial r=0.078 is under half the raw r=0.181 once pct_85plus+median_household_income is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_living_alone_65plus. Provenance: pct_living_alone_65plus is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-23", "level": "county", "kind": "exposure", "x": { "name": "pct_single_person_households", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.0719, "p": 0.000261, "ci95": [ 0.0334, 0.1103 ] }, "spearman": { "rho": 0.0678, "p": 0.000583 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": -0.047, "p": 0.0171, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0352, "p": 0.0772, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0961, "p": 3.12e-06, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure", "urbanicity" ], "note": "", "fdr_q_within_level": 0.000308, "classification": "negligible", "verdict": "Raw negligible positive correlation r=0.07 (95% CI 0.03 to 0.11), Spearman rho=0.07, n=2572 counties, p=0.000261, BH q=0.00031. NEGLIGIBLE: raw r=0.072 is statistically non-zero at n=2572 but below 0.10 in size; too small to interpret. Not a finding worth reporting as an association. Provenance: pct_single_person_households is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-24", "level": "county", "kind": "exposure", "x": { "name": "social_association_index", "unit": "count per 10,000 population", "year": "~2022 (County Business Patterns vintage embedded in CHR 2024)", "source": "County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release via data/panel_county.csv", "tag": "administrative compilation (CHR pass-through)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 3143, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.0304, "p": 0.123, "ci95": [ -0.0689, 0.0082 ] }, "spearman": { "rho": -0.0214, "p": 0.277 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": -0.0016, "p": 0.936, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0226, "p": 0.257, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0531, "p": 0.0101, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "negative", "why": "social engagement is protective" }, "confounders_named": [ "rurality (small counties have high per-capita association counts)", "income" ], "note": "", "fdr_q_within_level": 0.134, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.03 (95% CI -0.07 to 0.01), Spearman rho=-0.02, n=2580 counties, p=0.123, BH q=0.13. NULL: raw association not distinguishable from zero (BH q=0.13 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: social_association_index is administrative compilation (CHR pass-through); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-25", "level": "county", "kind": "exposure", "x": { "name": "exp_svi_theme2_household_percentile", "unit": "percentile (0-1)", "year": "2022 (single cross-section; placed only on each county's 2022 row)", "source": "CDC/ATSDR Social Vulnerability Index, 2022 database, County via data/panel_exposure_county.csv", "tag": "derived index (from ACS survey estimates)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.323, "p": 1.62e-63, "ci95": [ 0.2879, 0.3572 ] }, "spearman": { "rho": 0.3373, "p": 1.9e-69 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1908, "p": 1.62e-22, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1348, "p": 1.07e-11, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1657, "p": 6.51e-16, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure (theme 2 includes % 65+ and disability)", "poverty" ], "note": "", "fdr_q_within_level": 5.15e-63, "classification": "collapses_extended", "verdict": "Raw moderate positive correlation r=0.32 (95% CI 0.29 to 0.36), Spearman rho=0.34, n=2572 counties, p=1.62e-63, BH q=5.1e-63. COLLAPSES under the extended controls: survives the primary control (partial r=0.191) but falls to 0.135 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. Provenance: exp_svi_theme2_household_percentile is derived index (from ACS survey estimates); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-26", "level": "county", "kind": "exposure", "x": { "name": "pct_diabetes", "unit": "percent (0-100)", "year": "2023 (BRFSS)", "source": "CDC PLACES: County Data (GIS-Friendly Format), 2025 release via data/panel_county.csv", "tag": "modelled (PLACES small-area estimate)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2392, "n_unit": "counties", "x_filled": 2947, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.3784, "p": 2.62e-82, "ci95": [ 0.3436, 0.4123 ] }, "spearman": { "rho": 0.3891, "p": 2.65e-87 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0843, "p": 4.34e-05, "n": 2348 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0843, "p": 4.34e-05, "n": 2348 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0114, "p": 0.582, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "midlife diabetes raises dementia risk" }, "confounders_named": [ "poverty", "region", "obesity/inactivity (r>0.85 with these)", "PLACES model shares ACS inputs with the controls" ], "note": "", "fdr_q_within_level": 1.2e-81, "classification": "collapses", "verdict": "Raw moderate positive correlation r=0.38 (95% CI 0.34 to 0.41), Spearman rho=0.39, n=2392 counties, p=2.62e-82, BH q=1.2e-81. COLLAPSES: partial r=0.084 is under half the raw r=0.378 once pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_diabetes. Provenance: pct_diabetes is modelled (PLACES small-area estimate); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-27", "level": "county", "kind": "exposure", "x": { "name": "pct_obesity", "unit": "percent (0-100)", "year": "2023 (BRFSS)", "source": "CDC PLACES: County Data (GIS-Friendly Format), 2025 release via data/panel_county.csv", "tag": "modelled (PLACES small-area estimate)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2392, "n_unit": "counties", "x_filled": 2947, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.3461, "p": 2.85e-68, "ci95": [ 0.3103, 0.3809 ] }, "spearman": { "rho": 0.3431, "p": 4.58e-67 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1616, "p": 3.39e-15, "n": 2348 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1616, "p": 3.39e-15, "n": 2348 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.144, "p": 2.42e-12, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "midlife obesity raises dementia risk" }, "confounders_named": [ "poverty", "region", "diabetes" ], "note": "", "fdr_q_within_level": 1.01e-67, "classification": "collapses", "verdict": "Raw moderate positive correlation r=0.35 (95% CI 0.31 to 0.38), Spearman rho=0.34, n=2392 counties, p=2.85e-68, BH q=1e-67. COLLAPSES: partial r=0.162 is under half the raw r=0.346 once pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_obesity. Provenance: pct_obesity is modelled (PLACES small-area estimate); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-28", "level": "county", "kind": "exposure", "x": { "name": "pct_hypertension", "unit": "percent (0-100)", "year": "2023 (BRFSS)", "source": "CDC PLACES: County Data (GIS-Friendly Format), 2025 release via data/panel_county.csv", "tag": "modelled (PLACES small-area estimate)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2392, "n_unit": "counties", "x_filled": 2947, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.3575, "p": 4.77e-73, "ci95": [ 0.3221, 0.392 ] }, "spearman": { "rho": 0.3642, "p": 6.05e-76 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0736, "p": 0.000358, "n": 2348 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0736, "p": 0.000358, "n": 2348 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0111, "p": 0.591, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "midlife hypertension raises dementia risk" }, "confounders_named": [ "poverty", "region", "diabetes", "age structure" ], "note": "", "fdr_q_within_level": 1.85e-72, "classification": "collapses", "verdict": "Raw moderate positive correlation r=0.36 (95% CI 0.32 to 0.39), Spearman rho=0.36, n=2392 counties, p=4.77e-73, BH q=1.8e-72. COLLAPSES: partial r=0.074 is under half the raw r=0.357 once pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_hypertension. Provenance: pct_hypertension is modelled (PLACES small-area estimate); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-29", "level": "county", "kind": "exposure", "x": { "name": "pct_smoking", "unit": "percent (0-100)", "year": "2023 (BRFSS)", "source": "CDC PLACES: County Data (GIS-Friendly Format), 2025 release via data/panel_county.csv", "tag": "modelled (PLACES small-area estimate)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2392, "n_unit": "counties", "x_filled": 2947, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.3253, "p": 4.57e-60, "ci95": [ 0.289, 0.3607 ] }, "spearman": { "rho": 0.3451, "p": 7.14e-68 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1101, "p": 8.97e-08, "n": 2348 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1101, "p": 8.97e-08, "n": 2348 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1086, "p": 1.32e-07, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "smoking raises dementia risk" }, "confounders_named": [ "poverty", "region", "competing mortality (smokers die before dementia age)" ], "note": "", "fdr_q_within_level": 1.35e-59, "classification": "collapses", "verdict": "Raw moderate positive correlation r=0.33 (95% CI 0.29 to 0.36), Spearman rho=0.35, n=2392 counties, p=4.57e-60, BH q=1.3e-59. COLLAPSES: partial r=0.110 is under half the raw r=0.325 once pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_smoking. Provenance: pct_smoking is modelled (PLACES small-area estimate); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-30", "level": "county", "kind": "exposure", "x": { "name": "pct_physical_inactivity", "unit": "percent (0-100)", "year": "2023 (BRFSS)", "source": "CDC PLACES: County Data (GIS-Friendly Format), 2025 release via data/panel_county.csv", "tag": "modelled (PLACES small-area estimate)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2392, "n_unit": "counties", "x_filled": 2947, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.4513, "p": 2.31e-120, "ci95": [ 0.4188, 0.4826 ] }, "spearman": { "rho": 0.4436, "p": 6.52e-116 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.2586, "p": 3.53e-37, "n": 2348 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.2586, "p": 3.53e-37, "n": 2348 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.2274, "p": 6.44e-29, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "inactivity raises dementia risk" }, "confounders_named": [ "poverty", "region", "diabetes" ], "note": "", "fdr_q_within_level": 1.79e-119, "classification": "survives", "verdict": "Raw moderate positive correlation r=0.45 (95% CI 0.42 to 0.48), Spearman rho=0.44, n=2392 counties, p=2.31e-120, BH q=1.8e-119. SURVIVES the confounder test: partial r=0.259 (p=3.53e-37) controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south; extended partial r=0.2586; strict partial r=0.2274 controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: poverty, region, diabetes. Provenance: pct_physical_inactivity is modelled (PLACES small-area estimate); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-31", "level": "county", "kind": "exposure", "x": { "name": "pct_depression", "unit": "percent (0-100)", "year": "2023 (BRFSS)", "source": "CDC PLACES: County Data (GIS-Friendly Format), 2025 release via data/panel_county.csv", "tag": "modelled (PLACES small-area estimate)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2392, "n_unit": "counties", "x_filled": 2947, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.0086, "p": 0.675, "ci95": [ -0.0486, 0.0315 ] }, "spearman": { "rho": 0.0041, "p": 0.842 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.1096, "p": 1.02e-07, "n": 2348 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.1096, "p": 1.02e-07, "n": 2348 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.076, "p": 0.000229, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "depression is a risk factor and prodrome" }, "confounders_named": [ "healthcare contact (depression is diagnosed where people see doctors)", "region" ], "note": "", "fdr_q_within_level": 0.703, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.01 (95% CI -0.05 to 0.03), Spearman rho=0.00, n=2392 counties, p=0.675, BH q=0.7. NULL: raw association not distinguishable from zero (BH q=0.7 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_depression is modelled (PLACES small-area estimate); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-32", "level": "county", "kind": "exposure", "x": { "name": "pct_binge_drinking", "unit": "percent (0-100)", "year": "2023 (BRFSS)", "source": "CDC PLACES: County Data (GIS-Friendly Format), 2025 release via data/panel_county.csv", "tag": "modelled (PLACES small-area estimate)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2392, "n_unit": "counties", "x_filled": 2947, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.2435, "p": 1.27e-33, "ci95": [ -0.2808, -0.2054 ] }, "spearman": { "rho": -0.2609, "p": 1.65e-38 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0481, "p": 0.0199, "n": 2348 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0481, "p": 0.0199, "n": 2348 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1003, "p": 1.12e-06, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region", "age structure (younger counties drink more)", "income" ], "note": "", "fdr_q_within_level": 2.67e-33, "classification": "collapses", "verdict": "Raw weak negative correlation r=-0.24 (95% CI -0.28 to -0.21), Spearman rho=-0.26, n=2392 counties, p=1.27e-33, BH q=2.7e-33. COLLAPSES: partial r=0.048 is under half the raw r=-0.243 once pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_binge_drinking. Provenance: pct_binge_drinking is modelled (PLACES small-area estimate); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-33", "level": "county", "kind": "exposure", "x": { "name": "pct_fair_poor_health", "unit": "percent (0-100)", "year": "2023 (BRFSS)", "source": "CDC PLACES: County Data (GIS-Friendly Format), 2025 release via data/panel_county.csv", "tag": "modelled (PLACES small-area estimate)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2392, "n_unit": "counties", "x_filled": 2947, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.4275, "p": 6.73e-107, "ci95": [ 0.3942, 0.4597 ] }, "spearman": { "rho": 0.44, "p": 8.36e-114 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1696, "p": 1.31e-16, "n": 2348 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1696, "p": 1.31e-16, "n": 2348 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1396, "p": 1.11e-11, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "poverty", "region", "diabetes cluster" ], "note": "", "fdr_q_within_level": 4.17e-106, "classification": "collapses", "verdict": "Raw moderate positive correlation r=0.43 (95% CI 0.39 to 0.46), Spearman rho=0.44, n=2392 counties, p=6.73e-107, BH q=4.2e-106. COLLAPSES: partial r=0.170 is under half the raw r=0.427 once pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_fair_poor_health. Provenance: pct_fair_poor_health is modelled (PLACES small-area estimate); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-34", "level": "county", "kind": "exposure", "x": { "name": "pct_65plus_with_disability", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.3876, "p": 6.05e-93, "ci95": [ 0.3542, 0.4199 ] }, "spearman": { "rho": 0.4071, "p": 2.82e-103 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1567, "p": 2.47e-15, "n": 2521 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1567, "p": 2.47e-15, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1915, "p": 8.1e-21, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "poverty", "region", "dementia itself causes disability (reverse direction)" ], "note": "", "fdr_q_within_level": 2.89e-92, "classification": "collapses", "verdict": "Raw moderate positive correlation r=0.39 (95% CI 0.35 to 0.42), Spearman rho=0.41, n=2572 counties, p=6.05e-93, BH q=2.9e-92. COLLAPSES: partial r=0.157 is under half the raw r=0.388 once pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_65plus_with_disability. Provenance: pct_65plus_with_disability is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-35", "level": "county", "kind": "exposure", "x": { "name": "pm25_avg_ugm3", "unit": "micrograms per cubic meter (ug/m3)", "year": "~2021 (CDC/EPA Environmental Public Health Tracking vintage embedded in CHR 2024)", "source": "County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release via data/panel_county.csv", "tag": "modelled (CDC/EPA modelled PM2.5 via CHR)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2579, "n_unit": "counties", "x_filled": 3115, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.3529, "p": 1.61e-76, "ci95": [ 0.3187, 0.3863 ] }, "spearman": { "rho": 0.3643, "p": 9.08e-82 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.2682, "p": 8.81e-43, "n": 2520 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.2682, "p": 8.81e-43, "n": 2520 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.2307, "p": 1.05e-29, "n": 2346 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "PM2.5 is a Lancet Commission risk factor" }, "confounders_named": [ "urban density", "region (Southeast PM2.5 is high)", "modelled input" ], "note": "Same variable the baseline C1 tested against case fatality; here against prevalence.", "fdr_q_within_level": 6.65e-76, "classification": "survives", "verdict": "Raw moderate positive correlation r=0.35 (95% CI 0.32 to 0.39), Spearman rho=0.36, n=2579 counties, p=1.61e-76, BH q=6.7e-76. SURVIVES the confounder test: partial r=0.268 (p=8.81e-43) controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south; extended partial r=0.2682; strict partial r=0.2307 controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: urban density, region (Southeast PM2.5 is high), modelled input. Same variable the baseline C1 tested against case fatality; here against prevalence. Provenance: pm25_avg_ugm3 is modelled (CDC/EPA modelled PM2.5 via CHR); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-36", "level": "county", "kind": "exposure", "x": { "name": "exp_pm25_modelled_mean_2016_2019", "unit": "mean of the four annual modelled population-weighted PM2.5 means 2016-2019, ug/m3", "year": "2016-2019", "source": "CDC Environmental Public Health Tracking downscaler via data/panel_exposure_county.csv", "tag": "modelled (statistical downscaler)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2571, "n_unit": "counties", "x_filled": 3107, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.3991, "p": 6.44e-99, "ci95": [ 0.3661, 0.4311 ] }, "spearman": { "rho": 0.4076, "p": 1.71e-103 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.3154, "p": 3.86e-59, "n": 2512 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.3154, "p": 3.86e-59, "n": 2512 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.2839, "p": 1.32e-44, "n": 2338 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "PM2.5 is a Lancet Commission risk factor" }, "confounders_named": [ "urban density", "region" ], "note": "Lagged: exposure years 2016-2019 all precede the 2020 outcome.", "fdr_q_within_level": 3.33e-98, "classification": "survives", "verdict": "Raw moderate positive correlation r=0.40 (95% CI 0.37 to 0.43), Spearman rho=0.41, n=2571 counties, p=6.44e-99, BH q=3.3e-98. SURVIVES the confounder test: partial r=0.315 (p=3.86e-59) controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south; extended partial r=0.3154; strict partial r=0.2839 controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: urban density, region. Lagged: exposure years 2016-2019 all precede the 2020 outcome. Provenance: exp_pm25_modelled_mean_2016_2019 is modelled (statistical downscaler); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-37", "level": "county", "kind": "exposure", "x": { "name": "exp_pm25_monitor_mean_2016_2019", "unit": "mean of the four annual monitor-based PM2.5 means 2016-2019, ug/m3 (all four years required)", "year": "2016-2019", "source": "US EPA AQS annual_conc_by_monitor via data/panel_exposure_county.csv", "tag": "measured (physical monitor)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 559, "n_unit": "counties", "x_filled": 585, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.1752, "p": 3.11e-05, "ci95": [ 0.0936, 0.2544 ] }, "spearman": { "rho": 0.2018, "p": 1.51e-06 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1363, "p": 0.0013, "n": 554 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1363, "p": 0.0013, "n": 554 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0527, "p": 0.233, "n": 513 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "PM2.5 is a Lancet Commission risk factor" }, "confounders_named": [ "monitors sit in urban/industrial counties (selection)", "region" ], "note": "Measured, lagged, but only counties with a monitor in all four years (selection towards urban counties).", "fdr_q_within_level": 3.78e-05, "classification": "collapses_strict", "verdict": "Raw weak positive correlation r=0.18 (95% CI 0.09 to 0.25), Spearman rho=0.20, n=559 counties, p=3.11e-05, BH q=3.8e-05. COLLAPSES under place / diagnosis-intensity controls: survives age/income/region controls (extended partial r=0.1363) but falls to 0.053 once place and diagnosis-intensity markers are added (pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual). What looked like an exposure effect is carried by where the county is and how intensively it codes diagnoses. Measured, lagged, but only counties with a monitor in all four years (selection towards urban counties). Provenance: exp_pm25_monitor_mean_2016_2019 is measured (physical monitor); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-38", "level": "county", "kind": "exposure", "x": { "name": "exp_ozone_monitor_mean_2016_2019", "unit": "mean of four annual monitor-based ozone means 2016-2019, ppm", "year": "2016-2019", "source": "US EPA AQS via data/panel_exposure_county.csv", "tag": "measured (physical monitor)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 720, "n_unit": "counties", "x_filled": 757, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.1726, "p": 3.19e-06, "ci95": [ -0.2426, -0.1008 ] }, "spearman": { "rho": -0.1363, "p": 0.000244 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.1352, "p": 0.00031, "n": 708 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.1352, "p": 0.00031, "n": 708 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0863, "p": 0.028, "n": 648 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "monitor siting", "climate/latitude", "urbanicity" ], "note": "", "fdr_q_within_level": 4.04e-06, "classification": "negligible", "verdict": "Raw weak negative correlation r=-0.17 (95% CI -0.24 to -0.10), Spearman rho=-0.14, n=720 counties, p=3.19e-06, BH q=4e-06. NEGLIGIBLE: survives every control but the fully-controlled partial r=-0.086 is below 0.10: statistically non-zero at this n, practically nothing. Not a finding worth reporting as an association. Provenance: exp_ozone_monitor_mean_2016_2019 is measured (physical monitor); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-39", "level": "county", "kind": "exposure", "x": { "name": "drinking_water_violations", "unit": "binary (1 = had a violation, 0 = did not)", "year": "~2022 (embedded in CHR 2024)", "source": "County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release via data/panel_county.csv", "tag": "administrative compilation (CHR pass-through)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2540, "n_unit": "counties", "x_filled": 3091, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.0269, "p": 0.176, "ci95": [ -0.0657, 0.012 ] }, "spearman": { "rho": -0.0155, "p": 0.434 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0189, "p": 0.347, "n": 2486 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0189, "p": 0.347, "n": 2486 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0127, "p": 0.543, "n": 2312 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "rurality (small systems violate more)", "region" ], "note": "Binary flag, not a dose.", "fdr_q_within_level": 0.188, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.03 (95% CI -0.07 to 0.01), Spearman rho=-0.02, n=2540 counties, p=0.176, BH q=0.19. NULL: raw association not distinguishable from zero (BH q=0.19 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Binary flag, not a dose. Provenance: drinking_water_violations is administrative compilation (CHR pass-through); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-40", "level": "county", "kind": "exposure", "x": { "name": "population_density_per_sqmi", "unit": "persons per square mile", "year": "2023 (population) / 2024 (land area)", "source": "Derived: population (see 'population' column) divided by ALAND_SQMI from the 2024 Census Gazetteer file via data/panel_county.csv", "tag": "modelled/derived" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.0599, "p": 0.00238, "ci95": [ 0.0213, 0.0983 ] }, "spearman": { "rho": 0.14, "p": 9.83e-13 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1198, "p": 1.62e-09, "n": 2521 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1198, "p": 1.62e-09, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0986, "p": 1.69e-06, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "urbanicity", "income", "specialist supply" ], "note": "", "fdr_q_within_level": 0.00276, "classification": "negligible", "verdict": "Raw negligible positive correlation r=0.06 (95% CI 0.02 to 0.10), Spearman rho=0.14, n=2572 counties, p=0.00238, BH q=0.0028. NEGLIGIBLE: raw r=0.060 is statistically non-zero at n=2572 but below 0.10 in size; too small to interpret. Not a finding worth reporting as an association. Provenance: population_density_per_sqmi is modelled/derived; norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-41", "level": "county", "kind": "exposure", "x": { "name": "county_type_metro", "unit": "binary (1/0)", "year": "2023", "source": "USDA Economic Research Service, Rural-Urban Continuum Codes 2023 via data/panel_county.csv", "tag": "modelled/derived" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.0009, "p": 0.964, "ci95": [ -0.0378, 0.0395 ] }, "spearman": { "rho": 0.0295, "p": 0.135 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1565, "p": 1.43e-15, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1156, "p": 5.85e-09, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0675, "p": 0.00106, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "income", "specialist supply", "region" ], "note": "Binary metro flag from USDA RUCC.", "fdr_q_within_level": 0.964, "classification": "null", "verdict": "Raw negligible positive correlation r=0.00 (95% CI -0.04 to 0.04), Spearman rho=0.03, n=2572 counties, p=0.964, BH q=0.96. NULL: raw association not distinguishable from zero (BH q=0.96 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Binary metro flag from USDA RUCC. Provenance: county_type_metro is modelled/derived; norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-42", "level": "county", "kind": "exposure", "x": { "name": "exp_pct_seniors_low_access_2019", "unit": "percent", "year": "2019 (single cross-section; placed only on each county's 2019 row)", "source": "USDA ERS Food Environment Atlas 2025 release, variable PCT_LACCESS_SENIORS19 via data/panel_exposure_county.csv", "tag": "measured/administrative (USDA establishment counts)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 3139, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.0092, "p": 0.642, "ci95": [ -0.0294, 0.0477 ] }, "spearman": { "rho": 0.0182, "p": 0.357 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.179, "p": 1.37e-19, "n": 2521 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.179, "p": 1.37e-19, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.1794, "p": 1.98e-18, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "rurality (distance thresholds are rural)", "region" ], "note": "", "fdr_q_within_level": 0.68, "classification": "null", "verdict": "Raw negligible positive correlation r=0.01 (95% CI -0.03 to 0.05), Spearman rho=0.02, n=2580 counties, p=0.642, BH q=0.68. NULL: raw association not distinguishable from zero (BH q=0.68 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: exp_pct_seniors_low_access_2019 is measured/administrative (USDA establishment counts); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-43", "level": "county", "kind": "exposure", "x": { "name": "exp_pct_low_access_to_store_2019", "unit": "percent", "year": "2019 (single cross-section; placed only on each county's 2019 row)", "source": "USDA ERS Food Environment Atlas 2025 release, variable PCT_LACCESS_POP19 via data/panel_exposure_county.csv", "tag": "measured/administrative (USDA establishment counts)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 3141, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.0061, "p": 0.755, "ci95": [ -0.0447, 0.0324 ] }, "spearman": { "rho": 0.0028, "p": 0.887 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0307, "p": 0.123, "n": 2521 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0307, "p": 0.123, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0475, "p": 0.0215, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "rurality", "region" ], "note": "", "fdr_q_within_level": 0.774, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.01 (95% CI -0.04 to 0.03), Spearman rho=0.00, n=2580 counties, p=0.755, BH q=0.77. NULL: raw association not distinguishable from zero (BH q=0.77 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: exp_pct_low_access_to_store_2019 is measured/administrative (USDA establishment counts); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-44", "level": "county", "kind": "exposure", "x": { "name": "exp_pct_low_income_low_access_2019", "unit": "percent", "year": "2019 (single cross-section; placed only on each county's 2019 row)", "source": "USDA ERS Food Environment Atlas 2025 release, variable PCT_LACCESS_LOWI19 via data/panel_exposure_county.csv", "tag": "measured/administrative (USDA establishment counts)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 3141, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.1516, "p": 9.86e-15, "ci95": [ 0.1137, 0.1891 ] }, "spearman": { "rho": 0.1558, "p": 1.72e-15 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0424, "p": 0.0334, "n": 2521 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0424, "p": 0.0334, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0548, "p": 0.00795, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "poverty", "rurality" ], "note": "", "fdr_q_within_level": 1.49e-14, "classification": "collapses", "verdict": "Raw weak positive correlation r=0.15 (95% CI 0.11 to 0.19), Spearman rho=0.16, n=2580 counties, p=9.86e-15, BH q=1.5e-14. COLLAPSES: partial r=0.042 is under half the raw r=0.152 once pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with exp_pct_low_income_low_access_2019. Provenance: exp_pct_low_income_low_access_2019 is measured/administrative (USDA establishment counts); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-45", "level": "county", "kind": "exposure", "x": { "name": "exp_dollar_stores_per_1000pop_2020", "unit": "stores per 1,000 population", "year": "2020 (single cross-section; placed only on each county's 2020 row)", "source": "USDA ERS Food Environment Atlas 2025 release, variable DSPTH20 via data/panel_exposure_county.csv", "tag": "measured/administrative (USDA establishment counts)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 3143, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.2983, "p": 3.55e-54, "ci95": [ 0.2627, 0.3331 ] }, "spearman": { "rho": 0.3061, "p": 4.27e-57 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0822, "p": 3.59e-05, "n": 2521 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0822, "p": 3.59e-05, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.054, "p": 0.00883, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "poverty", "region (South has the densest dollar-store networks)", "rurality" ], "note": "", "fdr_q_within_level": 9.37e-54, "classification": "collapses", "verdict": "Raw weak positive correlation r=0.30 (95% CI 0.26 to 0.33), Spearman rho=0.31, n=2580 counties, p=3.55e-54, BH q=9.4e-54. COLLAPSES: partial r=0.082 is under half the raw r=0.298 once pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with exp_dollar_stores_per_1000pop_2020. Provenance: exp_dollar_stores_per_1000pop_2020 is measured/administrative (USDA establishment counts); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-46", "level": "county", "kind": "exposure", "x": { "name": "exp_grocery_stores_per_1000pop_2020", "unit": "stores per 1,000 population", "year": "2020 (single cross-section; placed only on each county's 2020 row)", "source": "USDA ERS Food Environment Atlas 2025 release, variable GROCPTH20 via data/panel_exposure_county.csv", "tag": "measured/administrative (USDA establishment counts)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2078, "n_unit": "counties", "x_filled": 2234, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.1382, "p": 2.48e-10, "ci95": [ -0.1802, -0.0958 ] }, "spearman": { "rho": -0.1253, "p": 1.01e-08 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.1121, "p": 3.45e-07, "n": 2056 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.1121, "p": 3.45e-07, "n": 2056 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.1517, "p": 2.48e-11, "n": 1916 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "rurality (small counties have high per-capita store counts)", "suppression: blank in 913 counties" ], "note": "", "fdr_q_within_level": 3.42e-10, "classification": "survives", "verdict": "Raw weak negative correlation r=-0.14 (95% CI -0.18 to -0.10), Spearman rho=-0.13, n=2078 counties, p=2.48e-10, BH q=3.4e-10. SURVIVES the confounder test: partial r=-0.112 (p=3.45e-07) controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south; extended partial r=-0.1121; strict partial r=-0.1517 controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: rurality (small counties have high per-capita store counts), suppression: blank in 913 counties. Provenance: exp_grocery_stores_per_1000pop_2020 is measured/administrative (USDA establishment counts); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-47", "level": "county", "kind": "exposure", "x": { "name": "exp_fastfood_restaurants_per_1000pop_2020", "unit": "restaurants per 1,000 population", "year": "2020 (single cross-section; placed only on each county's 2020 row)", "source": "USDA ERS Food Environment Atlas 2025 release, variable FFRPTH20 via data/panel_exposure_county.csv", "tag": "measured/administrative (USDA establishment counts)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2465, "n_unit": "counties", "x_filled": 2672, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.1053, "p": 1.62e-07, "ci95": [ 0.0661, 0.1442 ] }, "spearman": { "rho": 0.1423, "p": 1.27e-12 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.108, "p": 9.7e-08, "n": 2424 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.108, "p": 9.7e-08, "n": 2424 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0851, "p": 5.16e-05, "n": 2256 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "urbanicity", "region" ], "note": "", "fdr_q_within_level": 2.14e-07, "classification": "negligible", "verdict": "Raw weak positive correlation r=0.11 (95% CI 0.07 to 0.14), Spearman rho=0.14, n=2465 counties, p=1.62e-07, BH q=2.1e-07. NEGLIGIBLE: survives every control but the fully-controlled partial r=0.085 is below 0.10: statistically non-zero at this n, practically nothing. Not a finding worth reporting as an association. Provenance: exp_fastfood_restaurants_per_1000pop_2020 is measured/administrative (USDA establishment counts); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-48", "level": "county", "kind": "exposure", "x": { "name": "exp_fullservice_restaurants_per_1000pop_2020", "unit": "restaurants per 1,000 population", "year": "2020 (single cross-section; placed only on each county's 2020 row)", "source": "USDA ERS Food Environment Atlas 2025 release, variable FSRPTH20 via data/panel_exposure_county.csv", "tag": "measured/administrative (USDA establishment counts)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2494, "n_unit": "counties", "x_filled": 2861, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.2992, "p": 9.51e-53, "ci95": [ -0.3345, -0.263 ] }, "spearman": { "rho": -0.3121, "p": 1.75e-57 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.2287, "p": 1.94e-30, "n": 2450 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.2287, "p": 1.94e-30, "n": 2450 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.2232, "p": 3.42e-27, "n": 2285 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "tourism/rurality", "income" ], "note": "", "fdr_q_within_level": 2.46e-52, "classification": "survives", "verdict": "Raw weak negative correlation r=-0.30 (95% CI -0.33 to -0.26), Spearman rho=-0.31, n=2494 counties, p=9.51e-53, BH q=2.5e-52. SURVIVES the confounder test: partial r=-0.229 (p=1.94e-30) controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south; extended partial r=-0.2287; strict partial r=-0.2232 controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: tourism/rurality, income. Provenance: exp_fullservice_restaurants_per_1000pop_2020 is measured/administrative (USDA establishment counts); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-49", "level": "county", "kind": "exposure", "x": { "name": "exp_convenience_stores_per_1000pop_2020", "unit": "stores per 1,000 population", "year": "2020 (single cross-section; placed only on each county's 2020 row)", "source": "USDA ERS Food Environment Atlas 2025 release, variable CONVSPTH20 via data/panel_exposure_county.csv", "tag": "measured/administrative (USDA establishment counts)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2542, "n_unit": "counties", "x_filled": 2859, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.139, "p": 1.96e-12, "ci95": [ 0.1006, 0.1769 ] }, "spearman": { "rho": 0.1643, "p": 7.67e-17 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0728, "p": 0.000279, "n": 2489 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0728, "p": 0.000279, "n": 2489 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.114, "p": 3.66e-08, "n": 2319 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "rurality", "region" ], "note": "", "fdr_q_within_level": 2.89e-12, "classification": "sign_reverses", "verdict": "Raw weak positive correlation r=0.14 (95% CI 0.10 to 0.18), Spearman rho=0.16, n=2542 counties, p=1.96e-12, BH q=2.9e-12. SIGN REVERSES under control: partial r=-0.073 has the opposite sign to raw r=0.139: the raw association was confounded in the other direction. Provenance: exp_convenience_stores_per_1000pop_2020 is measured/administrative (USDA establishment counts); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-50", "level": "county", "kind": "exposure", "x": { "name": "exp_snap_authorized_stores_per_1000pop_2023", "unit": "stores per 1,000 population", "year": "2023 (single cross-section; placed only on each county's 2023 row)", "source": "USDA ERS Food Environment Atlas 2025 release, variable SNAPSPTH23 via data/panel_exposure_county.csv", "tag": "measured/administrative (USDA establishment counts)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3114, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.2265, "p": 2.77e-31, "ci95": [ 0.1895, 0.2629 ] }, "spearman": { "rho": 0.2481, "p": 2.23e-37 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0378, "p": 0.0579, "n": 2521 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0378, "p": 0.0579, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0779, "p": 0.000157, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "poverty", "rurality" ], "note": "", "fdr_q_within_level": 5.45e-31, "classification": "collapses", "verdict": "Raw weak positive correlation r=0.23 (95% CI 0.19 to 0.26), Spearman rho=0.25, n=2572 counties, p=2.77e-31, BH q=5.4e-31. COLLAPSES: partial r=-0.038 is under half the raw r=0.227 once pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with exp_snap_authorized_stores_per_1000pop_2023. Provenance: exp_snap_authorized_stores_per_1000pop_2023 is measured/administrative (USDA establishment counts); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-51", "level": "county", "kind": "exposure", "x": { "name": "hospital_beds_per_1000", "unit": "beds per 1,000 population", "year": "2026 snapshot (facility roster) / population as dated above", "source": "CMS Provider of Services File - Quality Improvement and Evaluation System (QIES), current quarterly snapshot pulled 2026-09-21 via data/panel_county.csv", "tag": "measured/administrative" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2146, "n_unit": "counties", "x_filled": 2457, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.1108, "p": 2.65e-07, "ci95": [ 0.0688, 0.1524 ] }, "spearman": { "rho": 0.2337, "p": 5.31e-28 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.0816, "p": 0.00016, "n": 2138 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0536, "p": 0.0133, "n": 2130 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0305, "p": 0.173, "n": 1998 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "rurality (critical access hospitals)", "diagnosis intensity" ], "note": "", "fdr_q_within_level": 3.46e-07, "classification": "collapses_extended", "verdict": "Raw weak positive correlation r=0.11 (95% CI 0.07 to 0.15), Spearman rho=0.23, n=2146 counties, p=2.65e-07, BH q=3.5e-07. COLLAPSES under the extended controls: survives the primary control (partial r=0.082) but falls to 0.054 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. Provenance: hospital_beds_per_1000 is measured/administrative; norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-52", "level": "county", "kind": "exposure", "x": { "name": "exp_hospital_beds_per_1000pop_2023", "unit": "beds per 1,000 population", "year": "2023 (single cross-section; placed only on each county's 2023 row)", "source": "Derived: hospital_beds_count_2023 / AHRF popn_est_23 * 1000 via data/panel_exposure_county.csv", "tag": "modelled/derived" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.1236, "p": 3.23e-10, "ci95": [ 0.0853, 0.1614 ] }, "spearman": { "rho": 0.1579, "p": 7.91e-16 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1078, "p": 4.3e-08, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.094, "p": 2.29e-06, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0873, "p": 2.3e-05, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "rurality", "diagnosis intensity" ], "note": "", "fdr_q_within_level": 4.4e-10, "classification": "negligible", "verdict": "Raw weak positive correlation r=0.12 (95% CI 0.09 to 0.16), Spearman rho=0.16, n=2572 counties, p=3.23e-10, BH q=4.4e-10. NEGLIGIBLE: survives every control but the fully-controlled partial r=0.087 is below 0.10: statistically non-zero at this n, practically nothing. Not a finding worth reporting as an association. Provenance: exp_hospital_beds_per_1000pop_2023 is modelled/derived; norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-53", "level": "county", "kind": "exposure", "x": { "name": "mental_health_providers_per_100k", "unit": "rate per 100,000", "year": "~2023 (CMS National Provider Identification vintage embedded in CHR 2024)", "source": "County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release via data/panel_county.csv", "tag": "administrative compilation (CHR pass-through)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2552, "n_unit": "counties", "x_filled": 2958, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.098, "p": 7.01e-07, "ci95": [ -0.1363, -0.0594 ] }, "spearman": { "rho": -0.1385, "p": 2.08e-12 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": -0.0445, "p": 0.0248, "n": 2544 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0153, "p": 0.445, "n": 2498 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0441, "p": 0.0337, "n": 2324 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "urbanicity", "income", "diagnosis intensity" ], "note": "", "fdr_q_within_level": 9.05e-07, "classification": "negligible", "verdict": "Raw negligible negative correlation r=-0.10 (95% CI -0.14 to -0.06), Spearman rho=-0.14, n=2552 counties, p=7.01e-07, BH q=9.1e-07. NEGLIGIBLE: raw r=-0.098 is statistically non-zero at n=2552 but below 0.10 in size; too small to interpret. Not a finding worth reporting as an association. Provenance: mental_health_providers_per_100k is administrative compilation (CHR pass-through); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CD-1", "level": "county", "kind": "diagnosis_intensity", "x": { "name": "pct_checkup_annual", "unit": "percent (0-100)", "year": "2023 (BRFSS)", "source": "CDC PLACES: County Data (GIS-Friendly Format), 2025 release via data/panel_county.csv", "tag": "modelled (PLACES small-area estimate)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2392, "n_unit": "counties", "x_filled": 2947, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.2292, "p": 6.99e-30, "ci95": [ 0.1909, 0.2668 ] }, "spearman": { "rho": 0.2272, "p": 2.26e-29 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.21, "p": 3.04e-25, "n": 2392 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1116, "p": 5.85e-08, "n": 2348 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "health-system contact", "coding practice", "insurance type mix" ], "note": "Share of adults with a routine check-up in the past year (PLACES, modelled). If diagnosed prevalence rises with check-up rates, part of the county variation is detection, not disease.", "fdr_q_within_level": 1.27e-29, "classification": "collapses_extended", "verdict": "Raw weak positive correlation r=0.23 (95% CI 0.19 to 0.27), Spearman rho=0.23, n=2392 counties, p=6.99e-30, BH q=1.3e-29. COLLAPSES under the extended controls: survives the primary control (partial r=0.210) but falls to 0.112 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. Share of adults with a routine check-up in the past year (PLACES, modelled). If diagnosed prevalence rises with check-up rates, part of the county variation is detection, not disease. Provenance: pct_checkup_annual is modelled (PLACES small-area estimate); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CD-2", "level": "county", "kind": "diagnosis_intensity", "x": { "name": "primary_care_physicians_per_100k", "unit": "rate per 100,000", "year": "~2022 (Area Health Resources Files vintage embedded in CHR 2024)", "source": "County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release via data/panel_county.csv", "tag": "measured/administrative (AHRF headcount)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2529, "n_unit": "counties", "x_filled": 2986, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.0854, "p": 1.69e-05, "ci95": [ -0.124, -0.0466 ] }, "spearman": { "rho": -0.1077, "p": 5.65e-08 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.0458, "p": 0.0216, "n": 2521 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "region_south" ], "r": 0.08, "p": 5.84e-05, "n": 2521 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "health-system contact", "coding practice", "insurance type mix" ], "note": "Physician supply is a detection-capacity proxy.", "fdr_q_within_level": 2.1e-05, "classification": "negligible", "verdict": "Raw negligible negative correlation r=-0.09 (95% CI -0.12 to -0.05), Spearman rho=-0.11, n=2529 counties, p=1.69e-05, BH q=2.1e-05. NEGLIGIBLE: raw r=-0.085 is statistically non-zero at n=2529 but below 0.10 in size; too small to interpret. Not a finding worth reporting as an association. Physician supply is a detection-capacity proxy. Provenance: primary_care_physicians_per_100k is measured/administrative (AHRF headcount); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CD-3", "level": "county", "kind": "diagnosis_intensity", "x": { "name": "norc_ma_share_of_bene_2020", "unit": "Medicare Advantage beneficiaries as percent of all beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, INSURANCE=2 / total", "tag": "measured/administrative (derived ratio)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2579, "n_unit": "counties", "x_filled": 3087, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.148, "p": 4.2e-14, "ci95": [ 0.11, 0.1855 ] }, "spearman": { "rho": 0.166, "p": 2.15e-17 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.138, "p": 2.09e-12, "n": 2571 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1279, "p": 1.17e-10, "n": 2520 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "health-system contact", "coding practice", "insurance type mix" ], "note": "Medicare Advantage plans are paid by risk score and code diagnoses more intensively than FFS; a positive association is a coding-intensity signal.", "fdr_q_within_level": 6.27e-14, "classification": "survives", "verdict": "Raw weak positive correlation r=0.15 (95% CI 0.11 to 0.19), Spearman rho=0.17, n=2579 counties, p=4.2e-14, BH q=6.3e-14. SURVIVES the confounder test: partial r=0.138 (p=2.09e-12) controlling for pct_85plus+median_household_income; extended partial r=0.1279. A place-level association not explained by the named control(s); candidate third things still unaddressed: health-system contact, coding practice, insurance type mix. Medicare Advantage plans are paid by risk score and code diagnoses more intensively than FFS; a positive association is a coding-intensity signal. Provenance: norc_ma_share_of_bene_2020 is measured/administrative (derived ratio); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CD-4", "level": "county", "kind": "diagnosis_intensity", "x": { "name": "norc_share_highly_likely_2020", "unit": "percent of the all-determinations prevalent count that is in the 'Highly likely' tier (diagnostic-certainty mix)", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, DEMENTIACAT=1 / all-determinations", "tag": "measured/administrative (derived ratio)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2579, "n_unit": "counties", "x_filled": 3077, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.3223, "p": 2.08e-63, "ci95": [ -0.3564, -0.2873 ] }, "spearman": { "rho": -0.2256, "p": 4.07e-31 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": -0.3319, "p": 3.83e-67, "n": 2571 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.4117, "p": 1.07e-103, "n": 2520 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "health-system contact", "coding practice", "insurance type mix" ], "note": "If counties with HIGH prevalence have a LOWER share of 'highly likely' cases, the excess is in the softer 'likely/possible' tiers, i.e. diagnostic-certainty mix differs by place.", "fdr_q_within_level": 6.45e-63, "classification": "survives", "verdict": "Raw moderate negative correlation r=-0.32 (95% CI -0.36 to -0.29), Spearman rho=-0.23, n=2579 counties, p=2.08e-63, BH q=6.5e-63. SURVIVES the confounder test: partial r=-0.332 (p=3.83e-67) controlling for pct_85plus+median_household_income; extended partial r=-0.4117. A place-level association not explained by the named control(s); candidate third things still unaddressed: health-system contact, coding practice, insurance type mix. If counties with HIGH prevalence have a LOWER share of 'highly likely' cases, the excess is in the softer 'likely/possible' tiers, i.e. diagnostic-certainty mix differs by place. Provenance: norc_share_highly_likely_2020 is measured/administrative (derived ratio); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CD-5", "level": "county", "kind": "diagnosis_intensity", "x": { "name": "norc_ffs_avg_total_cost_usd_2020", "unit": "average Medicare payment per FFS beneficiary in the county total row, USD", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, atc field (FFS only)", "tag": "measured/administrative" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2564, "n_unit": "counties", "x_filled": 2883, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.2651, "p": 1.67e-42, "ci95": [ 0.2288, 0.3008 ] }, "spearman": { "rho": 0.313, "p": 2.09e-59 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.2235, "p": 2.59e-30, "n": 2556 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1633, "p": 1.94e-16, "n": 2505 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "health-system contact", "coding practice", "insurance type mix" ], "note": "Average Medicare payment per FFS beneficiary: overall claims intensity of the county.", "fdr_q_within_level": 3.98e-42, "classification": "survives", "verdict": "Raw weak positive correlation r=0.27 (95% CI 0.23 to 0.30), Spearman rho=0.31, n=2564 counties, p=1.67e-42, BH q=4e-42. SURVIVES the confounder test: partial r=0.224 (p=2.59e-30) controlling for pct_85plus+median_household_income; extended partial r=0.1633. A place-level association not explained by the named control(s); candidate third things still unaddressed: health-system contact, coding practice, insurance type mix. Average Medicare payment per FFS beneficiary: overall claims intensity of the county. Provenance: norc_ffs_avg_total_cost_usd_2020 is measured/administrative; norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CD-6", "level": "county", "kind": "diagnosis_intensity", "x": { "name": "pct_uninsured", "unit": "percent (0-100)", "year": "~2021 (Small Area Health Insurance Estimates vintage embedded in CHR 2024)", "source": "County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release via data/panel_county.csv", "tag": "administrative compilation (CHR pass-through)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 3142, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.3002, "p": 6.95e-55, "ci95": [ 0.2647, 0.3349 ] }, "spearman": { "rho": 0.3322, "p": 1.7e-67 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1759, "p": 2.54e-19, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0551, "p": 0.00565, "n": 2521 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "health-system contact", "coding practice", "insurance type mix" ], "note": "Working-age uninsurance as a marker of low health-system contact before Medicare age.", "fdr_q_within_level": 1.92e-54, "classification": "collapses_extended", "verdict": "Raw moderate positive correlation r=0.30 (95% CI 0.26 to 0.33), Spearman rho=0.33, n=2580 counties, p=6.95e-55, BH q=1.9e-54. COLLAPSES under the extended controls: survives the primary control (partial r=0.176) but falls to 0.055 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. Working-age uninsurance as a marker of low health-system contact before Medicare age. Provenance: pct_uninsured is administrative compilation (CHR pass-through); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CH-1", "level": "county", "kind": "robustness", "x": { "name": "pct_black", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_highlylikely_prevalence_rate_2020", "unit": "diagnosed prevalence using only the strictest 'Highly likely' case tier, percent of all-age beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, DEMENTIACAT=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 3070, "n_unit": "counties", "x_filled": 3135, "y_filled": 3078, "frame_rows": 3143, "pearson": { "r": 0.3339, "p": 7.74e-81, "ci95": [ 0.3021, 0.365 ] }, "spearman": { "rho": 0.4533, "p": 1.89e-155 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.3074, "p": 3.49e-68, "n": 3070 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1712, "p": 1.15e-20, "n": 2923 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1289, "p": 1.39e-11, "n": 2729 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for the all-determinations version" ], "note": "Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes.", "fdr_q_within_level": 3.43e-80, "classification": "collapses_strict", "verdict": "Raw moderate positive correlation r=0.33 (95% CI 0.30 to 0.36), Spearman rho=0.45, n=3070 counties, p=7.74e-81, BH q=3.4e-80. COLLAPSES under place / diagnosis-intensity controls: survives age/income/region controls (extended partial r=0.1712) but falls to 0.129 once place and diagnosis-intensity markers are added (pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+norc_ma_share_of_bene_2020+pct_checkup_annual). What looked like an exposure effect is carried by where the county is and how intensively it codes diagnoses. Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes. Provenance: pct_black is survey estimate (ACS 5-year); norc_highlylikely_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CI-1", "level": "county", "kind": "robustness", "x": { "name": "pct_black", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_incidence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ first diagnosed in 2020", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 1928, "n_unit": "counties", "x_filled": 3135, "y_filled": 1936, "frame_rows": 3143, "pearson": { "r": 0.2995, "p": 3.06e-41, "ci95": [ 0.2583, 0.3396 ] }, "spearman": { "rho": 0.3565, "p": 7.4e-59 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.2071, "p": 4.04e-20, "n": 1928 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1064, "p": 2.99e-06, "n": 1920 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0951, "p": 5.9e-05, "n": 1778 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for prevalence", "prevalence itself (incidence is only detectable where someone is looking)" ], "note": "First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties).", "fdr_q_within_level": 7.03e-41, "classification": "collapses_extended", "verdict": "Raw weak positive correlation r=0.30 (95% CI 0.26 to 0.34), Spearman rho=0.36, n=1928 counties, p=3.06e-41, BH q=7e-41. COLLAPSES under the extended controls: survives the primary control (partial r=0.207) but falls to 0.106 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties). Provenance: pct_black is survey estimate (ACS 5-year); norc65_incidence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CH-2", "level": "county", "kind": "robustness", "x": { "name": "pct_bachelors_plus", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_highlylikely_prevalence_rate_2020", "unit": "diagnosed prevalence using only the strictest 'Highly likely' case tier, percent of all-age beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, DEMENTIACAT=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 3070, "n_unit": "counties", "x_filled": 3135, "y_filled": 3078, "frame_rows": 3143, "pearson": { "r": -0.1231, "p": 7.61e-12, "ci95": [ -0.1578, -0.0881 ] }, "spearman": { "rho": -0.121, "p": 1.75e-11 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": -0.0041, "p": 0.822, "n": 3070 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0366, "p": 0.0477, "n": 2923 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0949, "p": 6.73e-07, "n": 2729 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for the all-determinations version" ], "note": "Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes.", "fdr_q_within_level": 1.1e-11, "classification": "collapses", "verdict": "Raw weak negative correlation r=-0.12 (95% CI -0.16 to -0.09), Spearman rho=-0.12, n=3070 counties, p=7.61e-12, BH q=1.1e-11. COLLAPSES: partial r=-0.004 is under half the raw r=-0.123 once pct_85plus+median_household_income is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_bachelors_plus. Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes. Provenance: pct_bachelors_plus is survey estimate (ACS 5-year); norc_highlylikely_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CI-2", "level": "county", "kind": "robustness", "x": { "name": "pct_bachelors_plus", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_incidence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ first diagnosed in 2020", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 1928, "n_unit": "counties", "x_filled": 3135, "y_filled": 1936, "frame_rows": 3143, "pearson": { "r": -0.3003, "p": 1.78e-41, "ci95": [ -0.3404, -0.2591 ] }, "spearman": { "rho": -0.3226, "p": 6.13e-48 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": -0.0491, "p": 0.031, "n": 1928 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0664, "p": 0.00361, "n": 1920 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0881, "p": 0.000199, "n": 1778 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for prevalence", "prevalence itself (incidence is only detectable where someone is looking)" ], "note": "First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties).", "fdr_q_within_level": 4.16e-41, "classification": "collapses", "verdict": "Raw moderate negative correlation r=-0.30 (95% CI -0.34 to -0.26), Spearman rho=-0.32, n=1928 counties, p=1.78e-41, BH q=4.2e-41. COLLAPSES: partial r=-0.049 is under half the raw r=-0.300 once pct_85plus+median_household_income is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_bachelors_plus. First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties). Provenance: pct_bachelors_plus is survey estimate (ACS 5-year); norc65_incidence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CH-3", "level": "county", "kind": "robustness", "x": { "name": "median_household_income", "unit": "US dollars (nominal, ACS inflation-adjusted to survey-period dollars)", "year": "2020-2024 (ACS) or CHR vintage for 8 CT counties", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_highlylikely_prevalence_rate_2020", "unit": "diagnosed prevalence using only the strictest 'Highly likely' case tier, percent of all-age beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, DEMENTIACAT=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 3078, "n_unit": "counties", "x_filled": 3143, "y_filled": 3078, "frame_rows": 3143, "pearson": { "r": -0.1627, "p": 1.06e-19, "ci95": [ -0.1969, -0.1281 ] }, "spearman": { "rho": -0.1986, "p": 9.67e-29 }, "partial_primary": { "controls": [ "pct_85plus" ], "r": -0.167, "p": 1.2e-20, "n": 3070 }, "partial_extended": { "controls": [ "pct_85plus", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0324, "p": 0.0803, "n": 2923 }, "partial_strict": { "controls": [ "pct_85plus", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0453, "p": 0.0181, "n": 2729 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for the all-determinations version" ], "note": "Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes.", "fdr_q_within_level": 1.71e-19, "classification": "collapses_extended", "verdict": "Raw weak negative correlation r=-0.16 (95% CI -0.20 to -0.13), Spearman rho=-0.20, n=3078 counties, p=1.06e-19, BH q=1.7e-19. COLLAPSES under the extended controls: survives the primary control (partial r=-0.167) but falls to -0.032 with the extended set pct_85plus+primary_care_physicians_per_100k+region_south. Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes. Provenance: median_household_income is survey estimate (ACS 5-year); norc_highlylikely_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CI-3", "level": "county", "kind": "robustness", "x": { "name": "median_household_income", "unit": "US dollars (nominal, ACS inflation-adjusted to survey-period dollars)", "year": "2020-2024 (ACS) or CHR vintage for 8 CT counties", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_incidence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ first diagnosed in 2020", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 1936, "n_unit": "counties", "x_filled": 3143, "y_filled": 1936, "frame_rows": 3143, "pearson": { "r": -0.355, "p": 1.37e-58, "ci95": [ -0.3933, -0.3154 ] }, "spearman": { "rho": -0.3926, "p": 2.34e-72 }, "partial_primary": { "controls": [ "pct_85plus" ], "r": -0.3703, "p": 1.04e-63, "n": 1928 }, "partial_extended": { "controls": [ "pct_85plus", "primary_care_physicians_per_100k", "region_south" ], "r": -0.2665, "p": 1.4e-32, "n": 1920 }, "partial_strict": { "controls": [ "pct_85plus", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.2346, "p": 1.17e-23, "n": 1778 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for prevalence", "prevalence itself (incidence is only detectable where someone is looking)" ], "note": "First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties).", "fdr_q_within_level": 3.95e-58, "classification": "survives", "verdict": "Raw moderate negative correlation r=-0.35 (95% CI -0.39 to -0.32), Spearman rho=-0.39, n=1936 counties, p=1.37e-58, BH q=3.9e-58. SURVIVES the confounder test: partial r=-0.370 (p=1.04e-63) controlling for pct_85plus; extended partial r=-0.2665; strict partial r=-0.2346 controlling for pct_85plus+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: as for prevalence, prevalence itself (incidence is only detectable where someone is looking). First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties). Provenance: median_household_income is survey estimate (ACS 5-year); norc65_incidence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CH-4", "level": "county", "kind": "robustness", "x": { "name": "pct_85plus", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_highlylikely_prevalence_rate_2020", "unit": "diagnosed prevalence using only the strictest 'Highly likely' case tier, percent of all-age beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, DEMENTIACAT=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 3070, "n_unit": "counties", "x_filled": 3135, "y_filled": 3078, "frame_rows": 3143, "pearson": { "r": -0.0077, "p": 0.67, "ci95": [ -0.0431, 0.0277 ] }, "spearman": { "rho": -0.0654, "p": 0.000286 }, "partial_primary": { "controls": [ "median_household_income" ], "r": -0.0284, "p": 0.115, "n": 3070 }, "partial_extended": { "controls": [ "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1265, "p": 6.73e-12, "n": 2923 }, "partial_strict": { "controls": [ "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1502, "p": 3.04e-15, "n": 2729 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for the all-determinations version" ], "note": "Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes.", "fdr_q_within_level": 0.703, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.01 (95% CI -0.04 to 0.03), Spearman rho=-0.07, n=3070 counties, p=0.67, BH q=0.7. NULL: raw association not distinguishable from zero (BH q=0.7 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes. Provenance: pct_85plus is survey estimate (ACS 5-year); norc_highlylikely_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CI-4", "level": "county", "kind": "robustness", "x": { "name": "pct_85plus", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_incidence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ first diagnosed in 2020", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 1928, "n_unit": "counties", "x_filled": 3135, "y_filled": 1936, "frame_rows": 3143, "pearson": { "r": -0.0874, "p": 0.000122, "ci95": [ -0.1315, -0.0429 ] }, "spearman": { "rho": -0.0986, "p": 1.44e-05 }, "partial_primary": { "controls": [ "median_household_income" ], "r": -0.1435, "p": 2.49e-10, "n": 1928 }, "partial_extended": { "controls": [ "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0479, "p": 0.0359, "n": 1920 }, "partial_strict": { "controls": [ "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0239, "p": 0.314, "n": 1778 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for prevalence", "prevalence itself (incidence is only detectable where someone is looking)" ], "note": "First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties).", "fdr_q_within_level": 0.000145, "classification": "negligible", "verdict": "Raw negligible negative correlation r=-0.09 (95% CI -0.13 to -0.04), Spearman rho=-0.10, n=1928 counties, p=0.000122, BH q=0.00015. NEGLIGIBLE: raw r=-0.087 is statistically non-zero at n=1928 but below 0.10 in size; too small to interpret. Not a finding worth reporting as an association. First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties). Provenance: pct_85plus is survey estimate (ACS 5-year); norc65_incidence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CH-5", "level": "county", "kind": "robustness", "x": { "name": "pct_diabetes", "unit": "percent (0-100)", "year": "2023 (BRFSS)", "source": "CDC PLACES: County Data (GIS-Friendly Format), 2025 release via data/panel_county.csv", "tag": "modelled (PLACES small-area estimate)" }, "y": { "name": "norc_highlylikely_prevalence_rate_2020", "unit": "diagnosed prevalence using only the strictest 'Highly likely' case tier, percent of all-age beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, DEMENTIACAT=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2883, "n_unit": "counties", "x_filled": 2947, "y_filled": 3078, "frame_rows": 3143, "pearson": { "r": 0.2678, "p": 1.52e-48, "ci95": [ 0.2336, 0.3014 ] }, "spearman": { "rho": 0.2579, "p": 4.9e-45 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.2077, "p": 1.86e-29, "n": 2883 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0018, "p": 0.923, "n": 2746 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0576, "p": 0.00261, "n": 2729 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for the all-determinations version" ], "note": "Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes.", "fdr_q_within_level": 3.77e-48, "classification": "collapses_extended", "verdict": "Raw weak positive correlation r=0.27 (95% CI 0.23 to 0.30), Spearman rho=0.26, n=2883 counties, p=1.52e-48, BH q=3.8e-48. COLLAPSES under the extended controls: survives the primary control (partial r=0.208) but falls to 0.002 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes. Provenance: pct_diabetes is modelled (PLACES small-area estimate); norc_highlylikely_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CI-5", "level": "county", "kind": "robustness", "x": { "name": "pct_diabetes", "unit": "percent (0-100)", "year": "2023 (BRFSS)", "source": "CDC PLACES: County Data (GIS-Friendly Format), 2025 release via data/panel_county.csv", "tag": "modelled (PLACES small-area estimate)" }, "y": { "name": "norc65_incidence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ first diagnosed in 2020", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 1787, "n_unit": "counties", "x_filled": 2947, "y_filled": 1936, "frame_rows": 3143, "pearson": { "r": 0.4034, "p": 6.81e-71, "ci95": [ 0.3638, 0.4415 ] }, "spearman": { "rho": 0.4176, "p": 2.39e-76 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.2458, "p": 5.44e-26, "n": 1787 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.101, "p": 1.98e-05, "n": 1779 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0493, "p": 0.0375, "n": 1778 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for prevalence", "prevalence itself (incidence is only detectable where someone is looking)" ], "note": "First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties).", "fdr_q_within_level": 2.56e-70, "classification": "collapses_extended", "verdict": "Raw moderate positive correlation r=0.40 (95% CI 0.36 to 0.44), Spearman rho=0.42, n=1787 counties, p=6.81e-71, BH q=2.6e-70. COLLAPSES under the extended controls: survives the primary control (partial r=0.246) but falls to 0.101 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties). Provenance: pct_diabetes is modelled (PLACES small-area estimate); norc65_incidence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CH-6", "level": "county", "kind": "robustness", "x": { "name": "pm25_avg_ugm3", "unit": "micrograms per cubic meter (ug/m3)", "year": "~2021 (CDC/EPA Environmental Public Health Tracking vintage embedded in CHR 2024)", "source": "County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release via data/panel_county.csv", "tag": "modelled (CDC/EPA modelled PM2.5 via CHR)" }, "y": { "name": "norc_highlylikely_prevalence_rate_2020", "unit": "diagnosed prevalence using only the strictest 'Highly likely' case tier, percent of all-age beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, DEMENTIACAT=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 3066, "n_unit": "counties", "x_filled": 3115, "y_filled": 3078, "frame_rows": 3143, "pearson": { "r": 0.4019, "p": 2.15e-119, "ci95": [ 0.3718, 0.4312 ] }, "spearman": { "rho": 0.4004, "p": 1.92e-118 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.4038, "p": 2.67e-120, "n": 3058 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.3142, "p": 1.04e-67, "n": 2911 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.2946, "p": 1.15e-55, "n": 2723 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for the all-determinations version" ], "note": "Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes.", "fdr_q_within_level": 1.57e-118, "classification": "survives", "verdict": "Raw moderate positive correlation r=0.40 (95% CI 0.37 to 0.43), Spearman rho=0.40, n=3066 counties, p=2.15e-119, BH q=1.6e-118. SURVIVES the confounder test: partial r=0.404 (p=2.67e-120) controlling for pct_85plus+median_household_income; extended partial r=0.3142; strict partial r=0.2946 controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: as for the all-determinations version. Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes. Provenance: pm25_avg_ugm3 is modelled (CDC/EPA modelled PM2.5 via CHR); norc_highlylikely_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CI-6", "level": "county", "kind": "robustness", "x": { "name": "pm25_avg_ugm3", "unit": "micrograms per cubic meter (ug/m3)", "year": "~2021 (CDC/EPA Environmental Public Health Tracking vintage embedded in CHR 2024)", "source": "County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release via data/panel_county.csv", "tag": "modelled (CDC/EPA modelled PM2.5 via CHR)" }, "y": { "name": "norc65_incidence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ first diagnosed in 2020", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 1936, "n_unit": "counties", "x_filled": 3115, "y_filled": 1936, "frame_rows": 3143, "pearson": { "r": 0.2817, "p": 1.21e-36, "ci95": [ 0.2402, 0.3222 ] }, "spearman": { "rho": 0.2889, "p": 1.54e-38 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.2656, "p": 1.73e-32, "n": 1928 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.2133, "p": 3.39e-21, "n": 1920 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1858, "p": 2.87e-15, "n": 1778 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for prevalence", "prevalence itself (incidence is only detectable where someone is looking)" ], "note": "First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties).", "fdr_q_within_level": 2.68e-36, "classification": "survives", "verdict": "Raw weak positive correlation r=0.28 (95% CI 0.24 to 0.32), Spearman rho=0.29, n=1936 counties, p=1.21e-36, BH q=2.7e-36. SURVIVES the confounder test: partial r=0.266 (p=1.73e-32) controlling for pct_85plus+median_household_income; extended partial r=0.2133; strict partial r=0.1858 controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: as for prevalence, prevalence itself (incidence is only detectable where someone is looking). First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties). Provenance: pm25_avg_ugm3 is modelled (CDC/EPA modelled PM2.5 via CHR); norc65_incidence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CH-7", "level": "county", "kind": "robustness", "x": { "name": "pct_below_poverty", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_highlylikely_prevalence_rate_2020", "unit": "diagnosed prevalence using only the strictest 'Highly likely' case tier, percent of all-age beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, DEMENTIACAT=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 3070, "n_unit": "counties", "x_filled": 3135, "y_filled": 3078, "frame_rows": 3143, "pearson": { "r": 0.2103, "p": 5.02e-32, "ci95": [ 0.1762, 0.2439 ] }, "spearman": { "rho": 0.2298, "p": 4.47e-38 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1312, "p": 2.91e-13, "n": 3070 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0594, "p": 0.00132, "n": 2923 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0005, "p": 0.98, "n": 2729 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for the all-determinations version" ], "note": "Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes.", "fdr_q_within_level": 1.02e-31, "classification": "collapses_extended", "verdict": "Raw weak positive correlation r=0.21 (95% CI 0.18 to 0.24), Spearman rho=0.23, n=3070 counties, p=5.02e-32, BH q=1e-31. COLLAPSES under the extended controls: survives the primary control (partial r=0.131) but falls to 0.059 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes. Provenance: pct_below_poverty is survey estimate (ACS 5-year); norc_highlylikely_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CI-7", "level": "county", "kind": "robustness", "x": { "name": "pct_below_poverty", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_incidence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ first diagnosed in 2020", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 1928, "n_unit": "counties", "x_filled": 3135, "y_filled": 1936, "frame_rows": 3143, "pearson": { "r": 0.3886, "p": 1.59e-70, "ci95": [ 0.35, 0.4259 ] }, "spearman": { "rho": 0.3942, "p": 1.04e-72 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.156, "p": 5.62e-12, "n": 1928 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1138, "p": 5.79e-07, "n": 1920 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0806, "p": 0.000668, "n": 1778 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for prevalence", "prevalence itself (incidence is only detectable where someone is looking)" ], "note": "First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties).", "fdr_q_within_level": 5.8e-70, "classification": "collapses", "verdict": "Raw moderate positive correlation r=0.39 (95% CI 0.35 to 0.43), Spearman rho=0.39, n=1928 counties, p=1.59e-70, BH q=5.8e-70. COLLAPSES: partial r=0.156 is under half the raw r=0.389 once pct_85plus+median_household_income is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_below_poverty. First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties). Provenance: pct_below_poverty is survey estimate (ACS 5-year); norc65_incidence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CH-8", "level": "county", "kind": "robustness", "x": { "name": "pct_hispanic", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_highlylikely_prevalence_rate_2020", "unit": "diagnosed prevalence using only the strictest 'Highly likely' case tier, percent of all-age beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, DEMENTIACAT=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 3070, "n_unit": "counties", "x_filled": 3135, "y_filled": 3078, "frame_rows": 3143, "pearson": { "r": 0.1722, "p": 7.23e-22, "ci95": [ 0.1377, 0.2063 ] }, "spearman": { "rho": 0.1159, "p": 1.17e-10 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.188, "p": 8.12e-26, "n": 3070 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1862, "p": 3.28e-24, "n": 2923 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.2041, "p": 4.82e-27, "n": 2729 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for the all-determinations version" ], "note": "Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes.", "fdr_q_within_level": 1.23e-21, "classification": "survives", "verdict": "Raw weak positive correlation r=0.17 (95% CI 0.14 to 0.21), Spearman rho=0.12, n=3070 counties, p=7.23e-22, BH q=1.2e-21. SURVIVES the confounder test: partial r=0.188 (p=8.12e-26) controlling for pct_85plus+median_household_income; extended partial r=0.1862; strict partial r=0.2041 controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: as for the all-determinations version. Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes. Provenance: pct_hispanic is survey estimate (ACS 5-year); norc_highlylikely_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CI-8", "level": "county", "kind": "robustness", "x": { "name": "pct_hispanic", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_incidence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ first diagnosed in 2020", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 1928, "n_unit": "counties", "x_filled": 3135, "y_filled": 1936, "frame_rows": 3143, "pearson": { "r": 0.0481, "p": 0.0347, "ci95": [ 0.0035, 0.0925 ] }, "spearman": { "rho": -0.0587, "p": 0.00993 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1008, "p": 9.18e-06, "n": 1928 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1034, "p": 5.64e-06, "n": 1920 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1363, "p": 7.99e-09, "n": 1778 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for prevalence", "prevalence itself (incidence is only detectable where someone is looking)" ], "note": "First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties).", "fdr_q_within_level": 0.0388, "classification": "negligible", "verdict": "Raw negligible positive correlation r=0.05 (95% CI 0.00 to 0.09), Spearman rho=-0.06, n=1928 counties, p=0.0347, BH q=0.039. NEGLIGIBLE: raw r=0.048 is statistically non-zero at n=1928 but below 0.10 in size; too small to interpret. Not a finding worth reporting as an association. First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties). Provenance: pct_hispanic is survey estimate (ACS 5-year); norc65_incidence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CH-9", "level": "county", "kind": "robustness", "x": { "name": "exp_svi_overall_percentile", "unit": "percentile (0-1)", "year": "2022 (single cross-section; placed only on each county's 2022 row)", "source": "CDC/ATSDR Social Vulnerability Index, 2022 database, County via data/panel_exposure_county.csv", "tag": "derived index (from ACS survey estimates)" }, "y": { "name": "norc_highlylikely_prevalence_rate_2020", "unit": "diagnosed prevalence using only the strictest 'Highly likely' case tier, percent of all-age beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, DEMENTIACAT=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 3070, "n_unit": "counties", "x_filled": 3135, "y_filled": 3078, "frame_rows": 3143, "pearson": { "r": 0.3005, "p": 4.47e-65, "ci95": [ 0.2679, 0.3323 ] }, "spearman": { "rho": 0.3168, "p": 1.63e-72 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.2635, "p": 6.03e-50, "n": 3070 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1639, "p": 4.77e-19, "n": 2923 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0772, "p": 5.42e-05, "n": 2729 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for the all-determinations version" ], "note": "Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes.", "fdr_q_within_level": 1.46e-64, "classification": "collapses_strict", "verdict": "Raw moderate positive correlation r=0.30 (95% CI 0.27 to 0.33), Spearman rho=0.32, n=3070 counties, p=4.47e-65, BH q=1.5e-64. COLLAPSES under place / diagnosis-intensity controls: survives age/income/region controls (extended partial r=0.1639) but falls to 0.077 once place and diagnosis-intensity markers are added (pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual). What looked like an exposure effect is carried by where the county is and how intensively it codes diagnoses. Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes. Provenance: exp_svi_overall_percentile is derived index (from ACS survey estimates); norc_highlylikely_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CI-9", "level": "county", "kind": "robustness", "x": { "name": "exp_svi_overall_percentile", "unit": "percentile (0-1)", "year": "2022 (single cross-section; placed only on each county's 2022 row)", "source": "CDC/ATSDR Social Vulnerability Index, 2022 database, County via data/panel_exposure_county.csv", "tag": "derived index (from ACS survey estimates)" }, "y": { "name": "norc65_incidence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ first diagnosed in 2020", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 1928, "n_unit": "counties", "x_filled": 3135, "y_filled": 1936, "frame_rows": 3143, "pearson": { "r": 0.3823, "p": 4.1e-68, "ci95": [ 0.3435, 0.4197 ] }, "spearman": { "rho": 0.3978, "p": 3.95e-74 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.2088, "p": 1.96e-20, "n": 1928 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1439, "p": 2.39e-10, "n": 1920 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1263, "p": 9.19e-08, "n": 1778 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for prevalence", "prevalence itself (incidence is only detectable where someone is looking)" ], "note": "First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties).", "fdr_q_within_level": 1.41e-67, "classification": "collapses_extended", "verdict": "Raw moderate positive correlation r=0.38 (95% CI 0.34 to 0.42), Spearman rho=0.40, n=1928 counties, p=4.1e-68, BH q=1.4e-67. COLLAPSES under the extended controls: survives the primary control (partial r=0.209) but falls to 0.144 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties). Provenance: exp_svi_overall_percentile is derived index (from ACS survey estimates); norc65_incidence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CH-10", "level": "county", "kind": "robustness", "x": { "name": "pct_checkup_annual", "unit": "percent (0-100)", "year": "2023 (BRFSS)", "source": "CDC PLACES: County Data (GIS-Friendly Format), 2025 release via data/panel_county.csv", "tag": "modelled (PLACES small-area estimate)" }, "y": { "name": "norc_highlylikely_prevalence_rate_2020", "unit": "diagnosed prevalence using only the strictest 'Highly likely' case tier, percent of all-age beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, DEMENTIACAT=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2883, "n_unit": "counties", "x_filled": 2947, "y_filled": 3078, "frame_rows": 3143, "pearson": { "r": 0.1918, "p": 2.68e-25, "ci95": [ 0.1564, 0.2268 ] }, "spearman": { "rho": 0.1781, "p": 5.5e-22 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1693, "p": 5.49e-20, "n": 2883 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0314, "p": 0.1, "n": 2746 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020" ], "r": -0.0537, "p": 0.00499, "n": 2729 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for the all-determinations version" ], "note": "Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes.", "fdr_q_within_level": 4.68e-25, "classification": "collapses_extended", "verdict": "Raw weak positive correlation r=0.19 (95% CI 0.16 to 0.23), Spearman rho=0.18, n=2883 counties, p=2.68e-25, BH q=4.7e-25. COLLAPSES under the extended controls: survives the primary control (partial r=0.169) but falls to 0.031 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes. Provenance: pct_checkup_annual is modelled (PLACES small-area estimate); norc_highlylikely_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CI-10", "level": "county", "kind": "robustness", "x": { "name": "pct_checkup_annual", "unit": "percent (0-100)", "year": "2023 (BRFSS)", "source": "CDC PLACES: County Data (GIS-Friendly Format), 2025 release via data/panel_county.csv", "tag": "modelled (PLACES small-area estimate)" }, "y": { "name": "norc65_incidence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ first diagnosed in 2020", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 1787, "n_unit": "counties", "x_filled": 2947, "y_filled": 1936, "frame_rows": 3143, "pearson": { "r": 0.1853, "p": 2.88e-15, "ci95": [ 0.1401, 0.2297 ] }, "spearman": { "rho": 0.1858, "p": 2.41e-15 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1458, "p": 6e-10, "n": 1787 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.0448, "p": 0.0591, "n": 1779 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020" ], "r": -0.006, "p": 0.801, "n": 1778 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for prevalence", "prevalence itself (incidence is only detectable where someone is looking)" ], "note": "First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties).", "fdr_q_within_level": 4.41e-15, "classification": "collapses_extended", "verdict": "Raw weak positive correlation r=0.19 (95% CI 0.14 to 0.23), Spearman rho=0.19, n=1787 counties, p=2.88e-15, BH q=4.4e-15. COLLAPSES under the extended controls: survives the primary control (partial r=0.146) but falls to 0.045 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties). Provenance: pct_checkup_annual is modelled (PLACES small-area estimate); norc65_incidence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CH-11", "level": "county", "kind": "robustness", "x": { "name": "norc_ma_share_of_bene_2020", "unit": "Medicare Advantage beneficiaries as percent of all beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, INSURANCE=2 / total", "tag": "measured/administrative (derived ratio)" }, "y": { "name": "norc_highlylikely_prevalence_rate_2020", "unit": "diagnosed prevalence using only the strictest 'Highly likely' case tier, percent of all-age beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, DEMENTIACAT=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 3057, "n_unit": "counties", "x_filled": 3087, "y_filled": 3078, "frame_rows": 3143, "pearson": { "r": 0.1615, "p": 2.61e-19, "ci95": [ 0.1268, 0.1958 ] }, "spearman": { "rho": 0.1439, "p": 1.31e-15 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1576, "p": 2.08e-18, "n": 3049 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1102, "p": 2.58e-09, "n": 2906 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "pct_checkup_annual" ], "r": 0.0992, "p": 2.1e-07, "n": 2729 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for the all-determinations version" ], "note": "Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes.", "fdr_q_within_level": 4.15e-19, "classification": "negligible", "verdict": "Raw weak positive correlation r=0.16 (95% CI 0.13 to 0.20), Spearman rho=0.14, n=3057 counties, p=2.61e-19, BH q=4.2e-19. NEGLIGIBLE: survives every control but the fully-controlled partial r=0.099 is below 0.10: statistically non-zero at this n, practically nothing. Not a finding worth reporting as an association. Strict 'Highly likely' case tier: if the association holds here it is not driven by soft 'possible' codes. Provenance: norc_ma_share_of_bene_2020 is measured/administrative (derived ratio); norc_highlylikely_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CI-11", "level": "county", "kind": "robustness", "x": { "name": "norc_ma_share_of_bene_2020", "unit": "Medicare Advantage beneficiaries as percent of all beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, INSURANCE=2 / total", "tag": "measured/administrative (derived ratio)" }, "y": { "name": "norc65_incidence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ first diagnosed in 2020", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 1935, "n_unit": "counties", "x_filled": 3087, "y_filled": 1936, "frame_rows": 3143, "pearson": { "r": 0.0945, "p": 3.11e-05, "ci95": [ 0.0502, 0.1385 ] }, "spearman": { "rho": 0.098, "p": 1.58e-05 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.101, "p": 8.91e-06, "n": 1927 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1108, "p": 1.13e-06, "n": 1919 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "pct_checkup_annual" ], "r": 0.0882, "p": 0.000195, "n": 1778 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for prevalence", "prevalence itself (incidence is only detectable where someone is looking)" ], "note": "First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties).", "fdr_q_within_level": 3.78e-05, "classification": "negligible", "verdict": "Raw negligible positive correlation r=0.09 (95% CI 0.05 to 0.14), Spearman rho=0.10, n=1935 counties, p=3.11e-05, BH q=3.8e-05. NEGLIGIBLE: raw r=0.095 is statistically non-zero at n=1935 but below 0.10 in size; too small to interpret. Not a finding worth reporting as an association. First-time diagnoses in 2020 among 65+ (n is smaller: incidence is suppressed in more counties). Provenance: norc_ma_share_of_bene_2020 is measured/administrative (derived ratio); norc65_incidence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CP-1", "level": "county", "kind": "robustness", "x": { "name": "pct_black", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_pooled_prevalence_rate_2020_2022", "unit": "3-year pooled diagnosed prevalence, percent (numerator and denominator ROUNDED by NORC)", "year": "2020-2022", "source": "NORC Dementia DataHub pooled 2020-2022 PUF, county AGECAT=6 total row", "tag": "measured/administrative (rounded counts)" }, "n": 3117, "n_unit": "counties", "x_filled": 3135, "y_filled": 3125, "frame_rows": 3143, "pearson": { "r": 0.3695, "p": 1.87e-101, "ci95": [ 0.3388, 0.3995 ] }, "spearman": { "rho": 0.4957, "p": 5.12e-193 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.2968, "p": 2e-64, "n": 3117 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1511, "p": 1.36e-16, "n": 2964 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1124, "p": 3.34e-09, "n": 2754 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for the 2020 version" ], "note": "Pooled 2020-2022 rate from NORC-rounded counts; smooths single-year noise.", "fdr_q_within_level": 1.01e-100, "classification": "collapses_extended", "verdict": "Raw moderate positive correlation r=0.37 (95% CI 0.34 to 0.40), Spearman rho=0.50, n=3117 counties, p=1.87e-101, BH q=1e-100. COLLAPSES under the extended controls: survives the primary control (partial r=0.297) but falls to 0.151 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. Pooled 2020-2022 rate from NORC-rounded counts; smooths single-year noise. Provenance: pct_black is survey estimate (ACS 5-year); norc_pooled_prevalence_rate_2020_2022 is measured/administrative (rounded counts). Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CP-2", "level": "county", "kind": "robustness", "x": { "name": "pct_bachelors_plus", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_pooled_prevalence_rate_2020_2022", "unit": "3-year pooled diagnosed prevalence, percent (numerator and denominator ROUNDED by NORC)", "year": "2020-2022", "source": "NORC Dementia DataHub pooled 2020-2022 PUF, county AGECAT=6 total row", "tag": "measured/administrative (rounded counts)" }, "n": 3117, "n_unit": "counties", "x_filled": 3135, "y_filled": 3125, "frame_rows": 3143, "pearson": { "r": -0.2163, "p": 2.62e-34, "ci95": [ -0.2495, -0.1825 ] }, "spearman": { "rho": -0.2545, "p": 2.89e-47 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": -0.0226, "p": 0.207, "n": 3117 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0684, "p": 0.000195, "n": 2964 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.1159, "p": 1.07e-09, "n": 2754 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for the 2020 version" ], "note": "Pooled 2020-2022 rate from NORC-rounded counts; smooths single-year noise.", "fdr_q_within_level": 5.6e-34, "classification": "collapses", "verdict": "Raw weak negative correlation r=-0.22 (95% CI -0.25 to -0.18), Spearman rho=-0.25, n=3117 counties, p=2.62e-34, BH q=5.6e-34. COLLAPSES: partial r=-0.023 is under half the raw r=-0.216 once pct_85plus+median_household_income is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_bachelors_plus. Pooled 2020-2022 rate from NORC-rounded counts; smooths single-year noise. Provenance: pct_bachelors_plus is survey estimate (ACS 5-year); norc_pooled_prevalence_rate_2020_2022 is measured/administrative (rounded counts). Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CP-3", "level": "county", "kind": "robustness", "x": { "name": "pct_85plus", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_pooled_prevalence_rate_2020_2022", "unit": "3-year pooled diagnosed prevalence, percent (numerator and denominator ROUNDED by NORC)", "year": "2020-2022", "source": "NORC Dementia DataHub pooled 2020-2022 PUF, county AGECAT=6 total row", "tag": "measured/administrative (rounded counts)" }, "n": 3117, "n_unit": "counties", "x_filled": 3135, "y_filled": 3125, "frame_rows": 3143, "pearson": { "r": -0.1239, "p": 3.91e-12, "ci95": [ -0.1583, -0.0892 ] }, "spearman": { "rho": -0.1504, "p": 3.13e-17 }, "partial_primary": { "controls": [ "median_household_income" ], "r": -0.1652, "p": 1.65e-20, "n": 3117 }, "partial_extended": { "controls": [ "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0311, "p": 0.0907, "n": 2964 }, "partial_strict": { "controls": [ "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0058, "p": 0.761, "n": 2754 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for the 2020 version" ], "note": "Pooled 2020-2022 rate from NORC-rounded counts; smooths single-year noise.", "fdr_q_within_level": 5.7e-12, "classification": "collapses_extended", "verdict": "Raw weak negative correlation r=-0.12 (95% CI -0.16 to -0.09), Spearman rho=-0.15, n=3117 counties, p=3.91e-12, BH q=5.7e-12. COLLAPSES under the extended controls: survives the primary control (partial r=-0.165) but falls to -0.031 with the extended set median_household_income+primary_care_physicians_per_100k+region_south. Pooled 2020-2022 rate from NORC-rounded counts; smooths single-year noise. Provenance: pct_85plus is survey estimate (ACS 5-year); norc_pooled_prevalence_rate_2020_2022 is measured/administrative (rounded counts). Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CP-4", "level": "county", "kind": "robustness", "x": { "name": "median_household_income", "unit": "US dollars (nominal, ACS inflation-adjusted to survey-period dollars)", "year": "2020-2024 (ACS) or CHR vintage for 8 CT counties", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_pooled_prevalence_rate_2020_2022", "unit": "3-year pooled diagnosed prevalence, percent (numerator and denominator ROUNDED by NORC)", "year": "2020-2022", "source": "NORC Dementia DataHub pooled 2020-2022 PUF, county AGECAT=6 total row", "tag": "measured/administrative (rounded counts)" }, "n": 3125, "n_unit": "counties", "x_filled": 3143, "y_filled": 3125, "frame_rows": 3143, "pearson": { "r": -0.2727, "p": 1.98e-54, "ci95": [ -0.3049, -0.24 ] }, "spearman": { "rho": -0.3181, "p": 1.94e-74 }, "partial_primary": { "controls": [ "pct_85plus" ], "r": -0.293, "p": 9.03e-63, "n": 3117 }, "partial_extended": { "controls": [ "pct_85plus", "primary_care_physicians_per_100k", "region_south" ], "r": -0.1958, "p": 5.45e-27, "n": 2964 }, "partial_strict": { "controls": [ "pct_85plus", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.1821, "p": 5.82e-22, "n": 2754 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "as for the 2020 version" ], "note": "Pooled 2020-2022 rate from NORC-rounded counts; smooths single-year noise.", "fdr_q_within_level": 5.34e-54, "classification": "survives", "verdict": "Raw weak negative correlation r=-0.27 (95% CI -0.30 to -0.24), Spearman rho=-0.32, n=3125 counties, p=1.98e-54, BH q=5.3e-54. SURVIVES the confounder test: partial r=-0.293 (p=9.03e-63) controlling for pct_85plus; extended partial r=-0.1958; strict partial r=-0.1821 controlling for pct_85plus+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: as for the 2020 version. Pooled 2020-2022 rate from NORC-rounded counts; smooths single-year noise. Provenance: median_household_income is survey estimate (ACS 5-year); norc_pooled_prevalence_rate_2020_2022 is measured/administrative (rounded counts). Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CF-1", "level": "county", "kind": "exposure", "x": { "name": "pct_85plus", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_dementia_mortality_rate_2020", "unit": "percent (0-100)", "year": "2020", "source": "Derived from NORC Dementia DataHub counts per the DataHub's own published rate formula (mortality and COVID measures are divided by prev_cnt, not bene_cnt, per the codebook) via data/panel_county.csv (joined from data/county_dementia_outcomes.csv)", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2997, "n_unit": "counties", "x_filled": 3135, "y_filled": 3005, "frame_rows": 3143, "pearson": { "r": 0.1704, "p": 5.93e-21, "ci95": [ 0.1354, 0.2049 ] }, "spearman": { "rho": 0.1659, "p": 6.24e-20 }, "partial_primary": { "controls": [ "median_household_income" ], "r": 0.1585, "p": 2.52e-18, "n": 2997 }, "partial_extended": { "controls": [ "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1165, "p": 3.97e-10, "n": 2865 }, "partial_strict": { "controls": [ "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0627, "p": 0.00117, "n": 2679 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "older diagnosed cases have higher mortality" }, "confounders_named": [ "age of the diagnosed (older cases die sooner)" ], "note": "Outcome is all-cause deaths in 2020 as a share of DIAGNOSED prevalent cases (case fatality among the diagnosed), in the first COVID year.", "fdr_q_within_level": 9.94e-21, "classification": "collapses_strict", "verdict": "Raw weak positive correlation r=0.17 (95% CI 0.14 to 0.20), Spearman rho=0.17, n=2997 counties, p=5.93e-21, BH q=9.9e-21. COLLAPSES under place / diagnosis-intensity controls: survives age/income/region controls (extended partial r=0.1165) but falls to 0.063 once place and diagnosis-intensity markers are added (median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual). What looked like an exposure effect is carried by where the county is and how intensively it codes diagnoses. Outcome is all-cause deaths in 2020 as a share of DIAGNOSED prevalent cases (case fatality among the diagnosed), in the first COVID year. Provenance: pct_85plus is survey estimate (ACS 5-year); norc_dementia_mortality_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CF-2", "level": "county", "kind": "exposure", "x": { "name": "median_household_income", "unit": "US dollars (nominal, ACS inflation-adjusted to survey-period dollars)", "year": "2020-2024 (ACS) or CHR vintage for 8 CT counties", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_dementia_mortality_rate_2020", "unit": "percent (0-100)", "year": "2020", "source": "Derived from NORC Dementia DataHub counts per the DataHub's own published rate formula (mortality and COVID measures are divided by prev_cnt, not bene_cnt, per the codebook) via data/panel_county.csv (joined from data/county_dementia_outcomes.csv)", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 3005, "n_unit": "counties", "x_filled": 3143, "y_filled": 3005, "frame_rows": 3143, "pearson": { "r": -0.1113, "p": 9.53e-10, "ci95": [ -0.1465, -0.0758 ] }, "spearman": { "rho": -0.0936, "p": 2.72e-07 }, "partial_primary": { "controls": [ "pct_85plus" ], "r": -0.0926, "p": 3.82e-07, "n": 2997 }, "partial_extended": { "controls": [ "pct_85plus", "primary_care_physicians_per_100k", "region_south" ], "r": -0.1162, "p": 4.42e-10, "n": 2865 }, "partial_strict": { "controls": [ "pct_85plus", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.1157, "p": 1.89e-09, "n": 2679 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "nursing-home supply", "COVID-2020 exposure", "diagnosis timing (late diagnosis = sicker cases)" ], "note": "Outcome is all-cause deaths in 2020 as a share of DIAGNOSED prevalent cases (case fatality among the diagnosed), in the first COVID year.", "fdr_q_within_level": 1.28e-09, "classification": "survives", "verdict": "Raw weak negative correlation r=-0.11 (95% CI -0.15 to -0.08), Spearman rho=-0.09, n=3005 counties, p=9.53e-10, BH q=1.3e-09. SURVIVES the confounder test: partial r=-0.093 (p=3.82e-07) controlling for pct_85plus; extended partial r=-0.1162; strict partial r=-0.1157 controlling for pct_85plus+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: nursing-home supply, COVID-2020 exposure, diagnosis timing (late diagnosis = sicker cases). Outcome is all-cause deaths in 2020 as a share of DIAGNOSED prevalent cases (case fatality among the diagnosed), in the first COVID year. Provenance: median_household_income is survey estimate (ACS 5-year); norc_dementia_mortality_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CF-3", "level": "county", "kind": "exposure", "x": { "name": "pct_black", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_dementia_mortality_rate_2020", "unit": "percent (0-100)", "year": "2020", "source": "Derived from NORC Dementia DataHub counts per the DataHub's own published rate formula (mortality and COVID measures are divided by prev_cnt, not bene_cnt, per the codebook) via data/panel_county.csv (joined from data/county_dementia_outcomes.csv)", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2997, "n_unit": "counties", "x_filled": 3135, "y_filled": 3005, "frame_rows": 3143, "pearson": { "r": -0.0852, "p": 3.03e-06, "ci95": [ -0.1206, -0.0495 ] }, "spearman": { "rho": -0.1099, "p": 1.64e-09 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": -0.0857, "p": 2.61e-06, "n": 2997 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0224, "p": 0.231, "n": 2865 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.0195, "p": 0.312, "n": 2679 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region", "COVID-2020 mortality gradient", "diagnosis timing" ], "note": "Outcome is all-cause deaths in 2020 as a share of DIAGNOSED prevalent cases (case fatality among the diagnosed), in the first COVID year.", "fdr_q_within_level": 3.87e-06, "classification": "negligible", "verdict": "Raw negligible negative correlation r=-0.09 (95% CI -0.12 to -0.05), Spearman rho=-0.11, n=2997 counties, p=3.03e-06, BH q=3.9e-06. NEGLIGIBLE: raw r=-0.085 is statistically non-zero at n=2997 but below 0.10 in size; too small to interpret. Not a finding worth reporting as an association. Outcome is all-cause deaths in 2020 as a share of DIAGNOSED prevalent cases (case fatality among the diagnosed), in the first COVID year. Provenance: pct_black is survey estimate (ACS 5-year); norc_dementia_mortality_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CF-4", "level": "county", "kind": "exposure", "x": { "name": "pct_bachelors_plus", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_dementia_mortality_rate_2020", "unit": "percent (0-100)", "year": "2020", "source": "Derived from NORC Dementia DataHub counts per the DataHub's own published rate formula (mortality and COVID measures are divided by prev_cnt, not bene_cnt, per the codebook) via data/panel_county.csv (joined from data/county_dementia_outcomes.csv)", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2997, "n_unit": "counties", "x_filled": 3135, "y_filled": 3005, "frame_rows": 3143, "pearson": { "r": -0.1765, "p": 2.1e-22, "ci95": [ -0.211, -0.1416 ] }, "spearman": { "rho": -0.1593, "p": 1.71e-18 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": -0.148, "p": 3.92e-16, "n": 2997 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.1352, "p": 3.73e-13, "n": 2865 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.1388, "p": 5.39e-13, "n": 2679 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "income", "diagnosis timing" ], "note": "Outcome is all-cause deaths in 2020 as a share of DIAGNOSED prevalent cases (case fatality among the diagnosed), in the first COVID year.", "fdr_q_within_level": 3.62e-22, "classification": "survives", "verdict": "Raw weak negative correlation r=-0.18 (95% CI -0.21 to -0.14), Spearman rho=-0.16, n=2997 counties, p=2.1e-22, BH q=3.6e-22. SURVIVES the confounder test: partial r=-0.148 (p=3.92e-16) controlling for pct_85plus+median_household_income; extended partial r=-0.1352; strict partial r=-0.1388 controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: income, diagnosis timing. Outcome is all-cause deaths in 2020 as a share of DIAGNOSED prevalent cases (case fatality among the diagnosed), in the first COVID year. Provenance: pct_bachelors_plus is survey estimate (ACS 5-year); norc_dementia_mortality_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CF-5", "level": "county", "kind": "exposure", "x": { "name": "pm25_avg_ugm3", "unit": "micrograms per cubic meter (ug/m3)", "year": "~2021 (CDC/EPA Environmental Public Health Tracking vintage embedded in CHR 2024)", "source": "County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release via data/panel_county.csv", "tag": "modelled (CDC/EPA modelled PM2.5 via CHR)" }, "y": { "name": "norc_dementia_mortality_rate_2020", "unit": "percent (0-100)", "year": "2020", "source": "Derived from NORC Dementia DataHub counts per the DataHub's own published rate formula (mortality and COVID measures are divided by prev_cnt, not bene_cnt, per the codebook) via data/panel_county.csv (joined from data/county_dementia_outcomes.csv)", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2997, "n_unit": "counties", "x_filled": 3115, "y_filled": 3005, "frame_rows": 3143, "pearson": { "r": -0.1223, "p": 1.87e-11, "ci95": [ -0.1574, -0.0868 ] }, "spearman": { "rho": -0.0921, "p": 4.38e-07 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": -0.0923, "p": 4.36e-07, "n": 2989 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0354, "p": 0.0583, "n": 2857 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0118, "p": 0.542, "n": 2675 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "urbanicity", "COVID-2020 urban mortality" ], "note": "Outcome is all-cause deaths in 2020 as a share of DIAGNOSED prevalent cases (case fatality among the diagnosed), in the first COVID year.", "fdr_q_within_level": 2.67e-11, "classification": "collapses_extended", "verdict": "Raw weak negative correlation r=-0.12 (95% CI -0.16 to -0.09), Spearman rho=-0.09, n=2997 counties, p=1.87e-11, BH q=2.7e-11. COLLAPSES under the extended controls: survives the primary control (partial r=-0.092) but falls to -0.035 with the extended set pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south. Outcome is all-cause deaths in 2020 as a share of DIAGNOSED prevalent cases (case fatality among the diagnosed), in the first COVID year. Provenance: pm25_avg_ugm3 is modelled (CDC/EPA modelled PM2.5 via CHR); norc_dementia_mortality_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CF-6", "level": "county", "kind": "exposure", "x": { "name": "hospital_beds_per_1000", "unit": "beds per 1,000 population", "year": "2026 snapshot (facility roster) / population as dated above", "source": "CMS Provider of Services File - Quality Improvement and Evaluation System (QIES), current quarterly snapshot pulled 2026-09-21 via data/panel_county.csv", "tag": "measured/administrative" }, "y": { "name": "norc_dementia_mortality_rate_2020", "unit": "percent (0-100)", "year": "2020", "source": "Derived from NORC Dementia DataHub counts per the DataHub's own published rate formula (mortality and COVID measures are divided by prev_cnt, not bene_cnt, per the codebook) via data/panel_county.csv (joined from data/county_dementia_outcomes.csv)", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2419, "n_unit": "counties", "x_filled": 2457, "y_filled": 3005, "frame_rows": 3143, "pearson": { "r": 0.1084, "p": 9e-08, "ci95": [ 0.0689, 0.1476 ] }, "spearman": { "rho": 0.1572, "p": 7.4e-15 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.0793, "p": 9.64e-05, "n": 2411 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1064, "p": 2e-07, "n": 2377 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.0951, "p": 6.62e-06, "n": 2239 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "rurality", "case mix" ], "note": "Outcome is all-cause deaths in 2020 as a share of DIAGNOSED prevalent cases (case fatality among the diagnosed), in the first COVID year.", "fdr_q_within_level": 1.2e-07, "classification": "negligible", "verdict": "Raw weak positive correlation r=0.11 (95% CI 0.07 to 0.15), Spearman rho=0.16, n=2419 counties, p=9e-08, BH q=1.2e-07. NEGLIGIBLE: survives every control but the fully-controlled partial r=0.095 is below 0.10: statistically non-zero at this n, practically nothing. Not a finding worth reporting as an association. Outcome is all-cause deaths in 2020 as a share of DIAGNOSED prevalent cases (case fatality among the diagnosed), in the first COVID year. Provenance: hospital_beds_per_1000 is measured/administrative; norc_dementia_mortality_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CF-7", "level": "county", "kind": "exposure", "x": { "name": "pct_smoking", "unit": "percent (0-100)", "year": "2023 (BRFSS)", "source": "CDC PLACES: County Data (GIS-Friendly Format), 2025 release via data/panel_county.csv", "tag": "modelled (PLACES small-area estimate)" }, "y": { "name": "norc_dementia_mortality_rate_2020", "unit": "percent (0-100)", "year": "2020", "source": "Derived from NORC Dementia DataHub counts per the DataHub's own published rate formula (mortality and COVID measures are divided by prev_cnt, not bene_cnt, per the codebook) via data/panel_county.csv (joined from data/county_dementia_outcomes.csv)", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2810, "n_unit": "counties", "x_filled": 2947, "y_filled": 3005, "frame_rows": 3143, "pearson": { "r": 0.1565, "p": 7.16e-17, "ci95": [ 0.1202, 0.1924 ] }, "spearman": { "rho": 0.1717, "p": 4.87e-20 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1315, "p": 2.61e-12, "n": 2810 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1304, "p": 1.16e-11, "n": 2688 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.1335, "p": 3.96e-12, "n": 2679 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "competing risks", "poverty" ], "note": "Outcome is all-cause deaths in 2020 as a share of DIAGNOSED prevalent cases (case fatality among the diagnosed), in the first COVID year.", "fdr_q_within_level": 1.11e-16, "classification": "survives", "verdict": "Raw weak positive correlation r=0.16 (95% CI 0.12 to 0.19), Spearman rho=0.17, n=2810 counties, p=7.16e-17, BH q=1.1e-16. SURVIVES the confounder test: partial r=0.132 (p=2.61e-12) controlling for pct_85plus+median_household_income; extended partial r=0.1304; strict partial r=0.1335 controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: competing risks, poverty. Outcome is all-cause deaths in 2020 as a share of DIAGNOSED prevalent cases (case fatality among the diagnosed), in the first COVID year. Provenance: pct_smoking is modelled (PLACES small-area estimate); norc_dementia_mortality_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CC-1", "level": "county", "kind": "composition", "x": { "name": "pct_black", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_white_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among White beneficiaries, all ages, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, RACECAT=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 3085, "n_unit": "counties", "x_filled": 3135, "y_filled": 3093, "frame_rows": 3143, "pearson": { "r": 0.3202, "p": 1.8e-74, "ci95": [ 0.2881, 0.3515 ] }, "spearman": { "rho": 0.4202, "p": 2.98e-132 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.2793, "p": 2.09e-56, "n": 3085 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1644, "p": 3.11e-19, "n": 2936 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region (South)", "poverty", "coding practice" ], "note": "If this is near zero, the pct_black ~ prevalence association is composition (Black beneficiaries have higher diagnosed prevalence). If it persists, something about the PLACE raises diagnosed prevalence for White beneficiaries too.", "fdr_q_within_level": 7.2e-74, "classification": "survives", "verdict": "Raw moderate positive correlation r=0.32 (95% CI 0.29 to 0.35), Spearman rho=0.42, n=3085 counties, p=1.8e-74, BH q=7.2e-74. SURVIVES the confounder test: partial r=0.279 (p=2.09e-56) controlling for pct_85plus+median_household_income; extended partial r=0.1644. A place-level association not explained by the named control(s); candidate third things still unaddressed: region (South), poverty, coding practice. If this is near zero, the pct_black ~ prevalence association is composition (Black beneficiaries have higher diagnosed prevalence). If it persists, something about the PLACE raises diagnosed prevalence for White beneficiaries too. Provenance: pct_black is survey estimate (ACS 5-year); norc_white_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CC-2", "level": "county", "kind": "composition", "x": { "name": "pct_black", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_black_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among African-American beneficiaries, all ages, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, RACECAT=2 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 1539, "n_unit": "counties", "x_filled": 3135, "y_filled": 1547, "frame_rows": 3143, "pearson": { "r": 0.0407, "p": 0.111, "ci95": [ -0.0093, 0.0905 ] }, "spearman": { "rho": 0.1026, "p": 5.55e-05 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.0016, "p": 0.951, "n": 1539 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.0276, "p": 0.283, "n": 1508 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "suppression: only 1,547 counties have an unsuppressed Black stratum (selection towards counties with many Black beneficiaries)", "region" ], "note": "Within-stratum test: prevalence among Black beneficiaries against the county's Black share.", "fdr_q_within_level": 0.122, "classification": "null", "verdict": "Raw negligible positive correlation r=0.04 (95% CI -0.01 to 0.09), Spearman rho=0.10, n=1539 counties, p=0.111, BH q=0.12. NULL: raw association not distinguishable from zero (BH q=0.12 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Within-stratum test: prevalence among Black beneficiaries against the county's Black share. Provenance: pct_black is survey estimate (ACS 5-year); norc_black_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CC-3", "level": "county", "kind": "composition", "x": { "name": "pct_hispanic", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_hispanic_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among Hispanic beneficiaries, all ages, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, RACECAT=5 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 1296, "n_unit": "counties", "x_filled": 3135, "y_filled": 1304, "frame_rows": 3143, "pearson": { "r": 0.1288, "p": 3.28e-06, "ci95": [ 0.0749, 0.182 ] }, "spearman": { "rho": -0.0322, "p": 0.246 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1232, "p": 8.67e-06, "n": 1296 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1126, "p": 5.51e-05, "n": 1276 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "suppression (1,304 counties)", "region", "language access" ], "note": "", "fdr_q_within_level": 4.11e-06, "classification": "survives", "verdict": "Raw weak positive correlation r=0.13 (95% CI 0.07 to 0.18), Spearman rho=-0.03, n=1296 counties, p=3.28e-06, BH q=4.1e-06. SURVIVES the confounder test: partial r=0.123 (p=8.67e-06) controlling for pct_85plus+median_household_income; extended partial r=0.1126. A place-level association not explained by the named control(s); candidate third things still unaddressed: suppression (1,304 counties), region, language access. Provenance: pct_hispanic is survey estimate (ACS 5-year); norc_hispanic_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CC-4", "level": "county", "kind": "consistency", "x": { "name": "norc_white_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among White beneficiaries, all ages, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, RACECAT=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "norc_black_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among African-American beneficiaries, all ages, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, RACECAT=2 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 1545, "n_unit": "counties", "x_filled": 3093, "y_filled": 1547, "frame_rows": 3143, "pearson": { "r": 0.495, "p": 2.67e-96, "ci95": [ 0.4564, 0.5318 ] }, "spearman": { "rho": 0.5312, "p": 3.12e-113 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "shared county diagnosis intensity" ], "note": "Do the two race-specific rates move together across counties? High agreement points to a county-level (diagnosis or environment) factor shared by both groups.", "fdr_q_within_level": 1.32e-95, "classification": "consistency_check", "verdict": "Raw moderate positive correlation r=0.49 (95% CI 0.46 to 0.53), Spearman rho=0.53, n=1545 counties, p=2.67e-96, BH q=1.3e-95. CONSISTENCY CHECK: r=0.495: agreement level between two versions of the outcome / measurement. Do the two race-specific rates move together across counties? High agreement points to a county-level (diagnosis or environment) factor shared by both groups. Provenance: norc_white_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases); norc_black_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CC-5", "level": "county", "kind": "exposure", "x": { "name": "region_south", "unit": "1 = state in US Census South region (Divisions 5-7), else 0", "year": "static", "source": "US Census Bureau region definition", "tag": "classification (not a measurement)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 3143, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.4257, "p": 4.4e-114, "ci95": [ 0.3936, 0.4568 ] }, "spearman": { "rho": 0.461, "p": 6.32e-136 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.3288, "p": 6.46e-66, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "pct_black" ], "r": 0.2599, "p": 3.48e-40, "n": 2521 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "stroke-belt vascular risk", "coding practice", "race composition" ], "note": "Census South region as a named third thing behind several county gradients.", "fdr_q_within_level": 3.03e-113, "classification": "survives", "verdict": "Raw moderate positive correlation r=0.43 (95% CI 0.39 to 0.46), Spearman rho=0.46, n=2580 counties, p=4.4e-114, BH q=3e-113. SURVIVES the confounder test: partial r=0.329 (p=6.46e-66) controlling for pct_85plus+median_household_income; extended partial r=0.2599. A place-level association not explained by the named control(s); candidate third things still unaddressed: stroke-belt vascular risk, coding practice, race composition. Census South region as a named third thing behind several county gradients. Provenance: region_south is classification (not a measurement); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CA-1", "level": "county", "kind": "artefact_test", "x": { "name": "norc_share_bene_65plus_2020", "unit": "beneficiaries aged 65+ as percent of all beneficiaries (rest are under-65 disability/ESRD entitlement)", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 / total", "tag": "measured/administrative (derived ratio)" }, "y": { "name": "norc_dementia_prevalence_rate_2020", "unit": "percent (0-100)", "year": "2020", "source": "Derived from NORC Dementia DataHub counts per the DataHub's own published rate formula (dementiadatahub.org/resources/outcome-measures.html) via data/panel_county.csv (joined from data/county_dementia_outcomes.csv)", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 3061, "n_unit": "counties", "x_filled": 3065, "y_filled": 3104, "frame_rows": 3143, "pearson": { "r": -0.2902, "p": 1.77e-60, "ci95": [ -0.3223, -0.2574 ] }, "spearman": { "rho": -0.3154, "p": 1.14e-71 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "disability entitlement (under-65 beneficiaries are a poverty/disability marker)" ], "note": "All-age prevalence rate mechanically depends on how many under-65 (disability) beneficiaries dilute the denominator. This is why the 65+-only rate is the primary outcome in this report.", "fdr_q_within_level": 5.35e-60, "classification": "artefact_documented", "verdict": "Raw weak negative correlation r=-0.29 (95% CI -0.32 to -0.26), Spearman rho=-0.32, n=3061 counties, p=1.77e-60, BH q=5.3e-60. STRUCTURAL DEPENDENCE DOCUMENTED: r=-0.290, q=5.3e-60: documents the mechanical/structural dependence named. All-age prevalence rate mechanically depends on how many under-65 (disability) beneficiaries dilute the denominator. This is why the 65+-only rate is the primary outcome in this report. Provenance: norc_share_bene_65plus_2020 is measured/administrative (derived ratio); norc_dementia_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CA-2", "level": "county", "kind": "artefact_test", "x": { "name": "norc_prev_cnt_2020", "unit": "count (persons)", "year": "2020", "source": "NORC Dementia DataHub (NORC at the University of Chicago), 2020 Public Use File via data/panel_county.csv (joined from data/county_dementia_outcomes.csv)", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "exp_primary_care_physicians_count_2023", "unit": "count (persons)", "year": "2023 (single cross-section; placed only on each county's 2023 row)", "source": "HRSA Area Health Resources Files (AHRF) 2024-2025, field phys_nf_prim_care_pc_exc_rsdt_23 (AMA Physician Master File) via data/panel_exposure_county.csv", "tag": "measured/administrative (AHRF headcount)" }, "n": 3104, "n_unit": "counties", "x_filled": 3104, "y_filled": 3143, "frame_rows": 3143, "pearson": { "r": 0.9446, "p": 0.0, "ci95": [ 0.9407, 0.9483 ] }, "spearman": { "rho": 0.9039, "p": 0.0 }, "partial_primary": { "controls": [ "population" ], "r": 0.1363, "p": 2.44e-14, "n": 3104 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "population size" ], "note": "Two raw counts. Both scale with population; the partial on population is the real test (replicates the baseline S11 lesson at county level).", "fdr_q_within_level": 0.0, "classification": "artefact_collapses", "verdict": "Raw strong positive correlation r=0.94 (95% CI 0.94 to 0.95), Spearman rho=0.90, n=3104 counties, p=0.0, BH q=0. POPULATION/STRUCTURE ARTEFACT: raw r=0.945 falls to partial r=0.136 controlling for population. The raw coefficient is not a relationship. Two raw counts. Both scale with population; the partial on population is the real test (replicates the baseline S11 lesson at county level). Provenance: norc_prev_cnt_2020 is measured/administrative (claims-based DIAGNOSED cases); exp_primary_care_physicians_count_2023 is measured/administrative (AHRF headcount). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CA-3", "level": "county", "kind": "artefact_test", "x": { "name": "norc_prev_cnt_2020", "unit": "count (persons)", "year": "2020", "source": "NORC Dementia DataHub (NORC at the University of Chicago), 2020 Public Use File via data/panel_county.csv (joined from data/county_dementia_outcomes.csv)", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "exp_hospital_beds_count_2023", "unit": "count", "year": "2023 (single cross-section; placed only on each county's 2023 row)", "source": "HRSA Area Health Resources Files (AHRF) 2024-2025, field hosp_beds_23 (AHA Survey Database) via data/panel_exposure_county.csv", "tag": "measured/administrative (AHRF headcount)" }, "n": 3104, "n_unit": "counties", "x_filled": 3104, "y_filled": 3143, "frame_rows": 3143, "pearson": { "r": 0.9257, "p": 0.0, "ci95": [ 0.9205, 0.9306 ] }, "spearman": { "rho": 0.7488, "p": 0.0 }, "partial_primary": { "controls": [ "population" ], "r": 0.2468, "p": 2.73e-44, "n": 3104 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "population size" ], "note": "Two raw counts, same test.", "fdr_q_within_level": 0.0, "classification": "artefact_collapses", "verdict": "Raw strong positive correlation r=0.93 (95% CI 0.92 to 0.93), Spearman rho=0.75, n=3104 counties, p=0.0, BH q=0. POPULATION/STRUCTURE ARTEFACT: raw r=0.926 falls to partial r=0.247 controlling for population. The raw coefficient is not a relationship. Two raw counts, same test. Provenance: norc_prev_cnt_2020 is measured/administrative (claims-based DIAGNOSED cases); exp_hospital_beds_count_2023 is measured/administrative (AHRF headcount). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CA-4", "level": "county", "kind": "artefact_test", "x": { "name": "norc65_prev_cnt_2020", "unit": "count of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis (raw count, scales with population)", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "pct_black", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "n": 2572, "n_unit": "counties", "x_filled": 2580, "y_filled": 3135, "frame_rows": 3143, "pearson": { "r": 0.0568, "p": 0.00393, "ci95": [ 0.0182, 0.0953 ] }, "spearman": { "rho": 0.2667, "p": 4.02e-43 }, "partial_primary": { "controls": [ "population" ], "r": 0.0356, "p": 0.0711, "n": 2572 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "population size" ], "note": "A count outcome against a share exposure: the raw r says little; the population-controlled partial is what the C3 baseline card should have used.", "fdr_q_within_level": 0.00451, "classification": "artefact_test_survives", "verdict": "Raw negligible positive correlation r=0.06 (95% CI 0.02 to 0.10), Spearman rho=0.27, n=2572 counties, p=0.00393, BH q=0.0045. ARTEFACT TEST, RESIDUAL REMAINS: partial r=0.036 controlling for population retains >=50% of raw. A count outcome against a share exposure: the raw r says little; the population-controlled partial is what the C3 baseline card should have used. Provenance: norc65_prev_cnt_2020 is measured/administrative (claims-based DIAGNOSED cases); pct_black is survey estimate (ACS 5-year). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CA-5", "level": "county", "kind": "artefact_test", "x": { "name": "pct_living_alone_65plus", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prev_cnt_2020", "unit": "count of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis (raw count, scales with population)", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.0815, "p": 3.49e-05, "ci95": [ -0.1198, -0.043 ] }, "spearman": { "rho": -0.1218, "p": 5.66e-10 }, "partial_primary": { "controls": [ "population" ], "r": 0.0957, "p": 1.15e-06, "n": 2572 }, "partial_extended": { "controls": [ "population", "pct_85plus" ], "r": 0.0607, "p": 0.00209, "n": 2572 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "population size" ], "note": "Re-run of baseline C3 (which used the raw prevalence COUNT as outcome) with population controlled.", "fdr_q_within_level": 4.2e-05, "classification": "artefact_test_survives", "verdict": "Raw negligible negative correlation r=-0.08 (95% CI -0.12 to -0.04), Spearman rho=-0.12, n=2572 counties, p=3.49e-05, BH q=4.2e-05. ARTEFACT TEST, RESIDUAL REMAINS: partial r=0.096 controlling for population retains >=50% of raw. Re-run of baseline C3 (which used the raw prevalence COUNT as outcome) with population controlled. Provenance: pct_living_alone_65plus is survey estimate (ACS 5-year); norc65_prev_cnt_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CA-6", "level": "county", "kind": "exposure", "x": { "name": "pct_living_alone_65plus", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3135, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.1812, "p": 2.04e-20, "ci95": [ 0.1435, 0.2183 ] }, "spearman": { "rho": 0.1842, "p": 4.52e-21 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.0782, "p": 7.2e-05, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1251, "p": 2.95e-10, "n": 2521 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "social isolation is a risk factor" }, "confounders_named": [ "age structure", "institutionalisation" ], "note": "Baseline C3 re-run with a RATE outcome instead of a count.", "fdr_q_within_level": 3.33e-20, "classification": "collapses", "verdict": "Raw weak positive correlation r=0.18 (95% CI 0.14 to 0.22), Spearman rho=0.18, n=2572 counties, p=2.04e-20, BH q=3.3e-20. COLLAPSES: partial r=0.078 is under half the raw r=0.181 once pct_85plus+median_household_income is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_living_alone_65plus. Baseline C3 re-run with a RATE outcome instead of a count. Provenance: pct_living_alone_65plus is survey estimate (ACS 5-year); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CM-1", "level": "county", "kind": "measured_vs_modelled", "x": { "name": "exp_pm25_monitor_mean_2016_2019", "unit": "mean of the four annual monitor-based PM2.5 means 2016-2019, ug/m3 (all four years required)", "year": "2016-2019", "source": "US EPA AQS annual_conc_by_monitor via data/panel_exposure_county.csv", "tag": "measured (physical monitor)" }, "y": { "name": "exp_pm25_modelled_mean_2016_2019", "unit": "mean of the four annual modelled population-weighted PM2.5 means 2016-2019, ug/m3", "year": "2016-2019", "source": "CDC Environmental Public Health Tracking downscaler via data/panel_exposure_county.csv", "tag": "modelled (statistical downscaler)" }, "n": 577, "n_unit": "counties", "x_filled": 585, "y_filled": 3107, "frame_rows": 3143, "pearson": { "r": 0.8733, "p": 9.46e-182, "ci95": [ 0.8525, 0.8914 ] }, "spearman": { "rho": 0.8529, "p": 2.06e-164 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "Agreement between measured (monitor) and modelled (downscaler) PM2.5 in the counties that have both. r below 0.95 means they are not twins and the modelled column carries model error.", "fdr_q_within_level": 7.82e-181, "classification": "consistency_check", "verdict": "Raw strong positive correlation r=0.87 (95% CI 0.85 to 0.89), Spearman rho=0.85, n=577 counties, p=9.46e-182, BH q=7.8e-181. CONSISTENCY CHECK: r=0.873: agreement level between two versions of the outcome / measurement. Agreement between measured (monitor) and modelled (downscaler) PM2.5 in the counties that have both. r below 0.95 means they are not twins and the modelled column carries model error. Provenance: exp_pm25_monitor_mean_2016_2019 is measured (physical monitor); exp_pm25_modelled_mean_2016_2019 is modelled (statistical downscaler). One variable is modelled and one measured: they are mixed in this coefficient. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CM-2", "level": "county", "kind": "measured_vs_modelled", "x": { "name": "pm25_avg_ugm3", "unit": "micrograms per cubic meter (ug/m3)", "year": "~2021 (CDC/EPA Environmental Public Health Tracking vintage embedded in CHR 2024)", "source": "County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release via data/panel_county.csv", "tag": "modelled (CDC/EPA modelled PM2.5 via CHR)" }, "y": { "name": "exp_pm25_modelled_mean_2016_2019", "unit": "mean of the four annual modelled population-weighted PM2.5 means 2016-2019, ug/m3", "year": "2016-2019", "source": "CDC Environmental Public Health Tracking downscaler via data/panel_exposure_county.csv", "tag": "modelled (statistical downscaler)" }, "n": 3107, "n_unit": "counties", "x_filled": 3115, "y_filled": 3107, "frame_rows": 3143, "pearson": { "r": 0.928, "p": 0.0, "ci95": [ 0.9229, 0.9327 ] }, "spearman": { "rho": 0.9219, "p": 0.0 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "The CHR PM2.5 pass-through (single ~2021 modelled value) against the four-year modelled mean from the same CDC downscaler family.", "fdr_q_within_level": 0.0, "classification": "consistency_check", "verdict": "Raw strong positive correlation r=0.93 (95% CI 0.92 to 0.93), Spearman rho=0.92, n=3107 counties, p=0.0, BH q=0. CONSISTENCY CHECK: r=0.928: agreement level between two versions of the outcome / measurement. The CHR PM2.5 pass-through (single ~2021 modelled value) against the four-year modelled mean from the same CDC downscaler family. Provenance: pm25_avg_ugm3 is modelled (CDC/EPA modelled PM2.5 via CHR); exp_pm25_modelled_mean_2016_2019 is modelled (statistical downscaler). Both variables are modelled: this coefficient compares two models, not two measurements. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CM-3", "level": "county", "kind": "measured_vs_modelled", "x": { "name": "pct_65plus", "unit": "percent (0-100)", "year": "~2018-2022 (ACS vintage embedded in CHR 2024)", "source": "County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release via data/panel_county.csv", "tag": "survey estimate (ACS via CHR)" }, "y": { "name": "exp_pct_65plus_2020", "unit": "percent", "year": "2020-2024 (per row)", "source": "US Census Bureau, Population Division, Vintage 2024 Population Estimates (PEP) via data/panel_exposure_county.csv", "tag": "modelled/derived" }, "n": 3135, "n_unit": "counties", "x_filled": 3143, "y_filled": 3135, "frame_rows": 3143, "pearson": { "r": 0.9917, "p": 0.0, "ci95": [ 0.9911, 0.9923 ] }, "spearman": { "rho": 0.9915, "p": 0.0 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "ACS 5-year share 65+ (CHR) against the Census PEP 2020 share: twin check (|r|>=0.95 = same quantity twice).", "fdr_q_within_level": 0.0, "classification": "twin", "verdict": "Raw strong positive correlation r=0.99 (95% CI 0.99 to 0.99), Spearman rho=0.99, n=3135 counties, p=0.0, BH q=0. TWIN: |r|=0.992 >= 0.95: the two columns are the same quantity twice; never use both in one model. ACS 5-year share 65+ (CHR) against the Census PEP 2020 share: twin check (|r|>=0.95 = same quantity twice). Provenance: pct_65plus is survey estimate (ACS via CHR); exp_pct_65plus_2020 is modelled/derived. One variable is modelled and one measured: they are mixed in this coefficient. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CM-4", "level": "county", "kind": "measured_vs_modelled", "x": { "name": "primary_care_physicians_per_100k", "unit": "rate per 100,000", "year": "~2022 (Area Health Resources Files vintage embedded in CHR 2024)", "source": "County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release via data/panel_county.csv", "tag": "measured/administrative (AHRF headcount)" }, "y": { "name": "exp_primary_care_physicians_per_1000pop_2023", "unit": "physicians per 1,000 population", "year": "2023 (single cross-section; placed only on each county's 2023 row)", "source": "Derived: primary_care_physicians_count_2023 / AHRF popn_est_23 * 1000 via data/panel_exposure_county.csv", "tag": "modelled/derived" }, "n": 2978, "n_unit": "counties", "x_filled": 2986, "y_filled": 3135, "frame_rows": 3143, "pearson": { "r": 0.951, "p": 0.0, "ci95": [ 0.9475, 0.9543 ] }, "spearman": { "rho": 0.943, "p": 0.0 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "CHR PCP rate (~2022 AHRF) against the direct AHRF 2023 rate: twin check.", "fdr_q_within_level": 0.0, "classification": "twin", "verdict": "Raw strong positive correlation r=0.95 (95% CI 0.95 to 0.95), Spearman rho=0.94, n=2978 counties, p=0.0, BH q=0. TWIN: |r|=0.951 >= 0.95: the two columns are the same quantity twice; never use both in one model. CHR PCP rate (~2022 AHRF) against the direct AHRF 2023 rate: twin check. Provenance: primary_care_physicians_per_100k is measured/administrative (AHRF headcount); exp_primary_care_physicians_per_1000pop_2023 is modelled/derived. One variable is modelled and one measured: they are mixed in this coefficient. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CM-5", "level": "county", "kind": "consistency", "x": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "norc_dementia_prevalence_rate_2020", "unit": "percent (0-100)", "year": "2020", "source": "Derived from NORC Dementia DataHub counts per the DataHub's own published rate formula (dementiadatahub.org/resources/outcome-measures.html) via data/panel_county.csv (joined from data/county_dementia_outcomes.csv)", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2579, "n_unit": "counties", "x_filled": 2580, "y_filled": 3104, "frame_rows": 3143, "pearson": { "r": 0.9752, "p": 0.0, "ci95": [ 0.9733, 0.9771 ] }, "spearman": { "rho": 0.9695, "p": 0.0 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "65+ rate against the all-age rate the baseline used.", "fdr_q_within_level": 0.0, "classification": "twin", "verdict": "Raw strong positive correlation r=0.98 (95% CI 0.97 to 0.98), Spearman rho=0.97, n=2579 counties, p=0.0, BH q=0. TWIN: |r|=0.975 >= 0.95: the two columns are the same quantity twice; never use both in one model. 65+ rate against the all-age rate the baseline used. Provenance: norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases); norc_dementia_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CM-6", "level": "county", "kind": "consistency", "x": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "norc_pooled_prevalence_rate_2020_2022", "unit": "3-year pooled diagnosed prevalence, percent (numerator and denominator ROUNDED by NORC)", "year": "2020-2022", "source": "NORC Dementia DataHub pooled 2020-2022 PUF, county AGECAT=6 total row", "tag": "measured/administrative (rounded counts)" }, "n": 2580, "n_unit": "counties", "x_filled": 2580, "y_filled": 3125, "frame_rows": 3143, "pearson": { "r": 0.9712, "p": 0.0, "ci95": [ 0.9689, 0.9733 ] }, "spearman": { "rho": 0.9638, "p": 0.0 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "2020 65+ rate against the pooled 2020-2022 all-age rate (rounded counts).", "fdr_q_within_level": 0.0, "classification": "twin", "verdict": "Raw strong positive correlation r=0.97 (95% CI 0.97 to 0.97), Spearman rho=0.96, n=2580 counties, p=0.0, BH q=0. TWIN: |r|=0.971 >= 0.95: the two columns are the same quantity twice; never use both in one model. 2020 65+ rate against the pooled 2020-2022 all-age rate (rounded counts). Provenance: norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases); norc_pooled_prevalence_rate_2020_2022 is measured/administrative (rounded counts). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CM-7", "level": "county", "kind": "consistency", "x": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "norc_highlylikely_prevalence_rate_2020", "unit": "diagnosed prevalence using only the strictest 'Highly likely' case tier, percent of all-age beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, DEMENTIACAT=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 2580, "y_filled": 3078, "frame_rows": 3143, "pearson": { "r": 0.7258, "p": 0.0, "ci95": [ 0.707, 0.7435 ] }, "spearman": { "rho": 0.756, "p": 0.0 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "All-determinations 65+ rate against the strict-tier all-age rate.", "fdr_q_within_level": 0.0, "classification": "consistency_check", "verdict": "Raw strong positive correlation r=0.73 (95% CI 0.71 to 0.74), Spearman rho=0.76, n=2580 counties, p=0.0, BH q=0. CONSISTENCY CHECK: r=0.726: agreement level between two versions of the outcome / measurement. All-determinations 65+ rate against the strict-tier all-age rate. Provenance: norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases); norc_highlylikely_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CM-8", "level": "county", "kind": "consistency", "x": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "norc65_incidence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ first diagnosed in 2020", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 1932, "n_unit": "counties", "x_filled": 2580, "y_filled": 1936, "frame_rows": 3143, "pearson": { "r": 0.9039, "p": 0.0, "ci95": [ 0.8954, 0.9118 ] }, "spearman": { "rho": 0.8909, "p": 0.0 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "detection: a county that diagnoses more also detects more new cases" ], "note": "Prevalence vs incidence in the same year and file. Very high agreement is expected mechanically AND from shared detection intensity.", "fdr_q_within_level": 0.0, "classification": "consistency_check", "verdict": "Raw strong positive correlation r=0.90 (95% CI 0.90 to 0.91), Spearman rho=0.89, n=1932 counties, p=0.0, BH q=0. CONSISTENCY CHECK: r=0.904: agreement level between two versions of the outcome / measurement. Prevalence vs incidence in the same year and file. Very high agreement is expected mechanically AND from shared detection intensity. Provenance: norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases); norc65_incidence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CU-1", "level": "county", "kind": "uninterpretable", "x": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "norc_dementia_mortality_rate_2020", "unit": "percent (0-100)", "year": "2020", "source": "Derived from NORC Dementia DataHub counts per the DataHub's own published rate formula (mortality and COVID measures are divided by prev_cnt, not bene_cnt, per the codebook) via data/panel_county.csv (joined from data/county_dementia_outcomes.csv)", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2576, "n_unit": "counties", "x_filled": 2580, "y_filled": 3005, "frame_rows": 3143, "pearson": { "r": -0.2225, "p": 2.87e-30, "ci95": [ -0.2589, -0.1855 ] }, "spearman": { "rho": -0.1953, "p": 1.5e-23 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "the prevalence count is the DENOMINATOR of the case-fatality rate" ], "note": "Case fatality = deaths / prevalent cases. Correlating it with the prevalence rate conditions on the outcome's own denominator: a place that diagnoses many mild cases will mechanically show low case fatality. Not a finding.", "fdr_q_within_level": 5.39e-30, "classification": "uninterpretable", "verdict": "Raw weak negative correlation r=-0.22 (95% CI -0.26 to -0.19), Spearman rho=-0.20, n=2576 counties, p=2.87e-30, BH q=5.4e-30. UNINTERPRETABLE: flagged a priori: the prevalence count is the DENOMINATOR of the case-fatality rate. Documented so that nobody publishes it as a finding. Case fatality = deaths / prevalent cases. Correlating it with the prevalence rate conditions on the outcome's own denominator: a place that diagnoses many mild cases will mechanically show low case fatality. Not a finding. Provenance: norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases); norc_dementia_mortality_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CU-2", "level": "county", "kind": "uninterpretable", "x": { "name": "pct_65plus_with_disability", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "y": { "name": "norc_dementia_mortality_rate_2020", "unit": "percent (0-100)", "year": "2020", "source": "Derived from NORC Dementia DataHub counts per the DataHub's own published rate formula (mortality and COVID measures are divided by prev_cnt, not bene_cnt, per the codebook) via data/panel_county.csv (joined from data/county_dementia_outcomes.csv)", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2997, "n_unit": "counties", "x_filled": 3135, "y_filled": 3005, "frame_rows": 3143, "pearson": { "r": -0.0035, "p": 0.849, "ci95": [ -0.0393, 0.0323 ] }, "spearman": { "rho": 0.0036, "p": 0.844 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "dementia causes disability; disability is measured in the same population as the deaths" ], "note": "Reverse causation and shared population: uninterpretable as an exposure.", "fdr_q_within_level": 0.863, "classification": "uninterpretable", "verdict": "Raw negligible negative correlation r=-0.00 (95% CI -0.04 to 0.03), Spearman rho=0.00, n=2997 counties, p=0.849, BH q=0.86. UNINTERPRETABLE: flagged a priori: dementia causes disability; disability is measured in the same population as the deaths. Documented so that nobody publishes it as a finding. Reverse causation and shared population: uninterpretable as an exposure. Provenance: pct_65plus_with_disability is survey estimate (ACS 5-year); norc_dementia_mortality_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CM-9", "level": "county", "kind": "measured_vs_modelled", "x": { "name": "grocery_stores_per_1000pop_2020", "unit": "grocery stores per 1,000 population", "year": "2020", "source": "USDA ERS Food Environment Atlas via data/panel_county.csv (Phase 4 direct join; identical values to exp_grocery_stores_per_1000pop_2020)", "tag": "measured/administrative (USDA establishment counts)" }, "y": { "name": "exp_grocery_stores_per_1000pop_2020", "unit": "stores per 1,000 population", "year": "2020 (single cross-section; placed only on each county's 2020 row)", "source": "USDA ERS Food Environment Atlas 2025 release, variable GROCPTH20 via data/panel_exposure_county.csv", "tag": "measured/administrative (USDA establishment counts)" }, "n": 2234, "n_unit": "counties", "x_filled": 2234, "y_filled": 2234, "frame_rows": 3143, "pearson": { "r": 1.0, "p": null, "ci95": [ null, null ] }, "spearman": { "rho": 1.0, "p": null }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "Phase 4's direct panel_county.csv join of USDA grocery-store density against the pre-existing exposure-panel snapshot CX-46 already tested: same source file, joined twice.", "fdr_q_within_level": null, "classification": "twin", "verdict": "Raw r=1.00, n=2234. TWIN: |r|=1.000 >= 0.95: the two columns are the same quantity twice; never use both in one model. Phase 4's direct panel_county.csv join of USDA grocery-store density against the pre-existing exposure-panel snapshot CX-46 already tested: same source file, joined twice. Provenance: grocery_stores_per_1000pop_2020 is measured/administrative (USDA establishment counts); exp_grocery_stores_per_1000pop_2020 is measured/administrative (USDA establishment counts). Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CM-10", "level": "county", "kind": "measured_vs_modelled", "x": { "name": "fastfood_restaurants_per_1000pop_2020", "unit": "fast-food restaurants per 1,000 population", "year": "2020", "source": "USDA ERS Food Environment Atlas via data/panel_county.csv (Phase 4 direct join; identical values to exp_fastfood_restaurants_per_1000pop_2020)", "tag": "measured/administrative (USDA establishment counts)" }, "y": { "name": "exp_fastfood_restaurants_per_1000pop_2020", "unit": "restaurants per 1,000 population", "year": "2020 (single cross-section; placed only on each county's 2020 row)", "source": "USDA ERS Food Environment Atlas 2025 release, variable FFRPTH20 via data/panel_exposure_county.csv", "tag": "measured/administrative (USDA establishment counts)" }, "n": 2672, "n_unit": "counties", "x_filled": 2672, "y_filled": 2672, "frame_rows": 3143, "pearson": { "r": 1.0, "p": null, "ci95": [ null, null ] }, "spearman": { "rho": 1.0, "p": null }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "Same duplication check for fast-food restaurant density (CX-47).", "fdr_q_within_level": null, "classification": "twin", "verdict": "Raw r=1.00, n=2672. TWIN: |r|=1.000 >= 0.95: the two columns are the same quantity twice; never use both in one model. Same duplication check for fast-food restaurant density (CX-47). Provenance: fastfood_restaurants_per_1000pop_2020 is measured/administrative (USDA establishment counts); exp_fastfood_restaurants_per_1000pop_2020 is measured/administrative (USDA establishment counts). Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CM-11", "level": "county", "kind": "measured_vs_modelled", "x": { "name": "fullservice_restaurants_per_1000pop_2020", "unit": "full-service restaurants per 1,000 population", "year": "2020", "source": "USDA ERS Food Environment Atlas via data/panel_county.csv (Phase 4 direct join; identical values to exp_fullservice_restaurants_per_1000pop_2020)", "tag": "measured/administrative (USDA establishment counts)" }, "y": { "name": "exp_fullservice_restaurants_per_1000pop_2020", "unit": "restaurants per 1,000 population", "year": "2020 (single cross-section; placed only on each county's 2020 row)", "source": "USDA ERS Food Environment Atlas 2025 release, variable FSRPTH20 via data/panel_exposure_county.csv", "tag": "measured/administrative (USDA establishment counts)" }, "n": 2861, "n_unit": "counties", "x_filled": 2861, "y_filled": 2861, "frame_rows": 3143, "pearson": { "r": 1.0, "p": null, "ci95": [ null, null ] }, "spearman": { "rho": 1.0, "p": null }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "Same duplication check for full-service restaurant density (CX-48, the strongest food-environment survivor).", "fdr_q_within_level": null, "classification": "twin", "verdict": "Raw r=1.00, n=2861. TWIN: |r|=1.000 >= 0.95: the two columns are the same quantity twice; never use both in one model. Same duplication check for full-service restaurant density (CX-48, the strongest food-environment survivor). Provenance: fullservice_restaurants_per_1000pop_2020 is measured/administrative (USDA establishment counts); exp_fullservice_restaurants_per_1000pop_2020 is measured/administrative (USDA establishment counts). Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CM-12", "level": "county", "kind": "measured_vs_modelled", "x": { "name": "dollar_stores_per_1000pop_2020", "unit": "dollar stores per 1,000 population", "year": "2020", "source": "USDA ERS Food Environment Atlas via data/panel_county.csv (Phase 4 direct join; identical values to exp_dollar_stores_per_1000pop_2020)", "tag": "measured/administrative (USDA establishment counts)" }, "y": { "name": "exp_dollar_stores_per_1000pop_2020", "unit": "stores per 1,000 population", "year": "2020 (single cross-section; placed only on each county's 2020 row)", "source": "USDA ERS Food Environment Atlas 2025 release, variable DSPTH20 via data/panel_exposure_county.csv", "tag": "measured/administrative (USDA establishment counts)" }, "n": 3143, "n_unit": "counties", "x_filled": 3143, "y_filled": 3143, "frame_rows": 3143, "pearson": { "r": 1.0, "p": null, "ci95": [ null, null ] }, "spearman": { "rho": 1.0, "p": null }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "Same duplication check for dollar-store density (CX-45, which collapsed).", "fdr_q_within_level": null, "classification": "twin", "verdict": "Raw r=1.00, n=3143. TWIN: |r|=1.000 >= 0.95: the two columns are the same quantity twice; never use both in one model. Same duplication check for dollar-store density (CX-45, which collapsed). Provenance: dollar_stores_per_1000pop_2020 is measured/administrative (USDA establishment counts); exp_dollar_stores_per_1000pop_2020 is measured/administrative (USDA establishment counts). Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CX-54", "level": "county", "kind": "exposure", "x": { "name": "pct_low_access_pop_1mi", "unit": "percent of tract population living more than 1 mile (urban) / 10 miles (rural) from the nearest large grocery store, population-weighted county rollup of 2010 census tracts", "year": "2019", "source": "USDA ERS Food Access Research Atlas (2019 LRAM vintage) via data/panel_county.csv (data/sources/usda_food_access_research_atlas)", "tag": "measured/administrative (USDA tract-distance flag, population-weighted county aggregate)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 3141, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.0294, "p": 0.136, "ci95": [ -0.0092, 0.0679 ] }, "spearman": { "rho": 0.0345, "p": 0.0799 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": -0.0932, "p": 2.22e-06, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.1559, "p": 3.45e-15, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.1528, "p": 9.81e-14, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "rurality (distance thresholds are rural-calibrated)", "region", "urbanicity" ], "note": "New in Phase 4: USDA Food Access Research Atlas, a tract-level population-weighted rollup, distinct from the Food Environment Atlas store-count columns tested as CX-42/CX-43 (exp_pct_*_low_access_2019, both null).", "fdr_q_within_level": 0.147, "classification": "null", "verdict": "Raw negligible positive correlation r=0.03 (95% CI -0.01 to 0.07), Spearman rho=0.03, n=2580 counties, p=0.136, BH q=0.15. NULL: raw association not distinguishable from zero (BH q=0.15 within level). Absence of a place-level correlation is not evidence of no individual-level effect. New in Phase 4: USDA Food Access Research Atlas, a tract-level population-weighted rollup, distinct from the Food Environment Atlas store-count columns tested as CX-42/CX-43 (exp_pct_*_low_access_2019, both null). Provenance: pct_low_access_pop_1mi is measured/administrative (USDA tract-distance flag, population-weighted county aggregate); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CX-55", "level": "county", "kind": "exposure", "x": { "name": "pct_seniors_low_access_1mi", "unit": "percent of tract population aged 65+ living more than 1 mile from the nearest large grocery store, population-weighted county rollup", "year": "2019", "source": "USDA ERS Food Access Research Atlas (2019 LRAM vintage) via data/panel_county.csv", "tag": "measured/administrative (USDA tract-distance flag, population-weighted county aggregate)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2580, "n_unit": "counties", "x_filled": 3141, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.0537, "p": 0.00641, "ci95": [ -0.0921, -0.0151 ] }, "spearman": { "rho": -0.0461, "p": 0.0192 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": -0.1471, "p": 6.49e-14, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": -0.2142, "p": 1.46e-27, "n": 2521 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": -0.2167, "p": 2.42e-26, "n": 2347 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure (seniors-specific low-access share)", "rurality", "region" ], "note": "Seniors-specific version of the same new Food Access Research Atlas measure.", "fdr_q_within_level": 0.00729, "classification": "negligible", "verdict": "Raw negligible negative correlation r=-0.05 (95% CI -0.09 to -0.02), Spearman rho=-0.05, n=2580 counties, p=0.00641, BH q=0.0073. NEGLIGIBLE: raw r=-0.054 is statistically non-zero at n=2580 but below 0.10 in size; too small to interpret. Not a finding worth reporting as an association. Seniors-specific version of the same new Food Access Research Atlas measure. Provenance: pct_seniors_low_access_1mi is measured/administrative (USDA tract-distance flag, population-weighted county aggregate); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CM-13", "level": "county", "kind": "measured_vs_modelled", "x": { "name": "poverty_rate_popweighted_avg_pct", "unit": "county poverty rate: population-weighted average of the census-tract PovertyRate field", "year": "2015-2019 ACS vintage (tract file)", "source": "USDA ERS Food Access Research Atlas tract file via data/panel_county.csv", "tag": "survey estimate (ACS, tract-weighted rollup; a different geography/vintage than the panel's own pct_below_poverty)" }, "y": { "name": "pct_below_poverty", "unit": "percent (0-100)", "year": "2020-2024", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "n": 3133, "n_unit": "counties", "x_filled": 3141, "y_filled": 3135, "frame_rows": 3143, "pearson": { "r": 0.8207, "p": 0.0, "ci95": [ 0.8089, 0.8318 ] }, "spearman": { "rho": 0.8248, "p": 0.0 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "different geography (tract-population-weighted vs county ACS) and vintage" ], "note": "New Food Access Atlas tract-weighted poverty rate against the panel's existing county-level ACS poverty rate: same concept, two pipelines.", "fdr_q_within_level": 0.0, "classification": "consistency_check", "verdict": "Raw strong positive correlation r=0.82 (95% CI 0.81 to 0.83), Spearman rho=0.82, n=3133 counties, p=0.0, BH q=0. CONSISTENCY CHECK: r=0.821: agreement level between two versions of the outcome / measurement. New Food Access Atlas tract-weighted poverty rate against the panel's existing county-level ACS poverty rate: same concept, two pipelines. Provenance: poverty_rate_popweighted_avg_pct is survey estimate (ACS, tract-weighted rollup; a different geography/vintage than the panel's own pct_below_poverty); pct_below_poverty is survey estimate (ACS 5-year). Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CM-14", "level": "county", "kind": "measured_vs_modelled", "x": { "name": "median_family_income_popweighted_avg_usd", "unit": "county median family income, USD: population-weighted average of the census-tract MedianFamilyIncome field", "year": "2015-2019 ACS vintage (tract file)", "source": "USDA ERS Food Access Research Atlas tract file via data/panel_county.csv", "tag": "survey estimate (ACS, tract-weighted rollup; a different geography/vintage than the panel's own median_household_income)" }, "y": { "name": "median_household_income", "unit": "US dollars (nominal, ACS inflation-adjusted to survey-period dollars)", "year": "2020-2024 (ACS) or CHR vintage for 8 CT counties", "source": "US Census Bureau, American Community Survey 5-Year Estimates 2020-2024 (detailed tables), retrieved via the Census Reporter API (a non-key-gated public proxy of the official Census Bureau ACS API) via data/panel_county.csv", "tag": "survey estimate (ACS 5-year)" }, "n": 3141, "n_unit": "counties", "x_filled": 3141, "y_filled": 3143, "frame_rows": 3143, "pearson": { "r": 0.897, "p": 0.0, "ci95": [ 0.89, 0.9036 ] }, "spearman": { "rho": 0.8666, "p": 0.0 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "family vs household income definitions differ", "different geography and vintage" ], "note": "New Food Access Atlas tract-weighted median family income against the panel's existing median household income.", "fdr_q_within_level": 0.0, "classification": "consistency_check", "verdict": "Raw strong positive correlation r=0.90 (95% CI 0.89 to 0.90), Spearman rho=0.87, n=3141 counties, p=0.0, BH q=0. CONSISTENCY CHECK: r=0.897: agreement level between two versions of the outcome / measurement. New Food Access Atlas tract-weighted median family income against the panel's existing median household income. Provenance: median_family_income_popweighted_avg_usd is survey estimate (ACS, tract-weighted rollup; a different geography/vintage than the panel's own median_household_income); median_household_income is survey estimate (ACS 5-year). Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CM-15", "level": "county", "kind": "measured_vs_modelled", "x": { "name": "nursing_home_beds_per_1000", "unit": "beds per 1,000 population", "year": "n/a (NOT SOURCED for any county)", "source": "NOT SOURCED via data/panel_county.csv", "tag": "not sourced (documented 0-of-3,143 blank column; see CM-15 and the CANNOT_ANSWER note)" }, "y": { "name": "cms_nh_beds_per_1000_pop", "unit": "CMS-certified nursing-home beds per 1,000 population (county: against panel population; state: beds summed by state then divided by panel_state.csv's own pop_total, not a sum of per-1000 rates)", "year": "2026", "source": "CMS Nursing Home Care Compare Provider Info, Aug 2026 via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio); blank for Connecticut (0 matching CMS rows under either FIPS scheme) and 316 other counties with no Care Compare facility" }, "n": 0, "n_unit": "counties", "x_filled": 0, "y_filled": 2826, "frame_rows": 3143, "pearson": { "r": null, "p": null, "ci95": [ null, null ] }, "spearman": { "rho": null, "p": null }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "nursing_home_beds_per_1000 is NOT SOURCED at county level (0 of 3,143 rows filled)" ], "note": "Documents that the panel's pre-existing county nursing-home-beds column cannot be checked against the new CMS one because it has no data to check: this is not a consistency result, it is proof the new column is the only usable one.", "fdr_q_within_level": null, "classification": "cannot_compute", "verdict": "No coefficient: n=0 too small or zero variance. CANNOT COMPUTE: n=0 too small or zero variance. Documents that the panel's pre-existing county nursing-home-beds column cannot be checked against the new CMS one because it has no data to check: this is not a consistency result, it is proof the new column is the only usable one. Provenance: nursing_home_beds_per_1000 is not sourced (documented 0-of-3,143 blank column; see CM-15 and the CANNOT_ANSWER note); cms_nh_beds_per_1000_pop is measured/administrative (derived ratio); blank for Connecticut (0 matching CMS rows under either FIPS scheme) and 316 other counties with no Care Compare facility. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CX-56", "level": "county", "kind": "exposure", "x": { "name": "cms_nh_beds_per_1000_pop", "unit": "CMS-certified nursing-home beds per 1,000 population (county: against panel population; state: beds summed by state then divided by panel_state.csv's own pop_total, not a sum of per-1000 rates)", "year": "2026", "source": "CMS Nursing Home Care Compare Provider Info, Aug 2026 via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio); blank for Connecticut (0 matching CMS rows under either FIPS scheme) and 316 other counties with no Care Compare facility" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2495, "n_unit": "counties", "x_filled": 2826, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.2258, "p": 3.2e-30, "ci95": [ 0.1883, 0.2628 ] }, "spearman": { "rho": 0.26, "p": 7.83e-40 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1907, "p": 7.18e-22, "n": 2495 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.2023, "p": 4.84e-24, "n": 2451 }, "partial_strict": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south", "population_density_per_sqmi", "pct_black", "norc_ma_share_of_bene_2020", "pct_checkup_annual" ], "r": 0.2312, "p": 5.09e-29, "n": 2278 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "institutional diagnosis (nursing-home residents are screened)", "region", "rurality" ], "note": "New in Phase 4 (CMS Care Compare certified beds): blank for Connecticut's 8 legacy counties and 309 further counties with no Care Compare facility (2,826 of 3,143 counties populated).", "fdr_q_within_level": 5.92e-30, "classification": "survives", "verdict": "Raw weak positive correlation r=0.23 (95% CI 0.19 to 0.26), Spearman rho=0.26, n=2495 counties, p=3.2e-30, BH q=5.9e-30. SURVIVES the confounder test: partial r=0.191 (p=7.18e-22) controlling for pct_85plus+median_household_income; extended partial r=0.2023; strict partial r=0.2312 controlling for pct_85plus+median_household_income+primary_care_physicians_per_100k+region_south+population_density_per_sqmi+pct_black+norc_ma_share_of_bene_2020+pct_checkup_annual. A place-level association not explained by the named control(s); candidate third things still unaddressed: institutional diagnosis (nursing-home residents are screened), region, rurality. New in Phase 4 (CMS Care Compare certified beds): blank for Connecticut's 8 legacy counties and 309 further counties with no Care Compare facility (2,826 of 3,143 counties populated). Provenance: cms_nh_beds_per_1000_pop is measured/administrative (derived ratio); blank for Connecticut (0 matching CMS rows under either FIPS scheme) and 316 other counties with no Care Compare facility; norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CA-7", "level": "county", "kind": "artefact_test", "x": { "name": "cms_nh_certified_beds_total", "unit": "count of CMS-certified nursing-home beds (facility-level Care Compare provider file, summed to county/state)", "year": "2026", "source": "CMS Nursing Home Care Compare Provider Info, Aug 2026 via data/panel_county.csv / data/panel_state.csv (data/sources/cms_nursing_home_provider_info)", "tag": "measured/administrative (CMS certification count)" }, "y": { "name": "norc65_prev_cnt_2020", "unit": "count of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis (raw count, scales with population)", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2495, "n_unit": "counties", "x_filled": 2826, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.9238, "p": 0.0, "ci95": [ 0.9178, 0.9293 ] }, "spearman": { "rho": 0.868, "p": 0.0 }, "partial_primary": { "controls": [ "population" ], "r": 0.3902, "p": 1.56e-91, "n": 2495 }, "partial_extended": { "controls": [ "population" ], "r": 0.3902, "p": 1.56e-91, "n": 2495 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "population size" ], "note": "New count-vs-count pair with the same population-artefact structure as CA-2/CA-3.", "fdr_q_within_level": 0.0, "classification": "artefact_collapses", "verdict": "Raw strong positive correlation r=0.92 (95% CI 0.92 to 0.93), Spearman rho=0.87, n=2495 counties, p=0.0, BH q=0. POPULATION/STRUCTURE ARTEFACT: raw r=0.924 falls to partial r=0.390 controlling for population. The raw coefficient is not a relationship. New count-vs-count pair with the same population-artefact structure as CA-2/CA-3. Provenance: cms_nh_certified_beds_total is measured/administrative (CMS certification count); norc65_prev_cnt_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CM-16", "level": "county", "kind": "measured_vs_modelled", "x": { "name": "norc_ma_share_of_bene_2020", "unit": "Medicare Advantage beneficiaries as percent of all beneficiaries", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, INSURANCE=2 / total", "tag": "measured/administrative (derived ratio)" }, "y": { "name": "medicare_advantage_penetration_pct", "unit": "Medicare Advantage + other plan beneficiaries as percent of total Medicare beneficiaries, June 2026 (direct division of two published CMS counts, not modelled); an independent enrollment-based MA share, distinct from norc_ma_share_of_bene_2020 which is the MA share among Medicare beneficiaries who appear in the 2020 dementia claims file", "year": "2026-06", "source": "CMS Medicare Monthly Enrollment via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio); blank for Connecticut's 8 legacy counties and small-cell-suppressed counties" }, "n": 3073, "n_unit": "counties", "x_filled": 3087, "y_filled": 3108, "frame_rows": 3143, "pearson": { "r": 0.9079, "p": 0.0, "ci95": [ 0.9015, 0.9139 ] }, "spearman": { "rho": 0.9138, "p": 0.0 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "different denominators: NORC's dementia-claims beneficiary population vs CMS's total Medicare population", "different years (2020 vs June 2026)" ], "note": "Do two independently-sourced county MA-penetration measures agree?", "fdr_q_within_level": 0.0, "classification": "consistency_check", "verdict": "Raw strong positive correlation r=0.91 (95% CI 0.90 to 0.91), Spearman rho=0.91, n=3073 counties, p=0.0, BH q=0. CONSISTENCY CHECK: r=0.908: agreement level between two versions of the outcome / measurement. Do two independently-sourced county MA-penetration measures agree? Provenance: norc_ma_share_of_bene_2020 is measured/administrative (derived ratio); medicare_advantage_penetration_pct is measured/administrative (derived ratio); blank for Connecticut's 8 legacy counties and small-cell-suppressed counties. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "CD-7", "level": "county", "kind": "diagnosis_intensity", "x": { "name": "medicare_advantage_penetration_pct", "unit": "Medicare Advantage + other plan beneficiaries as percent of total Medicare beneficiaries, June 2026 (direct division of two published CMS counts, not modelled); an independent enrollment-based MA share, distinct from norc_ma_share_of_bene_2020 which is the MA share among Medicare beneficiaries who appear in the 2020 dementia claims file", "year": "2026-06", "source": "CMS Medicare Monthly Enrollment via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio); blank for Connecticut's 8 legacy counties and small-cell-suppressed counties" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3108, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.2869, "p": 6.09e-50, "ci95": [ 0.2511, 0.322 ] }, "spearman": { "rho": 0.2867, "p": 7.58e-50 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1945, "p": 2.35e-23, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.1677, "p": 2.36e-17, "n": 2521 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "health-system contact", "coding practice", "insurance type mix" ], "note": "New in Phase 4: CMS enrollment-based MA share of ALL Medicare beneficiaries (independent of the NORC dementia file used to build norc_ma_share_of_bene_2020, CD-3). Tests whether the CD-3 coding-intensity signal replicates with a completely independent numerator and denominator.", "fdr_q_within_level": 1.54e-49, "classification": "survives", "verdict": "Raw weak positive correlation r=0.29 (95% CI 0.25 to 0.32), Spearman rho=0.29, n=2572 counties, p=6.09e-50, BH q=1.5e-49. SURVIVES the confounder test: partial r=0.195 (p=2.35e-23) controlling for pct_85plus+median_household_income; extended partial r=0.1677. A place-level association not explained by the named control(s); candidate third things still unaddressed: health-system contact, coding practice, insurance type mix. New in Phase 4: CMS enrollment-based MA share of ALL Medicare beneficiaries (independent of the NORC dementia file used to build norc_ma_share_of_bene_2020, CD-3). Tests whether the CD-3 coding-intensity signal replicates with a completely independent numerator and denominator. Provenance: medicare_advantage_penetration_pct is measured/administrative (derived ratio); blank for Connecticut's 8 legacy counties and small-cell-suppressed counties; norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CD-8", "level": "county", "kind": "diagnosis_intensity", "x": { "name": "naics62_health_socasst_emplvl_per_1000pop", "unit": "NAICS 62 Health Care and Social Assistance employment level per 1,000 population (county: panel population; state: pop_total)", "year": "2024 employment / panel population year", "source": "derived in analysis/build_analysis_set.py from naics62_health_socasst_emplvl_annual_avg and population, BLS QCEW via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio; inherits the suppression floor of the employment count)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2572, "n_unit": "counties", "x_filled": 3121, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": 0.0336, "p": 0.0885, "ci95": [ -0.0051, 0.0721 ] }, "spearman": { "rho": 0.0119, "p": 0.548 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.1217, "p": 5.96e-10, "n": 2572 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.156, "p": 3.3e-15, "n": 2521 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "urbanicity", "income", "detection capacity (broad health + social-assistance sector, not physicians specifically)" ], "note": "New in Phase 4 (BLS QCEW). Employment counts are a suppression FLOOR where naics62_any_ownership_cell_suppressed='Y' (2,794 of 7,749 county-ownership cells nationally).", "fdr_q_within_level": 0.098, "classification": "null", "verdict": "Raw negligible positive correlation r=0.03 (95% CI -0.01 to 0.07), Spearman rho=0.01, n=2572 counties, p=0.0885, BH q=0.098. NULL: raw association not distinguishable from zero (BH q=0.098 within level). Absence of a place-level correlation is not evidence of no individual-level effect. New in Phase 4 (BLS QCEW). Employment counts are a suppression FLOOR where naics62_any_ownership_cell_suppressed='Y' (2,794 of 7,749 county-ownership cells nationally). Provenance: naics62_health_socasst_emplvl_per_1000pop is measured/administrative (derived ratio; inherits the suppression floor of the employment count); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CD-9", "level": "county", "kind": "diagnosis_intensity", "x": { "name": "naics623_nursing_residential_care_emplvl_per_1000pop", "unit": "NAICS 623 Nursing and Residential Care Facilities employment level per 1,000 population (county: panel population; state: pop_total)", "year": "2024 employment / panel population year", "source": "derived in analysis/build_analysis_set.py from naics623_nursing_residential_care_emplvl_annual_avg and population, BLS QCEW via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio; inherits the suppression floor of the employment count)" }, "y": { "name": "norc65_prevalence_rate_2020", "unit": "percent of Medicare beneficiaries aged 65+ with a prevalent dementia diagnosis", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF (NORC at the University of Chicago), Medicare FFS+MA claims, AGECAT2=1 county row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 2547, "n_unit": "counties", "x_filled": 2937, "y_filled": 2580, "frame_rows": 3143, "pearson": { "r": -0.0067, "p": 0.736, "ci95": [ -0.0455, 0.0322 ] }, "spearman": { "rho": 0.007, "p": 0.725 }, "partial_primary": { "controls": [ "pct_85plus", "median_household_income" ], "r": 0.0614, "p": 0.00193, "n": 2547 }, "partial_extended": { "controls": [ "pct_85plus", "median_household_income", "primary_care_physicians_per_100k", "region_south" ], "r": 0.128, "p": 1.35e-10, "n": 2499 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "long-term-care workforce presence", "overlaps conceptually with cms_nh_beds_per_1000_pop from a different agency" ], "note": "New in Phase 4 (BLS QCEW, NAICS 623 nursing/residential-care subsector). Same suppression-floor caveat as naics62.", "fdr_q_within_level": 0.761, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.01 (95% CI -0.05 to 0.03), Spearman rho=0.01, n=2547 counties, p=0.736, BH q=0.76. NULL: raw association not distinguishable from zero (BH q=0.76 within level). Absence of a place-level correlation is not evidence of no individual-level effect. New in Phase 4 (BLS QCEW, NAICS 623 nursing/residential-care subsector). Same suppression-floor caveat as naics62. Provenance: naics623_nursing_residential_care_emplvl_per_1000pop is measured/administrative (derived ratio; inherits the suppression floor of the employment count); norc65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "CM-17", "level": "county", "kind": "consistency", "x": { "name": "primary_care_physicians_per_100k", "unit": "rate per 100,000", "year": "~2022 (Area Health Resources Files vintage embedded in CHR 2024)", "source": "County Health Rankings & Roadmaps (Univ. of Wisconsin Population Health Institute), 2024 release via data/panel_county.csv", "tag": "measured/administrative (AHRF headcount)" }, "y": { "name": "naics62_health_socasst_emplvl_per_1000pop", "unit": "NAICS 62 Health Care and Social Assistance employment level per 1,000 population (county: panel population; state: pop_total)", "year": "2024 employment / panel population year", "source": "derived in analysis/build_analysis_set.py from naics62_health_socasst_emplvl_annual_avg and population, BLS QCEW via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio; inherits the suppression floor of the employment count)" }, "n": 2964, "n_unit": "counties", "x_filled": 2986, "y_filled": 3121, "frame_rows": 3143, "pearson": { "r": 0.5817, "p": 4.69e-268, "ci95": [ 0.5574, 0.6051 ] }, "spearman": { "rho": 0.5229, "p": 1.06e-207 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "AHRF counts physicians only; QCEW counts every paid worker in the health+social-assistance sector" ], "note": "Do a physician-specific supply measure (AHRF, pre-existing) and a broad health-workforce density measure (BLS QCEW, new) agree spatially?", "fdr_q_within_level": 4.15e-267, "classification": "consistency_check", "verdict": "Raw strong positive correlation r=0.58 (95% CI 0.56 to 0.61), Spearman rho=0.52, n=2964 counties, p=4.69e-268, BH q=4.1e-267. CONSISTENCY CHECK: r=0.582: agreement level between two versions of the outcome / measurement. Do a physician-specific supply measure (AHRF, pre-existing) and a broad health-workforce density measure (BLS QCEW, new) agree spatially? Provenance: primary_care_physicians_per_100k is measured/administrative (AHRF headcount); naics62_health_socasst_emplvl_per_1000pop is measured/administrative (derived ratio; inherits the suppression floor of the employment count). Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SK-1", "level": "state", "kind": "artefact_test", "x": { "name": "ts_dementia_deaths_pct_g30_2017", "unit": "percent of all dementia-coded deaths (G30+F01+F03+G31) whose underlying cause was coded G30 Alzheimer's", "year": "2017", "source": "CDC WONDER via data/panel_state_timeseries.csv", "tag": "measured/administrative (coding share)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2017", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted", "year": "2017", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.8183, "p": 2.25e-13, "ci95": [ 0.7008, 0.8926 ] }, "spearman": { "rho": 0.8546, "p": 1.49e-15 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "certifier coding practice", "state vital-statistics query rules" ], "note": "Same year, same file: the share of a state's dementia deaths that certifiers code as G30 Alzheimer's against the state's age-adjusted 'Alzheimer's' death rate. A strong correlation means the state Alzheimer's mortality league table is largely a coding-practice table.", "fdr_q_within_level": 2.44e-12, "classification": "artefact_documented", "verdict": "Raw strong positive correlation r=0.82 (95% CI 0.70 to 0.89), Spearman rho=0.85, n=51 states (incl. DC), p=2.25e-13, BH q=2.4e-12. STRUCTURAL DEPENDENCE DOCUMENTED: r=0.818, q=2.4e-12: documents the mechanical/structural dependence named. Same year, same file: the share of a state's dementia deaths that certifiers code as G30 Alzheimer's against the state's age-adjusted 'Alzheimer's' death rate. A strong correlation means the state Alzheimer's mortality league table is largely a coding-practice table. Provenance: ts_dementia_deaths_pct_g30_2017 is measured/administrative (coding share); ts_alz_age_adjusted_death_rate_2017 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SK-2", "level": "state", "kind": "artefact_test", "x": { "name": "ts_dementia_deaths_pct_g30_2017", "unit": "percent of all dementia-coded deaths (G30+F01+F03+G31) whose underlying cause was coded G30 Alzheimer's", "year": "2017", "source": "CDC WONDER via data/panel_state_timeseries.csv", "tag": "measured/administrative (coding share)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.7892, "p": 6.02e-12, "ci95": [ 0.6563, 0.8745 ] }, "spearman": { "rho": 0.8185, "p": 2.21e-13 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "coding practice persists over years" ], "note": "2017 coding share against the 2022 age-adjusted rate.", "fdr_q_within_level": 5.72e-11, "classification": "artefact_documented", "verdict": "Raw strong positive correlation r=0.79 (95% CI 0.66 to 0.87), Spearman rho=0.82, n=51 states (incl. DC), p=6.02e-12, BH q=5.7e-11. STRUCTURAL DEPENDENCE DOCUMENTED: r=0.789, q=5.7e-11: documents the mechanical/structural dependence named. 2017 coding share against the 2022 age-adjusted rate. Provenance: ts_dementia_deaths_pct_g30_2017 is measured/administrative (coding share); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SK-3", "level": "state", "kind": "artefact_test", "x": { "name": "ts_dementia_deaths_pct_f03_2017", "unit": "percent of all dementia-coded deaths coded F03 unspecified dementia", "year": "2017", "source": "CDC WONDER via data/panel_state_timeseries.csv", "tag": "measured/administrative (coding share)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2017", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted", "year": "2017", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.7143, "p": 3.97e-09, "ci95": [ -0.8271, -0.5462 ] }, "spearman": { "rho": -0.7562, "p": 1.4e-10 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "coding practice" ], "note": "Mirror image: share coded F03 'unspecified dementia' (what G30 displaces).", "fdr_q_within_level": 3.02e-08, "classification": "artefact_documented", "verdict": "Raw strong negative correlation r=-0.71 (95% CI -0.83 to -0.55), Spearman rho=-0.76, n=51 states (incl. DC), p=3.97e-09, BH q=3e-08. STRUCTURAL DEPENDENCE DOCUMENTED: r=-0.714, q=3e-08: documents the mechanical/structural dependence named. Mirror image: share coded F03 'unspecified dementia' (what G30 displaces). Provenance: ts_dementia_deaths_pct_f03_2017 is measured/administrative (coding share); ts_alz_age_adjusted_death_rate_2017 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SK-4", "level": "state", "kind": "consistency", "x": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "y": { "name": "alz_deaths_per_100k", "unit": "rate per 100k", "year": "2022", "source": "Alzheimer's Association 2025 Facts and Figures, Table 6 (NCHS) via data/panel_state.csv", "tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.8311, "p": 4.41e-14, "ci95": [ 0.7206, 0.9005 ] }, "spearman": { "rho": 0.7712, "p": 3.55e-11 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure" ], "note": "Age-adjusted rate against the unadjusted rate the baseline used for S1, S4, S6, S7, S9, B1, B3, B5, B6, A7.", "fdr_q_within_level": 5.59e-13, "classification": "consistency_check", "verdict": "Raw strong positive correlation r=0.83 (95% CI 0.72 to 0.90), Spearman rho=0.77, n=51 states (incl. DC), p=4.41e-14, BH q=5.6e-13. CONSISTENCY CHECK: r=0.831: agreement level between two versions of the outcome / measurement. Age-adjusted rate against the unadjusted rate the baseline used for S1, S4, S6, S7, S9, B1, B3, B5, B6, A7. Provenance: ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration); alz_deaths_per_100k is modelled (Alzheimer's Association projection built from age/sex/race distributions). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SK-5", "level": "state", "kind": "measured_vs_modelled", "x": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "alz_prevalence_65plus_pct", "unit": "percent", "year": "2020", "source": "Alzheimer's Association 2025 Alzheimer's Disease Facts and Figures, Table 5 (data from Dhana et al., state/county prevalence model) via data/panel_state.csv", "tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.6447, "p": 3.3e-07, "ci95": [ 0.4488, 0.7814 ] }, "spearman": { "rho": 0.7153, "p": 3.69e-09 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "the AA model is built from age/sex/race distributions, not from claims" ], "note": "Measured claims-based diagnosed prevalence (NORC) against the Alzheimer's Association MODELLED prevalence used by the baseline for S2, S3, S10, S12, B2, B4, A3.", "fdr_q_within_level": 1.93e-06, "classification": "consistency_check", "verdict": "Raw strong positive correlation r=0.64 (95% CI 0.45 to 0.78), Spearman rho=0.72, n=51 states (incl. DC), p=3.3e-07, BH q=1.9e-06. CONSISTENCY CHECK: r=0.645: agreement level between two versions of the outcome / measurement. Measured claims-based diagnosed prevalence (NORC) against the Alzheimer's Association MODELLED prevalence used by the baseline for S2, S3, S10, S12, B2, B4, A3. Provenance: norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases); alz_prevalence_65plus_pct is modelled (Alzheimer's Association projection built from age/sex/race distributions). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SK-6", "level": "state", "kind": "consistency", "x": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.0266, "p": 0.853, "ci95": [ -0.3, 0.2508 ] }, "spearman": { "rho": -0.0444, "p": 0.757 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "coding practice", "diagnosis intensity" ], "note": "Two measured outcomes that should agree if both track underlying disease: diagnosed prevalence vs age-adjusted mortality.", "fdr_q_within_level": 0.939, "classification": "consistency_check", "verdict": "Raw negligible negative correlation r=-0.03 (95% CI -0.30 to 0.25), Spearman rho=-0.04, n=51 states (incl. DC), p=0.853, BH q=0.94. CONSISTENCY CHECK: r=-0.027: agreement level between two versions of the outcome / measurement. Two measured outcomes that should agree if both track underlying disease: diagnosed prevalence vs age-adjusted mortality. Provenance: norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SK-7", "level": "state", "kind": "consistency", "x": { "name": "alz_prevalence_65plus_pct", "unit": "percent", "year": "2020", "source": "Alzheimer's Association 2025 Alzheimer's Disease Facts and Figures, Table 5 (data from Dhana et al., state/county prevalence model) via data/panel_state.csv", "tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.3873, "p": 0.00499, "ci95": [ -0.5989, -0.1251 ] }, "spearman": { "rho": -0.2143, "p": 0.131 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "coding practice" ], "note": "Modelled prevalence vs measured age-adjusted mortality.", "fdr_q_within_level": 0.0119, "classification": "consistency_check", "verdict": "Raw moderate negative correlation r=-0.39 (95% CI -0.60 to -0.13), Spearman rho=-0.21, n=51 states (incl. DC), p=0.00499, BH q=0.012. CONSISTENCY CHECK: r=-0.387: agreement level between two versions of the outcome / measurement. Modelled prevalence vs measured age-adjusted mortality. Provenance: alz_prevalence_65plus_pct is modelled (Alzheimer's Association projection built from age/sex/race distributions); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SK-8", "level": "state", "kind": "consistency", "x": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "ts_scd_reported__65plus_pct_latest", "unit": "percent of adults 65+ reporting subjective cognitive decline, most recent BRFSS year available per state (2015-2022)", "year": "2015-2022 (latest per state)", "source": "CDC BRFSS Cognitive Decline module via data/panel_state_timeseries.csv", "tag": "survey estimate (self-report)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.1787, "p": 0.21, "ci95": [ -0.4329, 0.1019 ] }, "spearman": { "rho": -0.1888, "p": 0.185 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "self-report culture", "survey response" ], "note": "Diagnosed prevalence vs self-reported subjective cognitive decline among 65+.", "fdr_q_within_level": 0.332, "classification": "consistency_check", "verdict": "Raw weak negative correlation r=-0.18 (95% CI -0.43 to 0.10), Spearman rho=-0.19, n=51 states (incl. DC), p=0.21, BH q=0.33. CONSISTENCY CHECK: r=-0.179: agreement level between two versions of the outcome / measurement. Diagnosed prevalence vs self-reported subjective cognitive decline among 65+. Provenance: norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases); ts_scd_reported__65plus_pct_latest is survey estimate (self-report). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SK-9", "level": "state", "kind": "consistency", "x": { "name": "norc_state_prevalence_rate_2020", "unit": "diagnosed dementia prevalence, all-age Medicare beneficiaries, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, state total row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.9837, "p": 3.76e-38, "ci95": [ 0.9714, 0.9907 ] }, "spearman": { "rho": 0.9862, "p": 6.67e-40 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "All-age vs 65+ state rates: twin check.", "fdr_q_within_level": 2.86e-36, "classification": "twin", "verdict": "Raw strong positive correlation r=0.98 (95% CI 0.97 to 0.99), Spearman rho=0.99, n=51 states (incl. DC), p=3.76e-38, BH q=2.9e-36. TWIN: |r|=0.984 >= 0.95: the two columns are the same quantity twice; never use both in one model. All-age vs 65+ state rates: twin check. Provenance: norc_state_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SK-10", "level": "state", "kind": "consistency", "x": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "ts_care_for_dementia__overall_pct_latest", "unit": "percent of caregivers whose care recipient has dementia, latest BRFSS year per state", "year": "2015-2022 (latest per state)", "source": "CDC BRFSS Caregiver module via data/panel_state_timeseries.csv", "tag": "survey estimate (self-report)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.0287, "p": 0.842, "ci95": [ -0.3019, 0.2489 ] }, "spearman": { "rho": -0.0577, "p": 0.688 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "caregiver survey response" ], "note": "Diagnosed prevalence vs share of caregivers caring for someone with dementia.", "fdr_q_within_level": 0.939, "classification": "consistency_check", "verdict": "Raw negligible negative correlation r=-0.03 (95% CI -0.30 to 0.25), Spearman rho=-0.06, n=51 states (incl. DC), p=0.842, BH q=0.94. CONSISTENCY CHECK: r=-0.029: agreement level between two versions of the outcome / measurement. Diagnosed prevalence vs share of caregivers caring for someone with dementia. Provenance: norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases); ts_care_for_dementia__overall_pct_latest is survey estimate (self-report). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SU-1", "level": "state", "kind": "uninterpretable", "x": { "name": "pct_race_black", "unit": "percent", "year": "2023", "source": "U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT0202 via data/panel_state.csv", "tag": "survey estimate (ACS 1-year)" }, "y": { "name": "alz_prevalence_65plus_pct", "unit": "percent", "year": "2020", "source": "Alzheimer's Association 2025 Alzheimer's Disease Facts and Figures, Table 5 (data from Dhana et al., state/county prevalence model) via data/panel_state.csv", "tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.7668, "p": 5.36e-11, "ci95": [ 0.6228, 0.8605 ] }, "spearman": { "rho": 0.7869, "p": 7.59e-12 }, "partial_primary": { "controls": [ "pct_85plus" ], "r": 0.8023, "p": 1.46e-12, "n": 51 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "race is an INPUT to the outcome model" ], "note": "The Alzheimer's Association prevalence estimate is computed FROM each state's age/sex/race distribution using fixed race-specific rates. Correlating it with pct_race_black recovers the model's own coefficient. Cannot be read as evidence about place.", "fdr_q_within_level": 4.53e-10, "classification": "uninterpretable", "verdict": "Raw strong positive correlation r=0.77 (95% CI 0.62 to 0.86), Spearman rho=0.79, n=51 states (incl. DC), p=5.36e-11, BH q=4.5e-10. UNINTERPRETABLE: flagged a priori: race is an INPUT to the outcome model. Documented so that nobody publishes it as a finding. The Alzheimer's Association prevalence estimate is computed FROM each state's age/sex/race distribution using fixed race-specific rates. Correlating it with pct_race_black recovers the model's own coefficient. Cannot be read as evidence about place. Provenance: pct_race_black is survey estimate (ACS 1-year); alz_prevalence_65plus_pct is modelled (Alzheimer's Association projection built from age/sex/race distributions). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SU-2", "level": "state", "kind": "uninterpretable", "x": { "name": "pct_race_hispanic", "unit": "percent", "year": "2024", "source": "KFF State Health Facts, Distribution by Race/Ethnicity (ACS 1-Year Estimates) via data/panel_state.csv", "tag": "survey estimate (BRFSS direct)" }, "y": { "name": "alz_prevalence_65plus_pct", "unit": "percent", "year": "2020", "source": "Alzheimer's Association 2025 Alzheimer's Disease Facts and Figures, Table 5 (data from Dhana et al., state/county prevalence model) via data/panel_state.csv", "tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.2385, "p": 0.092, "ci95": [ -0.0397, 0.4823 ] }, "spearman": { "rho": 0.339, "p": 0.0149 }, "partial_primary": { "controls": [ "pct_85plus" ], "r": 0.2389, "p": 0.0914, "n": 51 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "race/ethnicity is an INPUT to the outcome model" ], "note": "Same circularity as SU-1 (baseline B2 used this pair).", "fdr_q_within_level": 0.171, "classification": "uninterpretable", "verdict": "Raw weak positive correlation r=0.24 (95% CI -0.04 to 0.48), Spearman rho=0.34, n=51 states (incl. DC), p=0.092, BH q=0.17. UNINTERPRETABLE: flagged a priori: race/ethnicity is an INPUT to the outcome model. Documented so that nobody publishes it as a finding. Same circularity as SU-1 (baseline B2 used this pair). Provenance: pct_race_hispanic is survey estimate (BRFSS direct); alz_prevalence_65plus_pct is modelled (Alzheimer's Association projection built from age/sex/race distributions). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SU-3", "level": "state", "kind": "uninterpretable", "x": { "name": "pct_85plus", "unit": "percent", "year": "2023", "source": "U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT0104 via data/panel_state.csv", "tag": "survey estimate (ACS 1-year)" }, "y": { "name": "alz_prevalence_65plus_pct", "unit": "percent", "year": "2020", "source": "Alzheimer's Association 2025 Alzheimer's Disease Facts and Figures, Table 5 (data from Dhana et al., state/county prevalence model) via data/panel_state.csv", "tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.005, "p": 0.972, "ci95": [ -0.2802, 0.271 ] }, "spearman": { "rho": 0.0655, "p": 0.648 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age is an INPUT to the outcome model" ], "note": "Same circularity: age structure is a model input.", "fdr_q_within_level": 0.982, "classification": "uninterpretable", "verdict": "Raw negligible negative correlation r=-0.01 (95% CI -0.28 to 0.27), Spearman rho=0.07, n=51 states (incl. DC), p=0.972, BH q=0.98. UNINTERPRETABLE: flagged a priori: age is an INPUT to the outcome model. Documented so that nobody publishes it as a finding. Same circularity: age structure is a model input. Provenance: pct_85plus is survey estimate (ACS 1-year); alz_prevalence_65plus_pct is modelled (Alzheimer's Association projection built from age/sex/race distributions). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SP-1", "level": "state", "kind": "exposure", "x": { "name": "pct_race_black", "unit": "percent", "year": "2023", "source": "U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT0202 via data/panel_state.csv", "tag": "survey estimate (ACS 1-year)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.6508, "p": 2.33e-07, "ci95": [ 0.4573, 0.7855 ] }, "spearman": { "rho": 0.7472, "p": 3.03e-10 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus" ], "r": 0.6823, "p": 3.51e-08, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.5167, "p": 0.000104, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd" ], "r": 0.4073, "p": 0.00301, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region (South)", "poverty", "coding by race" ], "note": "Measured claims-based replacement for the baseline's modelled-prevalence pairs.", "fdr_q_within_level": 1.48e-06, "classification": "survives", "verdict": "Raw strong positive correlation r=0.65 (95% CI 0.46 to 0.79), Spearman rho=0.75, n=51 states (incl. DC), p=2.33e-07, BH q=1.5e-06. SURVIVES the confounder test: partial r=0.682 (p=3.51e-08) controlling for median_household_income+pct_85plus; extended partial r=0.5167; strict partial r=0.4073 controlling for median_household_income+pct_85plus+region_south+medicare_percapita_ffs_payment_usd. A place-level association not explained by the named control(s); candidate third things still unaddressed: region (South), poverty, coding by race. Measured claims-based replacement for the baseline's modelled-prevalence pairs. Provenance: pct_race_black is survey estimate (ACS 1-year); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-2", "level": "state", "kind": "exposure", "x": { "name": "pct_race_hispanic", "unit": "percent", "year": "2024", "source": "KFF State Health Facts, Distribution by Race/Ethnicity (ACS 1-Year Estimates) via data/panel_state.csv", "tag": "survey estimate (BRFSS direct)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.0438, "p": 0.76, "ci95": [ -0.2346, 0.3156 ] }, "spearman": { "rho": 0.0675, "p": 0.638 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus" ], "r": 0.1093, "p": 0.445, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.1728, "p": 0.225, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.0437, "p": 0.761, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region", "under-diagnosis in Spanish-speaking populations" ], "note": "", "fdr_q_within_level": 0.902, "classification": "null", "verdict": "Raw negligible positive correlation r=0.04 (95% CI -0.23 to 0.32), Spearman rho=0.07, n=51 states (incl. DC), p=0.76, BH q=0.9. NULL: raw association not distinguishable from zero (BH q=0.9 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_race_hispanic is survey estimate (BRFSS direct); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-3", "level": "state", "kind": "exposure", "x": { "name": "pct_below_poverty", "unit": "percent", "year": "2023", "source": "U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT1701 via data/panel_state.csv", "tag": "survey estimate (ACS 1-year)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.5584, "p": 2.06e-05, "ci95": [ 0.3342, 0.7227 ] }, "spearman": { "rho": 0.5642, "p": 1.62e-05 }, "partial_primary": { "controls": [ "pct_85plus" ], "r": 0.5631, "p": 1.69e-05, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.422, "p": 0.00204, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.1503, "p": 0.292, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region", "race composition" ], "note": "", "fdr_q_within_level": 9.78e-05, "classification": "collapses_strict", "verdict": "Raw strong positive correlation r=0.56 (95% CI 0.33 to 0.72), Spearman rho=0.56, n=51 states (incl. DC), p=2.06e-05, BH q=9.8e-05. COLLAPSES under place / diagnosis-intensity controls: survives age/income/region controls (extended partial r=0.422) but falls to 0.150 once place and diagnosis-intensity markers are added (median_household_income+pct_85plus+region_south+medicare_percapita_ffs_payment_usd+pct_race_black). What looked like an exposure effect is carried by where the county is and how intensively it codes diagnoses. Provenance: pct_below_poverty is survey estimate (ACS 1-year); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-4", "level": "state", "kind": "exposure", "x": { "name": "median_household_income", "unit": "usd", "year": "2023", "source": "U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT1901 via data/panel_state.csv", "tag": "survey estimate (ACS 1-year)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.2695, "p": 0.0558, "ci95": [ -0.5074, 0.0066 ] }, "spearman": { "rho": -0.3062, "p": 0.0289 }, "partial_primary": { "controls": [ "pct_85plus" ], "r": -0.2704, "p": 0.055, "n": 51 }, "partial_extended": { "controls": [ "pct_85plus", "region_south" ], "r": -0.0933, "p": 0.515, "n": 51 }, "partial_strict": { "controls": [ "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": -0.3706, "p": 0.00742, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "negative", "why": "higher SES is protective at the individual level" }, "confounders_named": [ "region", "race composition", "diagnosis intensity" ], "note": "", "fdr_q_within_level": 0.116, "classification": "null", "verdict": "Raw weak negative correlation r=-0.27 (95% CI -0.51 to 0.01), Spearman rho=-0.31, n=51 states (incl. DC), p=0.0558, BH q=0.12. NULL: raw association not distinguishable from zero (BH q=0.12 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: median_household_income is survey estimate (ACS 1-year); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-5", "level": "state", "kind": "exposure", "x": { "name": "pct_bachelors_plus", "unit": "percent", "year": "2023", "source": "U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT1502 via data/panel_state.csv", "tag": "survey estimate (ACS 1-year)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.0875, "p": 0.541, "ci95": [ -0.3545, 0.1927 ] }, "spearman": { "rho": -0.1583, "p": 0.267 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus" ], "r": 0.2798, "p": 0.0468, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.1725, "p": 0.226, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.0663, "p": 0.644, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "negative", "why": "education is protective" }, "confounders_named": [ "income", "region" ], "note": "", "fdr_q_within_level": 0.685, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.09 (95% CI -0.35 to 0.19), Spearman rho=-0.16, n=51 states (incl. DC), p=0.541, BH q=0.69. NULL: raw association not distinguishable from zero (BH q=0.69 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_bachelors_plus is survey estimate (ACS 1-year); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-6", "level": "state", "kind": "exposure", "x": { "name": "pct_rural", "unit": "percent", "year": "2023", "source": "America's Health Rankings, Rural Population (Census ACS 1-Year) via data/panel_state.csv", "tag": "survey estimate (ACS)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.1491, "p": 0.296, "ci95": [ -0.4079, 0.1319 ] }, "spearman": { "rho": -0.1798, "p": 0.207 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus" ], "r": -0.4062, "p": 0.0031, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": -0.437, "p": 0.00134, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": -0.1908, "p": 0.18, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region", "specialist supply" ], "note": "", "fdr_q_within_level": 0.433, "classification": "null", "verdict": "Raw weak negative correlation r=-0.15 (95% CI -0.41 to 0.13), Spearman rho=-0.18, n=51 states (incl. DC), p=0.296, BH q=0.43. NULL: raw association not distinguishable from zero (BH q=0.43 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_rural is survey estimate (ACS); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-7", "level": "state", "kind": "exposure", "x": { "name": "pct_diabetes", "unit": "percent", "year": "2024", "source": "KFF State Health Facts, Adults with Diagnosed Diabetes (CDC BRFSS) via data/panel_state.csv", "tag": "survey estimate (BRFSS direct)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 50, "n_unit": "states (incl. DC)", "x_filled": 50, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.5572, "p": 2.63e-05, "ci95": [ 0.3301, 0.7234 ] }, "spearman": { "rho": 0.5348, "p": 6.3e-05 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.3473, "p": 0.0135, "n": 50 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.3473, "p": 0.0135, "n": 50 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.1742, "p": 0.226, "n": 50 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "diabetes raises risk" }, "confounders_named": [ "region", "obesity" ], "note": "", "fdr_q_within_level": 0.000118, "classification": "collapses_strict", "verdict": "Raw strong positive correlation r=0.56 (95% CI 0.33 to 0.72), Spearman rho=0.53, n=50 states (incl. DC), p=2.63e-05, BH q=0.00012. COLLAPSES under place / diagnosis-intensity controls: survives age/income/region controls (extended partial r=0.3473) but falls to 0.174 once place and diagnosis-intensity markers are added (median_household_income+pct_85plus+region_south+medicare_percapita_ffs_payment_usd+pct_race_black). What looked like an exposure effect is carried by where the county is and how intensively it codes diagnoses. Provenance: pct_diabetes is survey estimate (BRFSS direct); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-8", "level": "state", "kind": "exposure", "x": { "name": "pct_obesity", "unit": "percent", "year": "2024", "source": "KFF State Health Facts, Adult Obesity (CDC BRFSS) via data/panel_state.csv", "tag": "survey estimate (BRFSS direct)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 50, "n_unit": "states (incl. DC)", "x_filled": 50, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.2289, "p": 0.11, "ci95": [ -0.0528, 0.4769 ] }, "spearman": { "rho": 0.2622, "p": 0.0658 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": -0.0203, "p": 0.889, "n": 50 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": -0.0203, "p": 0.889, "n": 50 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": -0.0498, "p": 0.731, "n": 50 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "midlife obesity raises risk" }, "confounders_named": [ "region", "diabetes" ], "note": "", "fdr_q_within_level": 0.19, "classification": "null", "verdict": "Raw weak positive correlation r=0.23 (95% CI -0.05 to 0.48), Spearman rho=0.26, n=50 states (incl. DC), p=0.11, BH q=0.19. NULL: raw association not distinguishable from zero (BH q=0.19 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_obesity is survey estimate (BRFSS direct); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-9", "level": "state", "kind": "exposure", "x": { "name": "pct_hypertension", "unit": "percent", "year": "2023", "source": "KFF State Health Facts, Adults Told They Have Hypertension (CDC BRFSS) via data/panel_state.csv", "tag": "survey estimate (BRFSS direct)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 49, "n_unit": "states (incl. DC)", "x_filled": 49, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.5348, "p": 7.54e-05, "ci95": [ 0.2985, 0.7093 ] }, "spearman": { "rho": 0.4418, "p": 0.00148 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.2746, "p": 0.0562, "n": 49 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.2746, "p": 0.0562, "n": 49 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.1482, "p": 0.31, "n": 49 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "midlife hypertension raises risk" }, "confounders_named": [ "region", "diabetes" ], "note": "", "fdr_q_within_level": 0.000318, "classification": "attenuated", "verdict": "Raw strong positive correlation r=0.53 (95% CI 0.30 to 0.71), Spearman rho=0.44, n=49 states (incl. DC), p=7.54e-05, BH q=0.00032. ATTENUATED: partial r=0.275 keeps >=50% of raw but is no longer significant (p=0.0562). Weak evidence at best. Provenance: pct_hypertension is survey estimate (BRFSS direct); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-10", "level": "state", "kind": "exposure", "x": { "name": "pct_smoking", "unit": "percent", "year": "2024", "source": "KFF State Health Facts, Adult Smoking Rate (CDC BRFSS) via data/panel_state.csv", "tag": "survey estimate (BRFSS direct)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 50, "n_unit": "states (incl. DC)", "x_filled": 50, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.1804, "p": 0.21, "ci95": [ -0.1031, 0.4369 ] }, "spearman": { "rho": 0.1538, "p": 0.286 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": -0.0982, "p": 0.497, "n": 50 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": -0.0982, "p": 0.497, "n": 50 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": -0.0167, "p": 0.908, "n": 50 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "smoking raises risk" }, "confounders_named": [ "region", "competing mortality" ], "note": "", "fdr_q_within_level": 0.332, "classification": "null", "verdict": "Raw weak positive correlation r=0.18 (95% CI -0.10 to 0.44), Spearman rho=0.15, n=50 states (incl. DC), p=0.21, BH q=0.33. NULL: raw association not distinguishable from zero (BH q=0.33 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_smoking is survey estimate (BRFSS direct); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-11", "level": "state", "kind": "exposure", "x": { "name": "pct_hearing_loss", "unit": "percent", "year": "2016", "source": "CDC/NCHS, Any Hearing Loss by State, National Health Interview Survey 2014-2016 (Blackwell & Norris) via data/panel_state.csv", "tag": "survey estimate (self-report)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.2561, "p": 0.0696, "ci95": [ -0.4967, 0.0209 ] }, "spearman": { "rho": -0.2329, "p": 0.1 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": -0.3883, "p": 0.00487, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": -0.3883, "p": 0.00487, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.0552, "p": 0.7, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "hearing loss is the largest modifiable risk factor in the Lancet Commission model" }, "confounders_named": [ "2014-2016 NHIS vintage", "region (rural/white states report more hearing loss)" ], "note": "Baseline S2 found a sign flip against the modelled outcome; re-tested against measured claims prevalence.", "fdr_q_within_level": 0.139, "classification": "null", "verdict": "Raw weak negative correlation r=-0.26 (95% CI -0.50 to 0.02), Spearman rho=-0.23, n=51 states (incl. DC), p=0.0696, BH q=0.14. NULL: raw association not distinguishable from zero (BH q=0.14 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Baseline S2 found a sign flip against the modelled outcome; re-tested against measured claims prevalence. Provenance: pct_hearing_loss is survey estimate (self-report); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-12", "level": "state", "kind": "exposure", "x": { "name": "pct_depression", "unit": "percent", "year": "2024", "source": "America's Health Rankings, Depression (CDC BRFSS) via data/panel_state.csv", "tag": "survey estimate / administrative compilation" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.0709, "p": 0.621, "ci95": [ -0.3398, 0.2088 ] }, "spearman": { "rho": -0.0318, "p": 0.825 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": -0.2394, "p": 0.0907, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": -0.2394, "p": 0.0907, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": -0.0111, "p": 0.938, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "depression is a risk factor and prodrome" }, "confounders_named": [ "healthcare contact", "region" ], "note": "", "fdr_q_within_level": 0.774, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.07 (95% CI -0.34 to 0.21), Spearman rho=-0.03, n=51 states (incl. DC), p=0.621, BH q=0.77. NULL: raw association not distinguishable from zero (BH q=0.77 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_depression is survey estimate / administrative compilation; norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-13", "level": "state", "kind": "exposure", "x": { "name": "pct_uninsured", "unit": "percent", "year": "2024", "source": "KFF State Health Facts, Health Insurance Coverage of the Total Population (ACS-based) via data/panel_state.csv", "tag": "survey estimate (BRFSS direct)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.1152, "p": 0.421, "ci95": [ -0.1657, 0.3787 ] }, "spearman": { "rho": 0.0523, "p": 0.716 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": -0.072, "p": 0.615, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": -0.072, "p": 0.615, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": -0.0708, "p": 0.622, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region (Medicaid non-expansion)", "poverty" ], "note": "", "fdr_q_within_level": 0.542, "classification": "null", "verdict": "Raw weak positive correlation r=0.12 (95% CI -0.17 to 0.38), Spearman rho=0.05, n=51 states (incl. DC), p=0.421, BH q=0.54. NULL: raw association not distinguishable from zero (BH q=0.54 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_uninsured is survey estimate (BRFSS direct); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-14", "level": "state", "kind": "exposure", "x": { "name": "primary_care_physicians_per_100k", "unit": "rate per 100k", "year": "2025", "source": "America's Health Rankings, Primary Care Providers (CMS National Plan and Provider Enumeration System) via data/panel_state.csv", "tag": "survey estimate / administrative compilation" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.0647, "p": 0.652, "ci95": [ -0.3343, 0.2147 ] }, "spearman": { "rho": -0.1192, "p": 0.405 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus" ], "r": 0.0304, "p": 0.833, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": -0.0497, "p": 0.729, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": -0.112, "p": 0.434, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "urbanicity", "diagnosis intensity" ], "note": "", "fdr_q_within_level": 0.787, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.06 (95% CI -0.33 to 0.21), Spearman rho=-0.12, n=51 states (incl. DC), p=0.652, BH q=0.79. NULL: raw association not distinguishable from zero (BH q=0.79 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: primary_care_physicians_per_100k is survey estimate / administrative compilation; norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-15", "level": "state", "kind": "exposure", "x": { "name": "neurologists_per_100k", "unit": "rate per 100k", "year": "2024", "source": "American Academy of Neurology, 2025 Member Demographics Report, Table 10b via data/panel_state.csv", "tag": "measured/administrative" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.1317, "p": 0.357, "ci95": [ -0.1493, 0.393 ] }, "spearman": { "rho": 0.105, "p": 0.463 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus" ], "r": 0.3873, "p": 0.00498, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.3025, "p": 0.031, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.1127, "p": 0.431, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "urbanicity", "diagnosis intensity" ], "note": "", "fdr_q_within_level": 0.487, "classification": "null", "verdict": "Raw weak positive correlation r=0.13 (95% CI -0.15 to 0.39), Spearman rho=0.10, n=51 states (incl. DC), p=0.357, BH q=0.49. NULL: raw association not distinguishable from zero (BH q=0.49 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: neurologists_per_100k is measured/administrative; norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-16", "level": "state", "kind": "exposure", "x": { "name": "pct_65plus", "unit": "percent", "year": "2023", "source": "U.S. Census Bureau, ACS 2023 1-Year Estimates, Geographic Comparison Table GCT0103 via data/panel_state.csv", "tag": "survey estimate (ACS 1-year)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.1227, "p": 0.391, "ci95": [ -0.3852, 0.1583 ] }, "spearman": { "rho": -0.1205, "p": 0.4 }, "partial_primary": { "controls": [ "median_household_income" ], "r": -0.2027, "p": 0.154, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": -0.3819, "p": 0.00568, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": -0.325, "p": 0.02, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "retirement migration" ], "note": "", "fdr_q_within_level": 0.521, "classification": "null", "verdict": "Raw weak negative correlation r=-0.12 (95% CI -0.39 to 0.16), Spearman rho=-0.12, n=51 states (incl. DC), p=0.391, BH q=0.52. NULL: raw association not distinguishable from zero (BH q=0.52 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_65plus is survey estimate (ACS 1-year); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-17", "level": "state", "kind": "exposure", "x": { "name": "pct_85plus", "unit": "percent", "year": "2023", "source": "U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT0104 via data/panel_state.csv", "tag": "survey estimate (ACS 1-year)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.0244, "p": 0.865, "ci95": [ -0.298, 0.2529 ] }, "spearman": { "rho": -0.1073, "p": 0.454 }, "partial_primary": { "controls": [ "median_household_income" ], "r": -0.0332, "p": 0.817, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "region_south" ], "r": 0.182, "p": 0.201, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.2856, "p": 0.0422, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "prevalence rises steeply with age" }, "confounders_named": [ "retirement migration", "institutionalisation" ], "note": "", "fdr_q_within_level": 0.939, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.02 (95% CI -0.30 to 0.25), Spearman rho=-0.11, n=51 states (incl. DC), p=0.865, BH q=0.94. NULL: raw association not distinguishable from zero (BH q=0.94 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_85plus is survey estimate (ACS 1-year); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-18", "level": "state", "kind": "exposure", "x": { "name": "pm25_avg_ugm3", "unit": "ug/m3", "year": "2023", "source": "America's Health Rankings, Air Pollution (source: EPA), 2022-2024 average via data/panel_state.csv", "tag": "survey estimate / administrative compilation" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.423, "p": 0.00198, "ci95": [ 0.1669, 0.6257 ] }, "spearman": { "rho": 0.3919, "p": 0.00446 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.3993, "p": 0.0037, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.3993, "p": 0.0037, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.0978, "p": 0.495, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "PM2.5 is a risk factor" }, "confounders_named": [ "urban density", "region" ], "note": "", "fdr_q_within_level": 0.00537, "classification": "collapses_strict", "verdict": "Raw moderate positive correlation r=0.42 (95% CI 0.17 to 0.63), Spearman rho=0.39, n=51 states (incl. DC), p=0.00198, BH q=0.0054. COLLAPSES under place / diagnosis-intensity controls: survives age/income/region controls (extended partial r=0.3993) but falls to 0.098 once place and diagnosis-intensity markers are added (median_household_income+pct_85plus+region_south+medicare_percapita_ffs_payment_usd+pct_race_black). What looked like an exposure effect is carried by where the county is and how intensively it codes diagnoses. Provenance: pm25_avg_ugm3 is survey estimate / administrative compilation; norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-19", "level": "state", "kind": "exposure", "x": { "name": "gini", "unit": "index (0-1)", "year": "2019", "source": "Wikipedia 'List of U.S. states and territories by income inequality', citing Census ACS table B19083 via data/panel_state.csv", "tag": "survey estimate (ACS)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.6858, "p": 2.8e-08, "ci95": [ 0.5058, 0.8086 ] }, "spearman": { "rho": 0.7033, "p": 8.67e-09 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.6055, "p": 2.5e-06, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.6055, "p": 2.5e-06, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.2908, "p": 0.0384, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "income", "race composition" ], "note": "", "fdr_q_within_level": 1.93e-07, "classification": "collapses_strict", "verdict": "Raw strong positive correlation r=0.69 (95% CI 0.51 to 0.81), Spearman rho=0.70, n=51 states (incl. DC), p=2.8e-08, BH q=1.9e-07. COLLAPSES under place / diagnosis-intensity controls: survives age/income/region controls (extended partial r=0.6055) but falls to 0.291 once place and diagnosis-intensity markers are added (median_household_income+pct_85plus+region_south+medicare_percapita_ffs_payment_usd+pct_race_black). What looked like an exposure effect is carried by where the county is and how intensively it codes diagnoses. Provenance: gini is survey estimate (ACS); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-20", "level": "state", "kind": "exposure", "x": { "name": "medicare_percapita_ffs_payment_usd", "unit": "usd per capita", "year": "2022", "source": "Alzheimer's Association 2025 Facts and Figures, Table 19, citing CMS via data/panel_state.csv", "tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.4274, "p": 0.00176, "ci95": [ 0.172, 0.6289 ] }, "spearman": { "rho": 0.5251, "p": 7.63e-05 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus" ], "r": 0.5622, "p": 1.76e-05, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.5059, "p": 0.000152, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "pct_race_black" ], "r": 0.3915, "p": 0.0045, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "claims intensity", "prices" ], "note": "Per-capita Medicare FFS payment: a claims-intensity marker, so a positive association points at detection.", "fdr_q_within_level": 0.00495, "classification": "survives", "verdict": "Raw moderate positive correlation r=0.43 (95% CI 0.17 to 0.63), Spearman rho=0.53, n=51 states (incl. DC), p=0.00176, BH q=0.005. SURVIVES the confounder test: partial r=0.562 (p=1.76e-05) controlling for median_household_income+pct_85plus; extended partial r=0.5059; strict partial r=0.3915 controlling for median_household_income+pct_85plus+region_south+pct_race_black. A place-level association not explained by the named control(s); candidate third things still unaddressed: claims intensity, prices. Per-capita Medicare FFS payment: a claims-intensity marker, so a positive association points at detection. Provenance: medicare_percapita_ffs_payment_usd is modelled (Alzheimer's Association projection built from age/sex/race distributions); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-21", "level": "state", "kind": "exposure", "x": { "name": "nursing_home_beds_per_1000", "unit": "rate per 1000 (65+ population)", "year": "2026", "source": "Becker's Hospital Review, States ranked by nursing home beds per capita (May 2026), using Census population + KFF CNF bed data via data/panel_state.csv", "tag": "measured/administrative" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.4771, "p": 0.0004, "ci95": [ 0.232, 0.6652 ] }, "spearman": { "rho": 0.495, "p": 0.000222 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus" ], "r": 0.4401, "p": 0.00123, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.4798, "p": 0.000367, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.4127, "p": 0.00262, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "institutional diagnosis (nursing-home residents are screened)", "region" ], "note": "", "fdr_q_within_level": 0.00132, "classification": "survives", "verdict": "Raw moderate positive correlation r=0.48 (95% CI 0.23 to 0.67), Spearman rho=0.49, n=51 states (incl. DC), p=0.0004, BH q=0.0013. SURVIVES the confounder test: partial r=0.440 (p=0.00123) controlling for median_household_income+pct_85plus; extended partial r=0.4798; strict partial r=0.4127 controlling for median_household_income+pct_85plus+region_south+medicare_percapita_ffs_payment_usd+pct_race_black. A place-level association not explained by the named control(s); candidate third things still unaddressed: institutional diagnosis (nursing-home residents are screened), region. Provenance: nursing_home_beds_per_1000 is measured/administrative; norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-22", "level": "state", "kind": "exposure", "x": { "name": "pct_physical_inactivity", "unit": "percent", "year": "2024", "source": "Derived from KFF State Health Facts, Participation in Physical Activity (CDC BRFSS) via data/panel_state.csv", "tag": "survey estimate (BRFSS direct)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 50, "n_unit": "states (incl. DC)", "x_filled": 50, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.5906, "p": 6.33e-06, "ci95": [ 0.3737, 0.7463 ] }, "spearman": { "rho": 0.5962, "p": 4.9e-06 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.4856, "p": 0.00035, "n": 50 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.4856, "p": 0.00035, "n": 50 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.3913, "p": 0.00496, "n": 50 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "inactivity raises risk" }, "confounders_named": [ "region", "obesity" ], "note": "", "fdr_q_within_level": 3.44e-05, "classification": "survives", "verdict": "Raw strong positive correlation r=0.59 (95% CI 0.37 to 0.75), Spearman rho=0.60, n=50 states (incl. DC), p=6.33e-06, BH q=3.4e-05. SURVIVES the confounder test: partial r=0.486 (p=0.00035) controlling for median_household_income+pct_85plus+region_south; extended partial r=0.4856; strict partial r=0.3913 controlling for median_household_income+pct_85plus+region_south+medicare_percapita_ffs_payment_usd+pct_race_black. A place-level association not explained by the named control(s); candidate third things still unaddressed: region, obesity. Provenance: pct_physical_inactivity is survey estimate (BRFSS direct); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SM-1", "level": "state", "kind": "exposure", "x": { "name": "pct_bachelors_plus", "unit": "percent", "year": "2023", "source": "U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT1502 via data/panel_state.csv", "tag": "survey estimate (ACS 1-year)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.5598, "p": 1.95e-05, "ci95": [ -0.7237, -0.336 ] }, "spearman": { "rho": -0.4425, "p": 0.00115 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": -0.2323, "p": 0.101, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": -0.2489, "p": 0.0782, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": -0.2014, "p": 0.157, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "negative", "why": "education is protective" }, "confounders_named": [ "income", "coding practice" ], "note": "Baseline S4 (unadjusted rate, r=-0.58) re-run with the AGE-ADJUSTED rate and coding share controlled.", "fdr_q_within_level": 9.78e-05, "classification": "collapses", "verdict": "Raw strong negative correlation r=-0.56 (95% CI -0.72 to -0.34), Spearman rho=-0.44, n=51 states (incl. DC), p=1.95e-05, BH q=9.8e-05. COLLAPSES: partial r=-0.232 is under half the raw r=-0.560 once median_household_income+ts_dementia_deaths_pct_g30_2017 is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_bachelors_plus. Baseline S4 (unadjusted rate, r=-0.58) re-run with the AGE-ADJUSTED rate and coding share controlled. Provenance: pct_bachelors_plus is survey estimate (ACS 1-year); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-2", "level": "state", "kind": "exposure", "x": { "name": "pct_rural", "unit": "percent", "year": "2023", "source": "America's Health Rankings, Rural Population (Census ACS 1-Year) via data/panel_state.csv", "tag": "survey estimate (ACS)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.3986, "p": 0.00377, "ci95": [ 0.1382, 0.6074 ] }, "spearman": { "rho": 0.3667, "p": 0.00813 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": 0.225, "p": 0.112, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": 0.2288, "p": 0.106, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": 0.1912, "p": 0.179, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "income", "coding practice" ], "note": "Baseline S9 (unadjusted rate, r=0.54) re-run age-adjusted.", "fdr_q_within_level": 0.00924, "classification": "attenuated", "verdict": "Raw moderate positive correlation r=0.40 (95% CI 0.14 to 0.61), Spearman rho=0.37, n=51 states (incl. DC), p=0.00377, BH q=0.0092. ATTENUATED: partial r=0.225 keeps >=50% of raw but is no longer significant (p=0.112). Weak evidence at best. Baseline S9 (unadjusted rate, r=0.54) re-run age-adjusted. Provenance: pct_rural is survey estimate (ACS); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-3", "level": "state", "kind": "exposure", "x": { "name": "pm25_avg_ugm3", "unit": "ug/m3", "year": "2023", "source": "America's Health Rankings, Air Pollution (source: EPA), 2022-2024 average via data/panel_state.csv", "tag": "survey estimate / administrative compilation" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.0667, "p": 0.642, "ci95": [ -0.2128, 0.3361 ] }, "spearman": { "rho": 0.1033, "p": 0.471 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": -0.0837, "p": 0.559, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": -0.0906, "p": 0.527, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": -0.0243, "p": 0.866, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "PM2.5 is a risk factor" }, "confounders_named": [ "region", "coding practice" ], "note": "Baseline S1 re-run age-adjusted.", "fdr_q_within_level": 0.787, "classification": "null", "verdict": "Raw negligible positive correlation r=0.07 (95% CI -0.21 to 0.34), Spearman rho=0.10, n=51 states (incl. DC), p=0.642, BH q=0.79. NULL: raw association not distinguishable from zero (BH q=0.79 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Baseline S1 re-run age-adjusted. Provenance: pm25_avg_ugm3 is survey estimate / administrative compilation; ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-4", "level": "state", "kind": "exposure", "x": { "name": "median_household_income", "unit": "usd", "year": "2023", "source": "U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT1901 via data/panel_state.csv", "tag": "survey estimate (ACS 1-year)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.4778, "p": 0.000392, "ci95": [ -0.6657, -0.2329 ] }, "spearman": { "rho": -0.3883, "p": 0.00486 }, "partial_primary": { "controls": [ "ts_dementia_deaths_pct_g30_2017" ], "r": -0.3447, "p": 0.0132, "n": 51 }, "partial_extended": { "controls": [ "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": -0.3227, "p": 0.0209, "n": 51 }, "partial_strict": { "controls": [ "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": -0.2792, "p": 0.0472, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "negative", "why": "higher SES is protective" }, "confounders_named": [ "coding practice", "region" ], "note": "", "fdr_q_within_level": 0.00132, "classification": "survives", "verdict": "Raw moderate negative correlation r=-0.48 (95% CI -0.67 to -0.23), Spearman rho=-0.39, n=51 states (incl. DC), p=0.000392, BH q=0.0013. SURVIVES the confounder test: partial r=-0.345 (p=0.0132) controlling for ts_dementia_deaths_pct_g30_2017; extended partial r=-0.3227; strict partial r=-0.2792 controlling for ts_dementia_deaths_pct_g30_2017+region_south+pct_race_black. A place-level association not explained by the named control(s); candidate third things still unaddressed: coding practice, region. Provenance: median_household_income is survey estimate (ACS 1-year); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-5", "level": "state", "kind": "exposure", "x": { "name": "pct_race_black", "unit": "percent", "year": "2023", "source": "U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT0202 via data/panel_state.csv", "tag": "survey estimate (ACS 1-year)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.0032, "p": 0.982, "ci95": [ -0.2727, 0.2785 ] }, "spearman": { "rho": 0.0138, "p": 0.923 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": -0.0909, "p": 0.526, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": -0.1722, "p": 0.227, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": -0.1722, "p": 0.227, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region", "coding practice" ], "note": "Baseline B1 re-run age-adjusted.", "fdr_q_within_level": 0.982, "classification": "null", "verdict": "Raw negligible positive correlation r=0.00 (95% CI -0.27 to 0.28), Spearman rho=0.01, n=51 states (incl. DC), p=0.982, BH q=0.98. NULL: raw association not distinguishable from zero (BH q=0.98 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Baseline B1 re-run age-adjusted. Provenance: pct_race_black is survey estimate (ACS 1-year); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-6", "level": "state", "kind": "exposure", "x": { "name": "pct_race_hispanic", "unit": "percent", "year": "2024", "source": "KFF State Health Facts, Distribution by Race/Ethnicity (ACS 1-Year Estimates) via data/panel_state.csv", "tag": "survey estimate (BRFSS direct)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.1444, "p": 0.312, "ci95": [ -0.404, 0.1366 ] }, "spearman": { "rho": -0.1814, "p": 0.203 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": -0.1814, "p": 0.203, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": -0.1802, "p": 0.206, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": -0.1993, "p": 0.161, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "region", "coding practice" ], "note": "", "fdr_q_within_level": 0.447, "classification": "null", "verdict": "Raw weak negative correlation r=-0.14 (95% CI -0.40 to 0.14), Spearman rho=-0.18, n=51 states (incl. DC), p=0.312, BH q=0.45. NULL: raw association not distinguishable from zero (BH q=0.45 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_race_hispanic is survey estimate (BRFSS direct); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-7", "level": "state", "kind": "exposure", "x": { "name": "gini", "unit": "index (0-1)", "year": "2019", "source": "Wikipedia 'List of U.S. states and territories by income inequality', citing Census ACS table B19083 via data/panel_state.csv", "tag": "survey estimate (ACS)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.2346, "p": 0.0976, "ci95": [ -0.4792, 0.0439 ] }, "spearman": { "rho": -0.1439, "p": 0.314 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": -0.3295, "p": 0.0182, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": -0.3847, "p": 0.00531, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": -0.351, "p": 0.0116, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "coding practice" ], "note": "Baseline B3 re-run age-adjusted.", "fdr_q_within_level": 0.177, "classification": "null", "verdict": "Raw weak negative correlation r=-0.23 (95% CI -0.48 to 0.04), Spearman rho=-0.14, n=51 states (incl. DC), p=0.0976, BH q=0.18. NULL: raw association not distinguishable from zero (BH q=0.18 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Baseline B3 re-run age-adjusted. Provenance: gini is survey estimate (ACS); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-8", "level": "state", "kind": "exposure", "x": { "name": "primary_care_physicians_per_100k", "unit": "rate per 100k", "year": "2025", "source": "America's Health Rankings, Primary Care Providers (CMS National Plan and Provider Enumeration System) via data/panel_state.csv", "tag": "survey estimate / administrative compilation" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.5047, "p": 0.000159, "ci95": [ -0.685, -0.2662 ] }, "spearman": { "rho": -0.4283, "p": 0.00172 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": -0.2024, "p": 0.154, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": -0.2058, "p": 0.147, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": -0.179, "p": 0.209, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "urbanicity", "coding practice" ], "note": "Baseline B5 re-run age-adjusted.", "fdr_q_within_level": 0.000604, "classification": "collapses", "verdict": "Raw strong negative correlation r=-0.50 (95% CI -0.69 to -0.27), Spearman rho=-0.43, n=51 states (incl. DC), p=0.000159, BH q=0.0006. COLLAPSES: partial r=-0.202 is under half the raw r=-0.505 once median_household_income+ts_dementia_deaths_pct_g30_2017 is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with primary_care_physicians_per_100k. Baseline B5 re-run age-adjusted. Provenance: primary_care_physicians_per_100k is survey estimate / administrative compilation; ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-9", "level": "state", "kind": "exposure", "x": { "name": "geriatricians_count", "unit": "count", "year": "2021", "source": "Alzheimer's Association 2025 Facts and Figures, Table 14, citing American Geriatrics Society via data/panel_state.csv", "tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.2685, "p": 0.0567, "ci95": [ -0.5066, 0.0076 ] }, "spearman": { "rho": -0.2173, "p": 0.126 }, "partial_primary": { "controls": [ "pop_total", "ts_dementia_deaths_pct_g30_2017" ], "r": -0.3559, "p": 0.0104, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": -0.2344, "p": 0.0978, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": -0.2169, "p": 0.126, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "population size (raw count)", "coding practice" ], "note": "Baseline S6 re-run age-adjusted, population controlled.", "fdr_q_within_level": 0.116, "classification": "null", "verdict": "Raw weak negative correlation r=-0.27 (95% CI -0.51 to 0.01), Spearman rho=-0.22, n=51 states (incl. DC), p=0.0567, BH q=0.12. NULL: raw association not distinguishable from zero (BH q=0.12 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Baseline S6 re-run age-adjusted, population controlled. Provenance: geriatricians_count is modelled (Alzheimer's Association projection built from age/sex/race distributions); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-10", "level": "state", "kind": "exposure", "x": { "name": "pct_diabetes", "unit": "percent", "year": "2024", "source": "KFF State Health Facts, Adults with Diagnosed Diabetes (CDC BRFSS) via data/panel_state.csv", "tag": "survey estimate (BRFSS direct)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 50, "n_unit": "states (incl. DC)", "x_filled": 50, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.2349, "p": 0.101, "ci95": [ -0.0464, 0.4818 ] }, "spearman": { "rho": 0.1977, "p": 0.169 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": -0.0571, "p": 0.694, "n": 50 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": -0.0837, "p": 0.563, "n": 50 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": -0.069, "p": 0.634, "n": 50 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "diabetes raises risk" }, "confounders_named": [ "region", "coding practice" ], "note": "", "fdr_q_within_level": 0.179, "classification": "null", "verdict": "Raw weak positive correlation r=0.23 (95% CI -0.05 to 0.48), Spearman rho=0.20, n=50 states (incl. DC), p=0.101, BH q=0.18. NULL: raw association not distinguishable from zero (BH q=0.18 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_diabetes is survey estimate (BRFSS direct); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-11", "level": "state", "kind": "exposure", "x": { "name": "pct_obesity", "unit": "percent", "year": "2024", "source": "KFF State Health Facts, Adult Obesity (CDC BRFSS) via data/panel_state.csv", "tag": "survey estimate (BRFSS direct)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 50, "n_unit": "states (incl. DC)", "x_filled": 50, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.5035, "p": 0.000193, "ci95": [ 0.2619, 0.6858 ] }, "spearman": { "rho": 0.4221, "p": 0.00226 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": 0.2739, "p": 0.0543, "n": 50 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": 0.2727, "p": 0.0554, "n": 50 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": 0.297, "p": 0.0362, "n": 50 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "obesity raises risk" }, "confounders_named": [ "region", "coding practice" ], "note": "", "fdr_q_within_level": 0.000698, "classification": "attenuated", "verdict": "Raw strong positive correlation r=0.50 (95% CI 0.26 to 0.69), Spearman rho=0.42, n=50 states (incl. DC), p=0.000193, BH q=0.0007. ATTENUATED: partial r=0.274 keeps >=50% of raw but is no longer significant (p=0.0543). Weak evidence at best. Provenance: pct_obesity is survey estimate (BRFSS direct); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-12", "level": "state", "kind": "exposure", "x": { "name": "pct_smoking", "unit": "percent", "year": "2024", "source": "KFF State Health Facts, Adult Smoking Rate (CDC BRFSS) via data/panel_state.csv", "tag": "survey estimate (BRFSS direct)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 50, "n_unit": "states (incl. DC)", "x_filled": 50, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.2562, "p": 0.0726, "ci95": [ -0.0239, 0.4989 ] }, "spearman": { "rho": 0.2729, "p": 0.0552 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": -0.0775, "p": 0.592, "n": 50 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": -0.0797, "p": 0.582, "n": 50 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": -0.1115, "p": 0.441, "n": 50 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "smoking raises risk" }, "confounders_named": [ "region", "coding practice", "competing mortality" ], "note": "", "fdr_q_within_level": 0.141, "classification": "null", "verdict": "Raw weak positive correlation r=0.26 (95% CI -0.02 to 0.50), Spearman rho=0.27, n=50 states (incl. DC), p=0.0726, BH q=0.14. NULL: raw association not distinguishable from zero (BH q=0.14 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_smoking is survey estimate (BRFSS direct); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-13", "level": "state", "kind": "exposure", "x": { "name": "pct_hypertension", "unit": "percent", "year": "2023", "source": "KFF State Health Facts, Adults Told They Have Hypertension (CDC BRFSS) via data/panel_state.csv", "tag": "survey estimate (BRFSS direct)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 49, "n_unit": "states (incl. DC)", "x_filled": 49, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.3618, "p": 0.0106, "ci95": [ 0.0897, 0.5836 ] }, "spearman": { "rho": 0.2217, "p": 0.126 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": 0.0678, "p": 0.644, "n": 49 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": 0.0638, "p": 0.663, "n": 49 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": 0.1148, "p": 0.432, "n": 49 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "hypertension raises risk" }, "confounders_named": [ "region", "coding practice" ], "note": "", "fdr_q_within_level": 0.0244, "classification": "collapses", "verdict": "Raw moderate positive correlation r=0.36 (95% CI 0.09 to 0.58), Spearman rho=0.22, n=49 states (incl. DC), p=0.0106, BH q=0.024. COLLAPSES: partial r=0.068 is under half the raw r=0.362 once median_household_income+ts_dementia_deaths_pct_g30_2017 is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_hypertension. Provenance: pct_hypertension is survey estimate (BRFSS direct); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-14", "level": "state", "kind": "exposure", "x": { "name": "pct_hearing_loss", "unit": "percent", "year": "2016", "source": "CDC/NCHS, Any Hearing Loss by State, National Health Interview Survey 2014-2016 (Blackwell & Norris) via data/panel_state.csv", "tag": "survey estimate (self-report)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.4325, "p": 0.00153, "ci95": [ 0.1782, 0.6327 ] }, "spearman": { "rho": 0.3627, "p": 0.00891 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": 0.3349, "p": 0.0163, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": 0.3815, "p": 0.00575, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": 0.3471, "p": 0.0126, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "hearing loss raises risk" }, "confounders_named": [ "region", "coding practice" ], "note": "", "fdr_q_within_level": 0.00447, "classification": "survives", "verdict": "Raw moderate positive correlation r=0.43 (95% CI 0.18 to 0.63), Spearman rho=0.36, n=51 states (incl. DC), p=0.00153, BH q=0.0045. SURVIVES the confounder test: partial r=0.335 (p=0.0163) controlling for median_household_income+ts_dementia_deaths_pct_g30_2017; extended partial r=0.3815; strict partial r=0.3471 controlling for median_household_income+ts_dementia_deaths_pct_g30_2017+region_south+pct_race_black. A place-level association not explained by the named control(s); candidate third things still unaddressed: region, coding practice. Provenance: pct_hearing_loss is survey estimate (self-report); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-15", "level": "state", "kind": "exposure", "x": { "name": "nursing_home_beds_per_1000", "unit": "rate per 1000 (65+ population)", "year": "2026", "source": "Becker's Hospital Review, States ranked by nursing home beds per capita (May 2026), using Census population + KFF CNF bed data via data/panel_state.csv", "tag": "measured/administrative" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.0071, "p": 0.96, "ci95": [ -0.269, 0.2822 ] }, "spearman": { "rho": -0.0056, "p": 0.969 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": -0.1234, "p": 0.388, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": -0.1234, "p": 0.388, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": -0.0837, "p": 0.559, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "place of death (nursing-home deaths get dementia coded)", "coding practice" ], "note": "Baseline B6 re-run age-adjusted.", "fdr_q_within_level": 0.982, "classification": "null", "verdict": "Raw negligible positive correlation r=0.01 (95% CI -0.27 to 0.28), Spearman rho=-0.01, n=51 states (incl. DC), p=0.96, BH q=0.98. NULL: raw association not distinguishable from zero (BH q=0.98 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Baseline B6 re-run age-adjusted. Provenance: nursing_home_beds_per_1000 is measured/administrative; ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-16", "level": "state", "kind": "exposure", "x": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.0266, "p": 0.853, "ci95": [ -0.3, 0.2508 ] }, "spearman": { "rho": -0.0444, "p": 0.757 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": -0.1099, "p": 0.443, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": -0.1483, "p": 0.299, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": -0.0711, "p": 0.62, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "diagnosis intensity", "coding practice" ], "note": "", "fdr_q_within_level": 0.939, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.03 (95% CI -0.30 to 0.25), Spearman rho=-0.04, n=51 states (incl. DC), p=0.853, BH q=0.94. NULL: raw association not distinguishable from zero (BH q=0.94 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SM-17", "level": "state", "kind": "exposure", "x": { "name": "pct_65plus", "unit": "percent", "year": "2023", "source": "U.S. Census Bureau, ACS 2023 1-Year Estimates, Geographic Comparison Table GCT0103 via data/panel_state.csv", "tag": "survey estimate (ACS 1-year)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.2152, "p": 0.129, "ci95": [ -0.4633, 0.0641 ] }, "spearman": { "rho": -0.3092, "p": 0.0273 }, "partial_primary": { "controls": [ "ts_dementia_deaths_pct_g30_2017" ], "r": -0.0842, "p": 0.557, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": -0.2325, "p": 0.101, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": -0.2684, "p": 0.0569, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "retirement migration" ], "note": "Age-adjusted rate should be independent of age structure; a residual association points at something else (e.g. retirement-destination states).", "fdr_q_within_level": 0.218, "classification": "null", "verdict": "Raw weak negative correlation r=-0.22 (95% CI -0.46 to 0.06), Spearman rho=-0.31, n=51 states (incl. DC), p=0.129, BH q=0.22. NULL: raw association not distinguishable from zero (BH q=0.22 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Age-adjusted rate should be independent of age structure; a residual association points at something else (e.g. retirement-destination states). Provenance: pct_65plus is survey estimate (ACS 1-year); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SM-18", "level": "state", "kind": "exposure", "x": { "name": "region_south", "unit": "1 = state in US Census South region (Divisions 5-7), else 0", "year": "static", "source": "US Census Bureau region definition", "tag": "classification (not a measurement)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.1849, "p": 0.194, "ci95": [ -0.0956, 0.4381 ] }, "spearman": { "rho": 0.2247, "p": 0.113 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": 0.0344, "p": 0.811, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": 0.0344, "p": 0.811, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "pct_race_black" ], "r": 0.1508, "p": 0.291, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "coding practice", "stroke-belt risk" ], "note": "", "fdr_q_within_level": 0.321, "classification": "null", "verdict": "Raw weak positive correlation r=0.18 (95% CI -0.10 to 0.44), Spearman rho=0.22, n=51 states (incl. DC), p=0.194, BH q=0.32. NULL: raw association not distinguishable from zero (BH q=0.32 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: region_south is classification (not a measurement); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "ST-1", "level": "state", "kind": "artefact_test", "x": { "name": "ts_age_adj_rate_2010", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted", "year": "2010", "source": "CDC WONDER via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "y": { "name": "ts_age_adj_rate_slope_2010_2019", "unit": "OLS slope of the age-adjusted G30 death rate, per 100,000 per year, over 2010-2019 (all 10 years required)", "year": "2010-2019", "source": "computed from data/panel_state_timeseries.csv", "tag": "measured/administrative (derived trend)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.0208, "p": 0.885, "ci95": [ -0.2947, 0.2562 ] }, "spearman": { "rho": 0.0152, "p": 0.916 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "regression to the mean", "coding convergence" ], "note": "Did states with high 2010 rates rise more slowly (convergence)? A negative r is the classic regression-to-the-mean pattern in coding-driven series.", "fdr_q_within_level": 0.947, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.02 (95% CI -0.29 to 0.26), Spearman rho=0.02, n=51 states (incl. DC), p=0.885, BH q=0.95. NULL: not distinguishable from zero after FDR. Absence of a place-level correlation is not evidence of no individual-level effect. Did states with high 2010 rates rise more slowly (convergence)? A negative r is the classic regression-to-the-mean pattern in coding-driven series. Provenance: ts_age_adj_rate_2010 is measured/administrative (vital registration); ts_age_adj_rate_slope_2010_2019 is measured/administrative (derived trend). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "ST-2", "level": "state", "kind": "artefact_test", "x": { "name": "ts_dementia_deaths_pct_g30_2017", "unit": "percent of all dementia-coded deaths (G30+F01+F03+G31) whose underlying cause was coded G30 Alzheimer's", "year": "2017", "source": "CDC WONDER via data/panel_state_timeseries.csv", "tag": "measured/administrative (coding share)" }, "y": { "name": "ts_age_adj_rate_slope_2010_2019", "unit": "OLS slope of the age-adjusted G30 death rate, per 100,000 per year, over 2010-2019 (all 10 years required)", "year": "2010-2019", "source": "computed from data/panel_state_timeseries.csv", "tag": "measured/administrative (derived trend)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.4557, "p": 0.000779, "ci95": [ 0.2059, 0.6497 ] }, "spearman": { "rho": 0.4973, "p": 0.000205 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "coding practice" ], "note": "States coding more G30 in 2017 vs their 2010-2019 trend.", "fdr_q_within_level": 0.00247, "classification": "artefact_documented", "verdict": "Raw moderate positive correlation r=0.46 (95% CI 0.21 to 0.65), Spearman rho=0.50, n=51 states (incl. DC), p=0.000779, BH q=0.0025. STRUCTURAL DEPENDENCE DOCUMENTED: r=0.456, q=0.0025: documents the mechanical/structural dependence named. States coding more G30 in 2017 vs their 2010-2019 trend. Provenance: ts_dementia_deaths_pct_g30_2017 is measured/administrative (coding share); ts_age_adj_rate_slope_2010_2019 is measured/administrative (derived trend). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "ST-3", "level": "state", "kind": "exposure", "x": { "name": "pct_bachelors_plus", "unit": "percent", "year": "2023", "source": "U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT1502 via data/panel_state.csv", "tag": "survey estimate (ACS 1-year)" }, "y": { "name": "ts_age_adj_rate_slope_2010_2019", "unit": "OLS slope of the age-adjusted G30 death rate, per 100,000 per year, over 2010-2019 (all 10 years required)", "year": "2010-2019", "source": "computed from data/panel_state_timeseries.csv", "tag": "measured/administrative (derived trend)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.4178, "p": 0.00228, "ci95": [ -0.6218, -0.1607 ] }, "spearman": { "rho": -0.354, "p": 0.0108 }, "partial_primary": { "controls": [ "median_household_income" ], "r": -0.2872, "p": 0.041, "n": 51 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "income", "coding shifts" ], "note": "", "fdr_q_within_level": 0.00598, "classification": "survives", "verdict": "Raw moderate negative correlation r=-0.42 (95% CI -0.62 to -0.16), Spearman rho=-0.35, n=51 states (incl. DC), p=0.00228, BH q=0.006. SURVIVES the confounder test: partial r=-0.287 (p=0.041) controlling for median_household_income. A place-level association not explained by the named control(s); candidate third things still unaddressed: income, coding shifts. Provenance: pct_bachelors_plus is survey estimate (ACS 1-year); ts_age_adj_rate_slope_2010_2019 is measured/administrative (derived trend). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "ST-4", "level": "state", "kind": "exposure", "x": { "name": "pct_race_black", "unit": "percent", "year": "2023", "source": "U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT0202 via data/panel_state.csv", "tag": "survey estimate (ACS 1-year)" }, "y": { "name": "ts_age_adj_rate_change_2019_2020_pct", "unit": "percent change in the age-adjusted G30 death rate from 2019 to 2020 (first COVID year)", "year": "2019-2020", "source": "computed from data/panel_state_timeseries.csv", "tag": "measured/administrative (derived change)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.2995, "p": 0.0327, "ci95": [ 0.0261, 0.5313 ] }, "spearman": { "rho": 0.3096, "p": 0.0271 }, "partial_primary": { "controls": [ "median_household_income" ], "r": 0.2949, "p": 0.0357, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "region_south" ], "r": 0.2526, "p": 0.0737, "n": 51 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "COVID-19 mortality gradient by race in 2020", "COVID deaths mis-attributed to dementia" ], "note": "First-COVID-year jump in the Alzheimer's death rate.", "fdr_q_within_level": 0.071, "classification": "null", "verdict": "Raw weak positive correlation r=0.30 (95% CI 0.03 to 0.53), Spearman rho=0.31, n=51 states (incl. DC), p=0.0327, BH q=0.071. NULL: raw association not distinguishable from zero (BH q=0.071 within level). Absence of a place-level correlation is not evidence of no individual-level effect. First-COVID-year jump in the Alzheimer's death rate. Provenance: pct_race_black is survey estimate (ACS 1-year); ts_age_adj_rate_change_2019_2020_pct is measured/administrative (derived change). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "ST-5", "level": "state", "kind": "exposure", "x": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "y": { "name": "ts_age_adj_rate_change_2019_2020_pct", "unit": "percent change in the age-adjusted G30 death rate from 2019 to 2020 (first COVID year)", "year": "2019-2020", "source": "computed from data/panel_state_timeseries.csv", "tag": "measured/administrative (derived change)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.2404, "p": 0.0892, "ci95": [ -0.0376, 0.484 ] }, "spearman": { "rho": 0.2996, "p": 0.0327 }, "partial_primary": { "controls": [ "median_household_income" ], "r": 0.2175, "p": 0.125, "n": 51 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "nursing-home COVID outbreaks", "diagnosis intensity" ], "note": "", "fdr_q_within_level": 0.169, "classification": "null", "verdict": "Raw weak positive correlation r=0.24 (95% CI -0.04 to 0.48), Spearman rho=0.30, n=51 states (incl. DC), p=0.0892, BH q=0.17. NULL: raw association not distinguishable from zero (BH q=0.17 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases); ts_age_adj_rate_change_2019_2020_pct is measured/administrative (derived change). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "ST-6", "level": "state", "kind": "exposure", "x": { "name": "nursing_home_beds_per_1000", "unit": "rate per 1000 (65+ population)", "year": "2026", "source": "Becker's Hospital Review, States ranked by nursing home beds per capita (May 2026), using Census population + KFF CNF bed data via data/panel_state.csv", "tag": "measured/administrative" }, "y": { "name": "ts_age_adj_rate_change_2019_2020_pct", "unit": "percent change in the age-adjusted G30 death rate from 2019 to 2020 (first COVID year)", "year": "2019-2020", "source": "computed from data/panel_state_timeseries.csv", "tag": "measured/administrative (derived change)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.0372, "p": 0.795, "ci95": [ -0.3096, 0.2408 ] }, "spearman": { "rho": 0.0595, "p": 0.678 }, "partial_primary": { "controls": [ "median_household_income" ], "r": -0.0815, "p": 0.57, "n": 51 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "nursing-home COVID outbreaks" ], "note": "", "fdr_q_within_level": 0.93, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.04 (95% CI -0.31 to 0.24), Spearman rho=0.06, n=51 states (incl. DC), p=0.795, BH q=0.93. NULL: raw association not distinguishable from zero (BH q=0.93 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: nursing_home_beds_per_1000 is measured/administrative; ts_age_adj_rate_change_2019_2020_pct is measured/administrative (derived change). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "ST-7", "level": "state", "kind": "exposure", "x": { "name": "pct_rural", "unit": "percent", "year": "2023", "source": "America's Health Rankings, Rural Population (Census ACS 1-Year) via data/panel_state.csv", "tag": "survey estimate (ACS)" }, "y": { "name": "ts_age_adj_rate_change_2019_2020_pct", "unit": "percent change in the age-adjusted G30 death rate from 2019 to 2020 (first COVID year)", "year": "2019-2020", "source": "computed from data/panel_state_timeseries.csv", "tag": "measured/administrative (derived change)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.3067, "p": 0.0286, "ci95": [ -0.5369, -0.034 ] }, "spearman": { "rho": -0.26, "p": 0.0654 }, "partial_primary": { "controls": [ "median_household_income" ], "r": -0.4803, "p": 0.000361, "n": 51 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "COVID wave timing in 2020 (urban first)" ], "note": "", "fdr_q_within_level": 0.0639, "classification": "null", "verdict": "Raw moderate negative correlation r=-0.31 (95% CI -0.54 to -0.03), Spearman rho=-0.26, n=51 states (incl. DC), p=0.0286, BH q=0.064. NULL: raw association not distinguishable from zero (BH q=0.064 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_rural is survey estimate (ACS); ts_age_adj_rate_change_2019_2020_pct is measured/administrative (derived change). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SA-1", "level": "state", "kind": "artefact_test", "x": { "name": "geriatricians_count", "unit": "count", "year": "2021", "source": "Alzheimer's Association 2025 Facts and Figures, Table 14, citing American Geriatrics Society via data/panel_state.csv", "tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, "y": { "name": "alz_count_65plus", "unit": "count", "year": "2020", "source": "Alzheimer's Association 2025 Alzheimer's Disease Facts and Figures, Table 5 via data/panel_state.csv", "tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.9256, "p": 2.57e-22, "ci95": [ 0.8726, 0.9571 ] }, "spearman": { "rho": 0.9341, "p": 1.44e-23 }, "partial_primary": { "controls": [ "pop_total" ], "r": 0.439, "p": 0.00127, "n": 51 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "population size" ], "note": "Two raw counts (baseline S5/S6 used counts).", "fdr_q_within_level": 4.88e-21, "classification": "artefact_collapses", "verdict": "Raw strong positive correlation r=0.93 (95% CI 0.87 to 0.96), Spearman rho=0.93, n=51 states (incl. DC), p=2.57e-22, BH q=4.9e-21. POPULATION/STRUCTURE ARTEFACT: raw r=0.926 falls to partial r=0.439 controlling for pop_total. The raw coefficient is not a relationship. Two raw counts (baseline S5/S6 used counts). Provenance: geriatricians_count is modelled (Alzheimer's Association projection built from age/sex/race distributions); alz_count_65plus is modelled (Alzheimer's Association projection built from age/sex/race distributions). Both variables are modelled: this coefficient compares two models, not two measurements. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SA-2", "level": "state", "kind": "artefact_test", "x": { "name": "medicaid_dementia_spend_total_usd", "unit": "usd", "year": "2025", "source": "Alzheimer's Association 2025 Facts and Figures, Table 22 (Lewin Model estimate, in 2024 dollars) via data/panel_state.csv", "tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, "y": { "name": "ts_dementia_deaths_total_2017", "unit": "count of deaths with any dementia underlying cause (G30, F01, F03, G31)", "year": "2017", "source": "CDC WONDER via data/panel_state_timeseries.csv", "tag": "measured/administrative (count)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.884, "p": 8.39e-18, "ci95": [ 0.8044, 0.9324 ] }, "spearman": { "rho": 0.941, "p": 1.05e-24 }, "partial_primary": { "controls": [ "pop_total" ], "r": 0.2554, "p": 0.0704, "n": 51 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "population size" ], "note": "Two raw totals.", "fdr_q_within_level": 1.28e-16, "classification": "artefact_collapses", "verdict": "Raw strong positive correlation r=0.88 (95% CI 0.80 to 0.93), Spearman rho=0.94, n=51 states (incl. DC), p=8.39e-18, BH q=1.3e-16. POPULATION/STRUCTURE ARTEFACT: raw r=0.884 falls to partial r=0.255 controlling for pop_total. The raw coefficient is not a relationship. Two raw totals. Provenance: medicaid_dementia_spend_total_usd is modelled (Alzheimer's Association projection built from age/sex/race distributions); ts_dementia_deaths_total_2017 is measured/administrative (count). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SA-3", "level": "state", "kind": "artefact_test", "x": { "name": "unpaid_caregivers_count", "unit": "count", "year": "2024", "source": "Alzheimer's Association 2025 Facts and Figures, Table 10 via data/panel_state.csv", "tag": "modelled (Alzheimer's Association projection built from age/sex/race distributions)" }, "y": { "name": "norc_state_prev_cnt_2020", "unit": "count of Medicare beneficiaries with a prevalent dementia diagnosis, state total (raw count, scales with population)", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, state total row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.9712, "p": 3.62e-32, "ci95": [ 0.9498, 0.9835 ] }, "spearman": { "rho": 0.9762, "p": 3.38e-34 }, "partial_primary": { "controls": [ "pop_total" ], "r": 0.2611, "p": 0.0642, "n": 51 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "population size" ], "note": "Baseline S11 lesson with measured NORC counts.", "fdr_q_within_level": 1.38e-30, "classification": "artefact_collapses", "verdict": "Raw strong positive correlation r=0.97 (95% CI 0.95 to 0.98), Spearman rho=0.98, n=51 states (incl. DC), p=3.62e-32, BH q=1.4e-30. POPULATION/STRUCTURE ARTEFACT: raw r=0.971 falls to partial r=0.261 controlling for pop_total. The raw coefficient is not a relationship. Baseline S11 lesson with measured NORC counts. Provenance: unpaid_caregivers_count is modelled (Alzheimer's Association projection built from age/sex/race distributions); norc_state_prev_cnt_2020 is measured/administrative (claims-based DIAGNOSED cases). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SS-1", "level": "state", "kind": "exposure", "x": { "name": "pct_depression", "unit": "percent", "year": "2024", "source": "America's Health Rankings, Depression (CDC BRFSS) via data/panel_state.csv", "tag": "survey estimate / administrative compilation" }, "y": { "name": "ts_scd_reported__65plus_pct_latest", "unit": "percent of adults 65+ reporting subjective cognitive decline, most recent BRFSS year available per state (2015-2022)", "year": "2015-2022 (latest per state)", "source": "CDC BRFSS Cognitive Decline module via data/panel_state_timeseries.csv", "tag": "survey estimate (self-report)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.1707, "p": 0.231, "ci95": [ -0.11, 0.4263 ] }, "spearman": { "rho": 0.1632, "p": 0.252 }, "partial_primary": { "controls": [ "median_household_income" ], "r": 0.2866, "p": 0.0415, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "region_south" ], "r": 0.2987, "p": 0.0332, "n": 51 }, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "depression and subjective decline co-occur" }, "confounders_named": [ "shared survey (both BRFSS self-report)", "reporting style" ], "note": "Baseline A1 used the overall-age SCD; here the 65+ stratum.", "fdr_q_within_level": 0.351, "classification": "null", "verdict": "Raw weak positive correlation r=0.17 (95% CI -0.11 to 0.43), Spearman rho=0.16, n=51 states (incl. DC), p=0.231, BH q=0.35. NULL: raw association not distinguishable from zero (BH q=0.35 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Baseline A1 used the overall-age SCD; here the 65+ stratum. Provenance: pct_depression is survey estimate / administrative compilation; ts_scd_reported__65plus_pct_latest is survey estimate (self-report). Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SS-2", "level": "state", "kind": "exposure", "x": { "name": "pct_bachelors_plus", "unit": "percent", "year": "2023", "source": "U.S. Census Bureau, ACS 2023 1-Year Estimates, GCT1502 via data/panel_state.csv", "tag": "survey estimate (ACS 1-year)" }, "y": { "name": "ts_scd_reported__65plus_pct_latest", "unit": "percent of adults 65+ reporting subjective cognitive decline, most recent BRFSS year available per state (2015-2022)", "year": "2015-2022 (latest per state)", "source": "CDC BRFSS Cognitive Decline module via data/panel_state_timeseries.csv", "tag": "survey estimate (self-report)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.1572, "p": 0.27, "ci95": [ -0.1237, 0.4149 ] }, "spearman": { "rho": 0.2868, "p": 0.0413 }, "partial_primary": { "controls": [ "median_household_income" ], "r": -0.0325, "p": 0.821, "n": 51 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "negative", "why": "education protects cognition" }, "confounders_named": [ "survey response", "income" ], "note": "", "fdr_q_within_level": 0.402, "classification": "null", "verdict": "Raw weak positive correlation r=0.16 (95% CI -0.12 to 0.41), Spearman rho=0.29, n=51 states (incl. DC), p=0.27, BH q=0.4. NULL: raw association not distinguishable from zero (BH q=0.4 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_bachelors_plus is survey estimate (ACS 1-year); ts_scd_reported__65plus_pct_latest is survey estimate (self-report). Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SS-3", "level": "state", "kind": "exposure", "x": { "name": "pct_hearing_loss", "unit": "percent", "year": "2016", "source": "CDC/NCHS, Any Hearing Loss by State, National Health Interview Survey 2014-2016 (Blackwell & Norris) via data/panel_state.csv", "tag": "survey estimate (self-report)" }, "y": { "name": "ts_scd_reported__65plus_pct_latest", "unit": "percent of adults 65+ reporting subjective cognitive decline, most recent BRFSS year available per state (2015-2022)", "year": "2015-2022 (latest per state)", "source": "CDC BRFSS Cognitive Decline module via data/panel_state_timeseries.csv", "tag": "survey estimate (self-report)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.034, "p": 0.813, "ci95": [ -0.3067, 0.2439 ] }, "spearman": { "rho": -0.0934, "p": 0.515 }, "partial_primary": { "controls": [ "median_household_income" ], "r": 0.1016, "p": 0.478, "n": 51 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "hearing loss is a risk factor" }, "confounders_named": [ "shared self-report" ], "note": "", "fdr_q_within_level": 0.936, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.03 (95% CI -0.31 to 0.24), Spearman rho=-0.09, n=51 states (incl. DC), p=0.813, BH q=0.94. NULL: raw association not distinguishable from zero (BH q=0.94 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: pct_hearing_loss is survey estimate (self-report); ts_scd_reported__65plus_pct_latest is survey estimate (self-report). Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SS-4", "level": "state", "kind": "consistency", "x": { "name": "ts_scd_reported__65plus_pct_latest", "unit": "percent of adults 65+ reporting subjective cognitive decline, most recent BRFSS year available per state (2015-2022)", "year": "2015-2022 (latest per state)", "source": "CDC BRFSS Cognitive Decline module via data/panel_state_timeseries.csv", "tag": "survey estimate (self-report)" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.1712, "p": 0.23, "ci95": [ -0.4267, 0.1096 ] }, "spearman": { "rho": -0.1669, "p": 0.242 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "coding practice" ], "note": "Self-reported decline vs age-adjusted mortality.", "fdr_q_within_level": 0.351, "classification": "consistency_check", "verdict": "Raw weak negative correlation r=-0.17 (95% CI -0.43 to 0.11), Spearman rho=-0.17, n=51 states (incl. DC), p=0.23, BH q=0.35. CONSISTENCY CHECK: r=-0.171: agreement level between two versions of the outcome / measurement. Self-reported decline vs age-adjusted mortality. Provenance: ts_scd_reported__65plus_pct_latest is survey estimate (self-report); ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SS-5", "level": "state", "kind": "consistency", "x": { "name": "ts_scd_reported__65plus_pct_mean_2015_2022", "unit": "mean over available BRFSS years of percent of adults 65+ reporting subjective cognitive decline", "year": "2015-2022", "source": "CDC BRFSS via data/panel_state_timeseries.csv", "tag": "survey estimate (self-report)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.1206, "p": 0.399, "ci95": [ -0.3834, 0.1603 ] }, "spearman": { "rho": -0.1534, "p": 0.283 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "survey response" ], "note": "Multi-year mean SCD (less single-year noise) vs diagnosed prevalence.", "fdr_q_within_level": 0.523, "classification": "consistency_check", "verdict": "Raw weak negative correlation r=-0.12 (95% CI -0.38 to 0.16), Spearman rho=-0.15, n=51 states (incl. DC), p=0.399, BH q=0.52. CONSISTENCY CHECK: r=-0.121: agreement level between two versions of the outcome / measurement. Multi-year mean SCD (less single-year noise) vs diagnosed prevalence. Provenance: ts_scd_reported__65plus_pct_mean_2015_2022 is survey estimate (self-report); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SK-11", "level": "state", "kind": "measured_vs_modelled", "x": { "name": "nursing_home_beds_per_1000", "unit": "rate per 1000 (65+ population)", "year": "2026", "source": "Becker's Hospital Review, States ranked by nursing home beds per capita (May 2026), using Census population + KFF CNF bed data via data/panel_state.csv", "tag": "measured/administrative" }, "y": { "name": "cms_nh_beds_per_1000_pop", "unit": "CMS-certified nursing-home beds per 1,000 population (county: against panel population; state: beds summed by state then divided by panel_state.csv's own pop_total, not a sum of per-1000 rates)", "year": "2026", "source": "CMS Nursing Home Care Compare Provider Info, Aug 2026 via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio); blank for Connecticut (0 matching CMS rows under either FIPS scheme) and 316 other counties with no Care Compare facility" }, "n": 50, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 50, "frame_rows": 51, "pearson": { "r": 0.9534, "p": 1.23e-26, "ci95": [ 0.919, 0.9734 ] }, "spearman": { "rho": 0.9401, "p": 4.41e-24 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "different source and vintage: CHR pass-through vs CMS Care Compare certified beds, 2026" ], "note": "Unlike at county level (where the old column is entirely blank), both nursing-home bed measures are populated at state level: do they agree?", "fdr_q_within_level": 3.12e-25, "classification": "twin", "verdict": "Raw strong positive correlation r=0.95 (95% CI 0.92 to 0.97), Spearman rho=0.94, n=50 states (incl. DC), p=1.23e-26, BH q=3.1e-25. TWIN: |r|=0.953 >= 0.95: the two columns are the same quantity twice; never use both in one model. Unlike at county level (where the old column is entirely blank), both nursing-home bed measures are populated at state level: do they agree? Provenance: nursing_home_beds_per_1000 is measured/administrative; cms_nh_beds_per_1000_pop is measured/administrative (derived ratio); blank for Connecticut (0 matching CMS rows under either FIPS scheme) and 316 other counties with no Care Compare facility. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SP-23", "level": "state", "kind": "exposure", "x": { "name": "cms_nh_beds_per_1000_pop", "unit": "CMS-certified nursing-home beds per 1,000 population (county: against panel population; state: beds summed by state then divided by panel_state.csv's own pop_total, not a sum of per-1000 rates)", "year": "2026", "source": "CMS Nursing Home Care Compare Provider Info, Aug 2026 via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio); blank for Connecticut (0 matching CMS rows under either FIPS scheme) and 316 other counties with no Care Compare facility" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 50, "n_unit": "states (incl. DC)", "x_filled": 50, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.4172, "p": 0.00258, "ci95": [ 0.157, 0.6231 ] }, "spearman": { "rho": 0.3984, "p": 0.00416 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus" ], "r": 0.3912, "p": 0.00497, "n": 50 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.4248, "p": 0.0021, "n": 50 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.3662, "p": 0.0089, "n": 50 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "institutional diagnosis (nursing-home residents are screened)", "region" ], "note": "Re-test of the baseline's strongest state-level survivor (SP-21, nursing_home_beds_per_1000: raw r=0.48, strict partial r=0.41) using the new Phase 4 CMS-sourced bed count instead of the CHR pass-through.", "fdr_q_within_level": 0.00654, "classification": "survives", "verdict": "Raw moderate positive correlation r=0.42 (95% CI 0.16 to 0.62), Spearman rho=0.40, n=50 states (incl. DC), p=0.00258, BH q=0.0065. SURVIVES the confounder test: partial r=0.391 (p=0.00497) controlling for median_household_income+pct_85plus; extended partial r=0.4248; strict partial r=0.3662 controlling for median_household_income+pct_85plus+region_south+medicare_percapita_ffs_payment_usd+pct_race_black. A place-level association not explained by the named control(s); candidate third things still unaddressed: institutional diagnosis (nursing-home residents are screened), region. Re-test of the baseline's strongest state-level survivor (SP-21, nursing_home_beds_per_1000: raw r=0.48, strict partial r=0.41) using the new Phase 4 CMS-sourced bed count instead of the CHR pass-through. Provenance: cms_nh_beds_per_1000_pop is measured/administrative (derived ratio); blank for Connecticut (0 matching CMS rows under either FIPS scheme) and 316 other counties with no Care Compare facility; norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SM-19", "level": "state", "kind": "exposure", "x": { "name": "cms_nh_beds_per_1000_pop", "unit": "CMS-certified nursing-home beds per 1,000 population (county: against panel population; state: beds summed by state then divided by panel_state.csv's own pop_total, not a sum of per-1000 rates)", "year": "2026", "source": "CMS Nursing Home Care Compare Provider Info, Aug 2026 via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio); blank for Connecticut (0 matching CMS rows under either FIPS scheme) and 316 other counties with no Care Compare facility" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 50, "n_unit": "states (incl. DC)", "x_filled": 50, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.0108, "p": 0.94, "ci95": [ -0.2883, 0.2683 ] }, "spearman": { "rho": -0.0481, "p": 0.74 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": -0.1588, "p": 0.271, "n": 50 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": -0.1565, "p": 0.278, "n": 50 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": -0.1266, "p": 0.381, "n": 50 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "place of death (nursing-home deaths get dementia coded)", "coding practice" ], "note": "Re-test of SM-15 (nursing_home_beds_per_1000 vs age-adjusted mortality, null) with the new CMS bed count.", "fdr_q_within_level": 0.979, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.01 (95% CI -0.29 to 0.27), Spearman rho=-0.05, n=50 states (incl. DC), p=0.94, BH q=0.98. NULL: raw association not distinguishable from zero (BH q=0.98 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Re-test of SM-15 (nursing_home_beds_per_1000 vs age-adjusted mortality, null) with the new CMS bed count. Provenance: cms_nh_beds_per_1000_pop is measured/administrative (derived ratio); blank for Connecticut (0 matching CMS rows under either FIPS scheme) and 316 other counties with no Care Compare facility; ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SP-24", "level": "state", "kind": "diagnosis_intensity", "x": { "name": "medicare_advantage_penetration_pct", "unit": "Medicare Advantage + other plan beneficiaries as percent of total Medicare beneficiaries, June 2026 (direct division of two published CMS counts, not modelled); an independent enrollment-based MA share, distinct from norc_ma_share_of_bene_2020 which is the MA share among Medicare beneficiaries who appear in the 2020 dementia claims file", "year": "2026-06", "source": "CMS Medicare Monthly Enrollment via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio); blank for Connecticut's 8 legacy counties and small-cell-suppressed counties" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.5105, "p": 0.00013, "ci95": [ 0.2733, 0.6891 ] }, "spearman": { "rho": 0.4325, "p": 0.00153 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus" ], "r": 0.4801, "p": 0.000363, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.5089, "p": 0.000137, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.48, "p": 0.000364, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "health-system contact", "coding practice", "insurance type mix" ], "note": "New in Phase 4: CMS enrollment-based MA share of ALL Medicare beneficiaries. No prior state-level MA measure existed to compare it against.", "fdr_q_within_level": 0.00052, "classification": "survives", "verdict": "Raw strong positive correlation r=0.51 (95% CI 0.27 to 0.69), Spearman rho=0.43, n=51 states (incl. DC), p=0.00013, BH q=0.00052. SURVIVES the confounder test: partial r=0.480 (p=0.000363) controlling for median_household_income+pct_85plus; extended partial r=0.5089; strict partial r=0.48 controlling for median_household_income+pct_85plus+region_south+medicare_percapita_ffs_payment_usd+pct_race_black. A place-level association not explained by the named control(s); candidate third things still unaddressed: health-system contact, coding practice, insurance type mix. New in Phase 4: CMS enrollment-based MA share of ALL Medicare beneficiaries. No prior state-level MA measure existed to compare it against. Provenance: medicare_advantage_penetration_pct is measured/administrative (derived ratio); blank for Connecticut's 8 legacy counties and small-cell-suppressed counties; norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SM-20", "level": "state", "kind": "diagnosis_intensity", "x": { "name": "medicare_advantage_penetration_pct", "unit": "Medicare Advantage + other plan beneficiaries as percent of total Medicare beneficiaries, June 2026 (direct division of two published CMS counts, not modelled); an independent enrollment-based MA share, distinct from norc_ma_share_of_bene_2020 which is the MA share among Medicare beneficiaries who appear in the 2020 dementia claims file", "year": "2026-06", "source": "CMS Medicare Monthly Enrollment via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio); blank for Connecticut's 8 legacy counties and small-cell-suppressed counties" }, "y": { "name": "ts_alz_age_adjusted_death_rate_2022", "unit": "Alzheimer's (G30) deaths per 100,000, age-adjusted to 2000 US standard", "year": "2022", "source": "CDC WONDER / NCHS via data/panel_state_timeseries.csv", "tag": "measured/administrative (vital registration)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.1361, "p": 0.341, "ci95": [ -0.1449, 0.3968 ] }, "spearman": { "rho": 0.1603, "p": 0.261 }, "partial_primary": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017" ], "r": 0.0376, "p": 0.794, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south" ], "r": 0.036, "p": 0.802, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "ts_dementia_deaths_pct_g30_2017", "region_south", "pct_race_black" ], "r": 0.0598, "p": 0.677, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "coding practice", "claims intensity" ], "note": "Does MA penetration behave anomalously against a G30-coded MORTALITY outcome the way it does as a diagnosed-PREVALENCE booster at county level (CD-3, CD's new medicare_advantage_penetration_pct card)?", "fdr_q_within_level": 0.48, "classification": "null", "verdict": "Raw weak positive correlation r=0.14 (95% CI -0.14 to 0.40), Spearman rho=0.16, n=51 states (incl. DC), p=0.341, BH q=0.48. NULL: raw association not distinguishable from zero (BH q=0.48 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Does MA penetration behave anomalously against a G30-coded MORTALITY outcome the way it does as a diagnosed-PREVALENCE booster at county level (CD-3, CD's new medicare_advantage_penetration_pct card)? Provenance: medicare_advantage_penetration_pct is measured/administrative (derived ratio); blank for Connecticut's 8 legacy counties and small-cell-suppressed counties; ts_alz_age_adjusted_death_rate_2022 is measured/administrative (vital registration). Outcome is cause-of-death coding (ICD-10 G30): state coding practice differs, so a high 'Alzheimer's' death rate can reflect certifiers writing G30 rather than F03. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "SP-25", "level": "state", "kind": "diagnosis_intensity", "x": { "name": "naics62_health_socasst_emplvl_per_1000pop", "unit": "NAICS 62 Health Care and Social Assistance employment level per 1,000 population (county: panel population; state: pop_total)", "year": "2024 employment / panel population year", "source": "derived in analysis/build_analysis_set.py from naics62_health_socasst_emplvl_annual_avg and population, BLS QCEW via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio; inherits the suppression floor of the employment count)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": -0.0126, "p": 0.93, "ci95": [ -0.2871, 0.2639 ] }, "spearman": { "rho": -0.0689, "p": 0.631 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus" ], "r": 0.1626, "p": 0.254, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.2389, "p": 0.0913, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.0144, "p": 0.92, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "urbanicity", "income", "detection capacity" ], "note": "New in Phase 4 (BLS QCEW), state-level per-1,000-population health+social-assistance employment.", "fdr_q_within_level": 0.979, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.01 (95% CI -0.29 to 0.26), Spearman rho=-0.07, n=51 states (incl. DC), p=0.93, BH q=0.98. NULL: raw association not distinguishable from zero (BH q=0.98 within level). Absence of a place-level correlation is not evidence of no individual-level effect. New in Phase 4 (BLS QCEW), state-level per-1,000-population health+social-assistance employment. Provenance: naics62_health_socasst_emplvl_per_1000pop is measured/administrative (derived ratio; inherits the suppression floor of the employment count); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SP-26", "level": "state", "kind": "diagnosis_intensity", "x": { "name": "naics623_nursing_residential_care_emplvl_per_1000pop", "unit": "NAICS 623 Nursing and Residential Care Facilities employment level per 1,000 population (county: panel population; state: pop_total)", "year": "2024 employment / panel population year", "source": "derived in analysis/build_analysis_set.py from naics623_nursing_residential_care_emplvl_annual_avg and population, BLS QCEW via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio; inherits the suppression floor of the employment count)" }, "y": { "name": "norc_state65_prevalence_rate_2020", "unit": "diagnosed dementia prevalence among beneficiaries 65+, percent", "year": "2020", "source": "NORC Dementia DataHub 2020 PUF, Nat_State file, AGECAT2=1 row", "tag": "measured/administrative (claims-based DIAGNOSED cases)" }, "n": 51, "n_unit": "states (incl. DC)", "x_filled": 51, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.1312, "p": 0.359, "ci95": [ -0.1498, 0.3926 ] }, "spearman": { "rho": 0.1959, "p": 0.168 }, "partial_primary": { "controls": [ "median_household_income", "pct_85plus" ], "r": 0.2454, "p": 0.0827, "n": 51 }, "partial_extended": { "controls": [ "median_household_income", "pct_85plus", "region_south" ], "r": 0.4037, "p": 0.00331, "n": 51 }, "partial_strict": { "controls": [ "median_household_income", "pct_85plus", "region_south", "medicare_percapita_ffs_payment_usd", "pct_race_black" ], "r": 0.2887, "p": 0.0399, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "long-term-care workforce presence" ], "note": "New in Phase 4 (BLS QCEW, NAICS 623 nursing/residential-care subsector), state-level per-1,000-population.", "fdr_q_within_level": 0.487, "classification": "null", "verdict": "Raw weak positive correlation r=0.13 (95% CI -0.15 to 0.39), Spearman rho=0.20, n=51 states (incl. DC), p=0.359, BH q=0.49. NULL: raw association not distinguishable from zero (BH q=0.49 within level). Absence of a place-level correlation is not evidence of no individual-level effect. New in Phase 4 (BLS QCEW, NAICS 623 nursing/residential-care subsector), state-level per-1,000-population. Provenance: naics623_nursing_residential_care_emplvl_per_1000pop is measured/administrative (derived ratio; inherits the suppression floor of the employment count); norc_state65_prevalence_rate_2020 is measured/administrative (claims-based DIAGNOSED cases). Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "Outcome is DIAGNOSED dementia from Medicare claims: a low rate can mean less disease or less diagnosis, and the two are indistinguishable in this file." ] }, { "id": "SK-12", "level": "state", "kind": "consistency", "x": { "name": "cms_nh_beds_per_1000_pop", "unit": "CMS-certified nursing-home beds per 1,000 population (county: against panel population; state: beds summed by state then divided by panel_state.csv's own pop_total, not a sum of per-1000 rates)", "year": "2026", "source": "CMS Nursing Home Care Compare Provider Info, Aug 2026 via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio); blank for Connecticut (0 matching CMS rows under either FIPS scheme) and 316 other counties with no Care Compare facility" }, "y": { "name": "naics623_nursing_residential_care_emplvl_per_1000pop", "unit": "NAICS 623 Nursing and Residential Care Facilities employment level per 1,000 population (county: panel population; state: pop_total)", "year": "2024 employment / panel population year", "source": "derived in analysis/build_analysis_set.py from naics623_nursing_residential_care_emplvl_annual_avg and population, BLS QCEW via data/panel_county.csv / data/panel_state.csv", "tag": "measured/administrative (derived ratio; inherits the suppression floor of the employment count)" }, "n": 50, "n_unit": "states (incl. DC)", "x_filled": 50, "y_filled": 51, "frame_rows": 51, "pearson": { "r": 0.4499, "p": 0.00105, "ci95": [ 0.1961, 0.6472 ] }, "spearman": { "rho": 0.4698, "p": 0.000578 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "two different federal agencies' proxies for the same underlying long-term-care capacity: CMS certified beds vs BLS nursing/residential-care employment" ], "note": "Cross-agency check on long-term-care capacity.", "fdr_q_within_level": 0.00319, "classification": "consistency_check", "verdict": "Raw moderate positive correlation r=0.45 (95% CI 0.20 to 0.65), Spearman rho=0.47, n=50 states (incl. DC), p=0.00105, BH q=0.0032. CONSISTENCY CHECK: r=0.450: agreement level between two versions of the outcome / measurement. Cross-agency check on long-term-care capacity. Provenance: cms_nh_beds_per_1000_pop is measured/administrative (derived ratio); blank for Connecticut (0 matching CMS rows under either FIPS scheme) and 316 other counties with no Care Compare facility; naics623_nursing_residential_care_emplvl_per_1000pop is measured/administrative (derived ratio; inherits the suppression floor of the employment count). Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KA-1", "level": "country", "kind": "artefact_test", "x": { "name": "ts_gbd_deaths_per_100k_age_std_2021", "unit": "dementia deaths per 100,000, labelled age-standardised upstream but numerically identical to the crude rate (see collinearity)", "year": "2021", "source": "IHME GBD via OWID mirror, data/panel_country_timeseries.csv", "tag": "modelled (GBD estimate)" }, "y": { "name": "ts_gbd_deaths_crude_per_100k_2021", "unit": "dementia deaths per 100,000 total population, crude (computed in the timeseries build)", "year": "2021", "source": "IHME/OWID deaths over World Bank population, data/panel_country_timeseries.csv", "tag": "modelled numerator / measured denominator" }, "n": 143, "n_unit": "countries", "x_filled": 143, "y_filled": 143, "frame_rows": 217, "pearson": { "r": 0.9994, "p": 2.01e-208, "ci95": [ 0.9992, 0.9996 ] }, "spearman": { "rho": 0.9985, "p": 1.36e-180 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "no age standardisation was applied upstream" ], "note": "The column labelled age-standardised against the crude rate computed from deaths and population. r near 1 means the label is wrong: there is NO age-standardised country outcome anywhere in the data.", "fdr_q_within_level": 6.23e-207, "classification": "artefact_documented", "verdict": "Raw strong positive correlation r=1.00 (95% CI 1.00 to 1.00), Spearman rho=1.00, n=143 countries, p=2.01e-208, BH q=6.2e-207. STRUCTURAL DEPENDENCE DOCUMENTED: r=0.999, q=6.2e-207: documents the mechanical/structural dependence named. The column labelled age-standardised against the crude rate computed from deaths and population. r near 1 means the label is wrong: there is NO age-standardised country outcome anywhere in the data. Provenance: ts_gbd_deaths_per_100k_age_std_2021 is modelled (GBD estimate); ts_gbd_deaths_crude_per_100k_2021 is modelled numerator / measured denominator. Both variables are modelled: this coefficient compares two models, not two measurements. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KA-2", "level": "country", "kind": "consistency", "x": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "y": { "name": "ts_gbd_deaths_per_100k_age_std_2021", "unit": "dementia deaths per 100,000, labelled age-standardised upstream but numerically identical to the crude rate (see collinearity)", "year": "2021", "source": "IHME GBD via OWID mirror, data/panel_country_timeseries.csv", "tag": "modelled (GBD estimate)" }, "n": 143, "n_unit": "countries", "x_filled": 192, "y_filled": 143, "frame_rows": 217, "pearson": { "r": 1.0, "p": null, "ci95": [ null, null ] }, "spearman": { "rho": 1.0, "p": null }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "WHO GHE rate (panel_country) vs IHME/OWID rate (timeseries): are the two 'sources' the same series?", "fdr_q_within_level": null, "classification": "twin", "verdict": "Raw r=1.00, n=143. TWIN: |r|=1.000 >= 0.95: the two columns are the same quantity twice; never use both in one model. WHO GHE rate (panel_country) vs IHME/OWID rate (timeseries): are the two 'sources' the same series? Provenance: dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble); ts_gbd_deaths_per_100k_age_std_2021 is modelled (GBD estimate). Both variables are modelled: this coefficient compares two models, not two measurements. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KA-3", "level": "country", "kind": "artefact_test", "x": { "name": "pct_65plus", "unit": "% of total population", "year": "latest per country, mostly 2025", "source": "World Bank via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 192, "n_unit": "countries", "x_filled": 217, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.7179, "p": 1.05e-31, "ci95": [ 0.6415, 0.7802 ] }, "spearman": { "rho": 0.7772, "p": 4.41e-40 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure IS the exposure here" ], "note": "How much of the country death-rate variation is age structure.", "fdr_q_within_level": 5.42e-31, "classification": "artefact_documented", "verdict": "Raw strong positive correlation r=0.72 (95% CI 0.64 to 0.78), Spearman rho=0.78, n=192 countries, p=1.05e-31, BH q=5.4e-31. STRUCTURAL DEPENDENCE DOCUMENTED: r=0.718, q=5.4e-31: documents the mechanical/structural dependence named. How much of the country death-rate variation is age structure. Provenance: pct_65plus is administrative compilation (World Bank); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KA-4", "level": "country", "kind": "artefact_test", "x": { "name": "pct_65plus", "unit": "% of total population", "year": "latest per country, mostly 2025", "source": "World Bank via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_daly_per_100k", "unit": "DALYs per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 192, "n_unit": "countries", "x_filled": 217, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.8381, "p": 6.9e-52, "ci95": [ 0.7902, 0.8757 ] }, "spearman": { "rho": 0.919, "p": 1.02e-78 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure" ], "note": "", "fdr_q_within_level": 7.13e-51, "classification": "artefact_documented", "verdict": "Raw strong positive correlation r=0.84 (95% CI 0.79 to 0.88), Spearman rho=0.92, n=192 countries, p=6.9e-52, BH q=7.1e-51. STRUCTURAL DEPENDENCE DOCUMENTED: r=0.838, q=7.1e-51: documents the mechanical/structural dependence named. Provenance: pct_65plus is administrative compilation (World Bank); dementia_daly_per_100k is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-1", "level": "country", "kind": "exposure", "x": { "name": "gdp_per_capita_usd", "unit": "current US$", "year": "latest per country, 2011-2025", "source": "World Bank via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 191, "n_unit": "countries", "x_filled": 214, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.8176, "p": 3.52e-47, "ci95": [ 0.7643, 0.8597 ] }, "spearman": { "rho": 0.6366, "p": 4.28e-23 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": 0.6878, "p": 4.3e-28, "n": 191 }, "partial_extended": { "controls": [ "pct_65plus" ], "r": 0.6878, "p": 4.3e-28, "n": 191 }, "partial_strict": { "controls": [ "pct_65plus", "life_expectancy" ], "r": 0.7013, "p": 1.37e-29, "n": 191 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure (rich countries are old)", "death-certification completeness", "competing mortality" ], "note": "Baseline I1 (r=0.82) with age structure controlled. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 2.73e-46, "classification": "survives", "verdict": "Raw strong positive correlation r=0.82 (95% CI 0.76 to 0.86), Spearman rho=0.64, n=191 countries, p=3.52e-47, BH q=2.7e-46. SURVIVES the confounder test: partial r=0.688 (p=4.3e-28) controlling for pct_65plus; extended partial r=0.6878; strict partial r=0.7013 controlling for pct_65plus+life_expectancy. A place-level association not explained by the named control(s); candidate third things still unaddressed: age structure (rich countries are old), death-certification completeness, competing mortality. Baseline I1 (r=0.82) with age structure controlled. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: gdp_per_capita_usd is administrative compilation (World Bank); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-2", "level": "country", "kind": "exposure", "x": { "name": "health_spend_per_capita_usd", "unit": "current US$", "year": "latest per country, 2020-2024", "source": "World Bank / WHO GHED via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 191, "n_unit": "countries", "x_filled": 193, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.7396, "p": 2.45e-34, "ci95": [ 0.6678, 0.7978 ] }, "spearman": { "rho": 0.6589, "p": 3.75e-25 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": 0.5191, "p": 1.44e-14, "n": 191 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": 0.029, "p": 0.691, "n": 191 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": 0.0489, "p": 0.502, "n": 191 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure", "GDP" ], "note": "Baseline I3 with age structure controlled. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 1.52e-33, "classification": "collapses_extended", "verdict": "Raw strong positive correlation r=0.74 (95% CI 0.67 to 0.80), Spearman rho=0.66, n=191 countries, p=2.45e-34, BH q=1.5e-33. COLLAPSES under the extended controls: survives the primary control (partial r=0.519) but falls to 0.029 with the extended set pct_65plus+gdp_per_capita_usd. Baseline I3 with age structure controlled. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: health_spend_per_capita_usd is administrative compilation (World Bank); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-3", "level": "country", "kind": "exposure", "x": { "name": "physicians_per_1000", "unit": "per 1,000 population", "year": "latest per country, 1980-2023", "source": "World Bank (WHO Global Health Workforce Statistics) via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 192, "n_unit": "countries", "x_filled": 208, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.6204, "p": 8.23e-22, "ci95": [ 0.5249, 0.7005 ] }, "spearman": { "rho": 0.6527, "p": 1.09e-24 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": 0.1333, "p": 0.0652, "n": 192 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.0539, "p": 0.459, "n": 191 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": 0.0159, "p": 0.828, "n": 191 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure", "GDP" ], "note": "Baseline J2 with age structure controlled. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 2.83e-21, "classification": "collapses", "verdict": "Raw strong positive correlation r=0.62 (95% CI 0.52 to 0.70), Spearman rho=0.65, n=192 countries, p=8.23e-22, BH q=2.8e-21. COLLAPSES: partial r=0.133 is under half the raw r=0.620 once pct_65plus is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with physicians_per_1000. Baseline J2 with age structure controlled. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: physicians_per_1000 is administrative compilation (World Bank); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-4", "level": "country", "kind": "exposure", "x": { "name": "pct_urban", "unit": "%", "year": "latest per country, mostly 2025", "source": "World Bank via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 192, "n_unit": "countries", "x_filled": 217, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.4319, "p": 3.96e-10, "ci95": [ 0.3092, 0.5405 ] }, "spearman": { "rho": 0.4377, "p": 2.18e-10 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": 0.1661, "p": 0.0213, "n": 192 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.0674, "p": 0.354, "n": 191 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": 0.012, "p": 0.869, "n": 191 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure", "GDP" ], "note": "Baseline J3. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 8.77e-10, "classification": "collapses", "verdict": "Raw moderate positive correlation r=0.43 (95% CI 0.31 to 0.54), Spearman rho=0.44, n=192 countries, p=3.96e-10, BH q=8.8e-10. COLLAPSES: partial r=0.166 is under half the raw r=0.432 once pct_65plus is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_urban. Baseline J3. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: pct_urban is administrative compilation (World Bank); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-5", "level": "country", "kind": "exposure", "x": { "name": "literacy_rate", "unit": "%", "year": "latest per country, 1970-2024", "source": "World Bank (UNESCO Institute for Statistics) via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 162, "n_unit": "countries", "x_filled": 170, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.3674, "p": 1.52e-06, "ci95": [ 0.226, 0.4936 ] }, "spearman": { "rho": 0.4707, "p": 2.6e-10 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": -0.0799, "p": 0.312, "n": 162 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.1143, "p": 0.149, "n": 161 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.1747, "p": 0.0266, "n": 161 }, "divergence_flag": false, "sign_flip_vs_literature": true, "literature_expectation": { "direction": "negative", "why": "education is protective" }, "confounders_named": [ "age structure", "GDP", "high-income countries do not report literacy (missing = rich)" ], "note": "Baseline J4. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 2.48e-06, "classification": "collapses", "verdict": "Raw moderate positive correlation r=0.37 (95% CI 0.23 to 0.49), Spearman rho=0.47, n=162 countries, p=1.52e-06, BH q=2.5e-06. COLLAPSES: partial r=-0.080 is under half the raw r=0.367 once pct_65plus is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with literacy_rate. SIGN FLIP versus individual-level literature (education is protective): read as a statement about measurement or place, not about risk. Baseline J4. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: literacy_rate is administrative compilation (World Bank); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-6", "level": "country", "kind": "exposure", "x": { "name": "education_years_mean", "unit": "years", "year": "2010 for all countries (Barro-Lee dataset last release)", "source": "World Bank, mirroring the Barro-Lee Educational Attainment dataset via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 142, "n_unit": "countries", "x_filled": 144, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.5745, "p": 7.68e-14, "ci95": [ 0.4526, 0.6753 ] }, "spearman": { "rho": 0.6234, "p": 1.17e-16 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": 0.0127, "p": 0.881, "n": 142 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.1698, "p": 0.0434, "n": 142 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.1621, "p": 0.0539, "n": 142 }, "divergence_flag": false, "sign_flip_vs_literature": true, "literature_expectation": { "direction": "negative", "why": "education is protective" }, "confounders_named": [ "age structure", "GDP", "2010 vintage" ], "note": "A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 1.98e-13, "classification": "collapses", "verdict": "Raw strong positive correlation r=0.57 (95% CI 0.45 to 0.68), Spearman rho=0.62, n=142 countries, p=7.68e-14, BH q=2e-13. COLLAPSES: partial r=0.013 is under half the raw r=0.575 once pct_65plus is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with education_years_mean. SIGN FLIP versus individual-level literature (education is protective): read as a statement about measurement or place, not about risk. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: education_years_mean is administrative compilation (World Bank); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-7", "level": "country", "kind": "exposure", "x": { "name": "pct_diabetes", "unit": "% of adults", "year": "latest per country, mostly 2022", "source": "WHO Global Health Observatory (NCD-RisC) via data/panel_country.csv", "tag": "modelled (NCD-RisC)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 190, "n_unit": "countries", "x_filled": 196, "y_filled": 192, "frame_rows": 217, "pearson": { "r": -0.4251, "p": 9.71e-10, "ci95": [ -0.5351, -0.301 ] }, "spearman": { "rho": -0.3159, "p": 8.97e-06 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": -0.2422, "p": 0.00076, "n": 190 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.2355, "p": 0.0011, "n": 189 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.2329, "p": 0.00126, "n": 189 }, "divergence_flag": false, "sign_flip_vs_literature": true, "literature_expectation": { "direction": "positive", "why": "diabetes raises risk" }, "confounders_named": [ "age structure", "GDP", "NCD-RisC model" ], "note": "Baseline I2 sign flip re-tested with age structure controlled. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 2.01e-09, "classification": "survives_sign_flipped", "verdict": "Raw moderate negative correlation r=-0.43 (95% CI -0.54 to -0.30), Spearman rho=-0.32, n=190 countries, p=9.71e-10, BH q=2e-09. SURVIVES BUT SIGN-FLIPPED: partial r=-0.242 (p=0.00076) controlling for pct_65plus; extended partial r=-0.2355; strict partial r=-0.2329 controlling for pct_65plus+gdp_per_capita_usd+life_expectancy. The association is robust to the controls but points the OPPOSITE way to individual-level evidence, so it is a statement about how the outcome is measured or certified in these places, not about risk. SIGN FLIP versus individual-level literature (diabetes raises risk): read as a statement about measurement or place, not about risk. Baseline I2 sign flip re-tested with age structure controlled. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: pct_diabetes is modelled (NCD-RisC); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). Both variables are modelled: this coefficient compares two models, not two measurements. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-8", "level": "country", "kind": "exposure", "x": { "name": "pct_obesity", "unit": "% of adults 18+", "year": "latest per country, mostly 2024", "source": "WHO Global Health Observatory (NCD-RisC) via data/panel_country.csv", "tag": "modelled (NCD-RisC)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 190, "n_unit": "countries", "x_filled": 196, "y_filled": 192, "frame_rows": 217, "pearson": { "r": -0.0176, "p": 0.81, "ci95": [ -0.1595, 0.1251 ] }, "spearman": { "rho": 0.1224, "p": 0.0926 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": -0.0803, "p": 0.271, "n": 190 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.1381, "p": 0.0581, "n": 189 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.1298, "p": 0.075, "n": 189 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "obesity raises risk" }, "confounders_named": [ "age structure", "GDP" ], "note": "A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 0.83, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.02 (95% CI -0.16 to 0.13), Spearman rho=0.12, n=190 countries, p=0.81, BH q=0.83. NULL: raw association not distinguishable from zero (BH q=0.83 within level). Absence of a place-level correlation is not evidence of no individual-level effect. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: pct_obesity is modelled (NCD-RisC); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). Both variables are modelled: this coefficient compares two models, not two measurements. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-9", "level": "country", "kind": "exposure", "x": { "name": "pct_hypertension", "unit": "% of adults 30-79", "year": "latest per country, mostly 2019", "source": "WHO Global Health Observatory (NCD-RisC) via data/panel_country.csv", "tag": "modelled (NCD-RisC)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 190, "n_unit": "countries", "x_filled": 192, "y_filled": 192, "frame_rows": 217, "pearson": { "r": -0.3186, "p": 7.44e-06, "ci95": [ -0.441, -0.1847 ] }, "spearman": { "rho": -0.1207, "p": 0.0973 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": -0.4427, "p": 1.6e-10, "n": 190 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.3579, "p": 4.29e-07, "n": 189 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.3611, "p": 3.32e-07, "n": 189 }, "divergence_flag": false, "sign_flip_vs_literature": true, "literature_expectation": { "direction": "positive", "why": "hypertension raises risk" }, "confounders_named": [ "age structure", "GDP" ], "note": "A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 1.1e-05, "classification": "survives_sign_flipped", "verdict": "Raw moderate negative correlation r=-0.32 (95% CI -0.44 to -0.18), Spearman rho=-0.12, n=190 countries, p=7.44e-06, BH q=1.1e-05. SURVIVES BUT SIGN-FLIPPED: partial r=-0.443 (p=1.6e-10) controlling for pct_65plus; extended partial r=-0.3579; strict partial r=-0.3611 controlling for pct_65plus+gdp_per_capita_usd+life_expectancy. The association is robust to the controls but points the OPPOSITE way to individual-level evidence, so it is a statement about how the outcome is measured or certified in these places, not about risk. SIGN FLIP versus individual-level literature (hypertension raises risk): read as a statement about measurement or place, not about risk. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: pct_hypertension is modelled (NCD-RisC); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). Both variables are modelled: this coefficient compares two models, not two measurements. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-10", "level": "country", "kind": "exposure", "x": { "name": "pct_smoking", "unit": "% of population", "year": "latest per country, mostly 2024", "source": "WHO Global Health Observatory (WHO tobacco trends estimates) via data/panel_country.csv", "tag": "modelled/derived" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 165, "n_unit": "countries", "x_filled": 166, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.1864, "p": 0.0165, "ci95": [ 0.0346, 0.3298 ] }, "spearman": { "rho": 0.4003, "p": 1e-07 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": -0.2589, "p": 0.000785, "n": 165 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.1861, "p": 0.017, "n": 164 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.1836, "p": 0.0186, "n": 164 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "smoking raises risk" }, "confounders_named": [ "age structure", "GDP", "competing mortality" ], "note": "A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 0.0189, "classification": "sign_reverses", "verdict": "Raw weak positive correlation r=0.19 (95% CI 0.03 to 0.33), Spearman rho=0.40, n=165 countries, p=0.0165, BH q=0.019. SIGN REVERSES under control: partial r=-0.259 has the opposite sign to raw r=0.186: the raw association was confounded in the other direction. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: pct_smoking is modelled/derived; dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). Both variables are modelled: this coefficient compares two models, not two measurements. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-11", "level": "country", "kind": "exposure", "x": { "name": "pct_physical_inactivity", "unit": "% of adults 18+", "year": "latest per country, mostly 2022", "source": "WHO Global Health Observatory via data/panel_country.csv", "tag": "modelled/derived" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 191, "n_unit": "countries", "x_filled": 193, "y_filled": 192, "frame_rows": 217, "pearson": { "r": -0.0156, "p": 0.83, "ci95": [ -0.1572, 0.1266 ] }, "spearman": { "rho": 0.1383, "p": 0.0564 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": -0.1549, "p": 0.0324, "n": 191 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.18, "p": 0.0129, "n": 190 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.182, "p": 0.012, "n": 190 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "inactivity raises risk" }, "confounders_named": [ "age structure", "GDP" ], "note": "A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 0.83, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.02 (95% CI -0.16 to 0.13), Spearman rho=0.14, n=191 countries, p=0.83, BH q=0.83. NULL: raw association not distinguishable from zero (BH q=0.83 within level). Absence of a place-level correlation is not evidence of no individual-level effect. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: pct_physical_inactivity is modelled/derived; dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). Both variables are modelled: this coefficient compares two models, not two measurements. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-12", "level": "country", "kind": "exposure", "x": { "name": "pct_alcohol_use_disorder", "unit": "% of population 15+", "year": "latest per country, mostly 2016", "source": "WHO Global Health Observatory (Global Status Report on Alcohol and Health) via data/panel_country.csv", "tag": "measured/administrative" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 188, "n_unit": "countries", "x_filled": 188, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.2942, "p": 4.16e-05, "ci95": [ 0.1578, 0.4197 ] }, "spearman": { "rho": 0.3063, "p": 1.91e-05 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": -0.1022, "p": 0.163, "n": 188 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.0861, "p": 0.242, "n": 187 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.0962, "p": 0.19, "n": 187 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "positive", "why": "heavy alcohol raises risk" }, "confounders_named": [ "age structure", "GDP", "2016 vintage" ], "note": "A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 5.37e-05, "classification": "collapses", "verdict": "Raw weak positive correlation r=0.29 (95% CI 0.16 to 0.42), Spearman rho=0.31, n=188 countries, p=4.16e-05, BH q=5.4e-05. COLLAPSES: partial r=-0.102 is under half the raw r=0.294 once pct_65plus is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_alcohol_use_disorder. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: pct_alcohol_use_disorder is measured/administrative; dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-13", "level": "country", "kind": "exposure", "x": { "name": "mean_bmi", "unit": "kg/m^2", "year": "latest per country, mostly 2016", "source": "WHO Global Health Observatory (NCD-RisC) via data/panel_country.csv", "tag": "modelled (NCD-RisC)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 188, "n_unit": "countries", "x_filled": 188, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.1587, "p": 0.0296, "ci95": [ 0.016, 0.2951 ] }, "spearman": { "rho": 0.2609, "p": 0.000299 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": -0.0441, "p": 0.548, "n": 188 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.1232, "p": 0.093, "n": 187 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.1146, "p": 0.118, "n": 187 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure", "GDP" ], "note": "A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 0.0328, "classification": "collapses", "verdict": "Raw weak positive correlation r=0.16 (95% CI 0.02 to 0.30), Spearman rho=0.26, n=188 countries, p=0.0296, BH q=0.033. COLLAPSES: partial r=-0.044 is under half the raw r=0.159 once pct_65plus is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with mean_bmi. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: mean_bmi is modelled (NCD-RisC); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). Both variables are modelled: this coefficient compares two models, not two measurements. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-14", "level": "country", "kind": "exposure", "x": { "name": "pct_air_pollution_exposure", "unit": "% of total population", "year": "2017 for all countries (single global vintage)", "source": "World Bank (based on satellite-derived PM2.5 modeling, Brauer et al./Global Burden of Disease-adjacent methodology but published directly by WB, not IHME) via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 186, "n_unit": "countries", "x_filled": 194, "y_filled": 192, "frame_rows": 217, "pearson": { "r": -0.6411, "p": 6.45e-23, "ci95": [ -0.7187, -0.5477 ] }, "spearman": { "rho": -0.4898, "p": 1.3e-12 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": -0.5562, "p": 1.7e-16, "n": 186 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.427, "p": 1.35e-09, "n": 185 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.4289, "p": 1.13e-09, "n": 185 }, "divergence_flag": false, "sign_flip_vs_literature": true, "literature_expectation": { "direction": "positive", "why": "PM2.5 raises risk" }, "confounders_named": [ "age structure (polluted countries are young)", "GDP" ], "note": "Baseline I5 sign flip re-tested. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 2.5e-22, "classification": "survives_sign_flipped", "verdict": "Raw strong negative correlation r=-0.64 (95% CI -0.72 to -0.55), Spearman rho=-0.49, n=186 countries, p=6.45e-23, BH q=2.5e-22. SURVIVES BUT SIGN-FLIPPED: partial r=-0.556 (p=1.7e-16) controlling for pct_65plus; extended partial r=-0.427; strict partial r=-0.4289 controlling for pct_65plus+gdp_per_capita_usd+life_expectancy. The association is robust to the controls but points the OPPOSITE way to individual-level evidence, so it is a statement about how the outcome is measured or certified in these places, not about risk. SIGN FLIP versus individual-level literature (PM2.5 raises risk): read as a statement about measurement or place, not about risk. Baseline I5 sign flip re-tested. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: pct_air_pollution_exposure is administrative compilation (World Bank); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-15", "level": "country", "kind": "exposure", "x": { "name": "pm25_avg_ugm3", "unit": "micrograms per cubic meter (ug/m3)", "year": "latest per country, mostly 2023", "source": "WHO Global Health Observatory (WHO Ambient Air Quality Database) via data/panel_country.csv", "tag": "modelled/derived" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 192, "n_unit": "countries", "x_filled": 212, "y_filled": 192, "frame_rows": 217, "pearson": { "r": -0.3678, "p": 1.54e-07, "ci95": [ -0.4842, -0.2386 ] }, "spearman": { "rho": -0.4806, "p": 1.74e-12 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": 0.0201, "p": 0.782, "n": 192 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": 0.0308, "p": 0.672, "n": 191 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": 0.0411, "p": 0.572, "n": 191 }, "divergence_flag": false, "sign_flip_vs_literature": true, "literature_expectation": { "direction": "positive", "why": "PM2.5 raises risk" }, "confounders_named": [ "age structure", "GDP" ], "note": "A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 2.81e-07, "classification": "collapses", "verdict": "Raw moderate negative correlation r=-0.37 (95% CI -0.48 to -0.24), Spearman rho=-0.48, n=192 countries, p=1.54e-07, BH q=2.8e-07. COLLAPSES: partial r=0.020 is under half the raw r=-0.368 once pct_65plus is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pm25_avg_ugm3. SIGN FLIP versus individual-level literature (PM2.5 raises risk): read as a statement about measurement or place, not about risk. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: pm25_avg_ugm3 is modelled/derived; dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). Both variables are modelled: this coefficient compares two models, not two measurements. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-16", "level": "country", "kind": "exposure", "x": { "name": "uhc_service_coverage_index", "unit": "index, 0-100", "year": "latest per country, mostly 2023", "source": "WHO Global Health Observatory via data/panel_country.csv", "tag": "modelled/derived" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 192, "n_unit": "countries", "x_filled": 193, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.4709, "p": 5.47e-12, "ci95": [ 0.3528, 0.5742 ] }, "spearman": { "rho": 0.6543, "p": 7.73e-25 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": -0.0397, "p": 0.584, "n": 192 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.2333, "p": 0.00116, "n": 191 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.1334, "p": 0.0657, "n": 191 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure", "GDP", "death certification" ], "note": "A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 1.3e-11, "classification": "collapses", "verdict": "Raw moderate positive correlation r=0.47 (95% CI 0.35 to 0.57), Spearman rho=0.65, n=192 countries, p=5.47e-12, BH q=1.3e-11. COLLAPSES: partial r=-0.040 is under half the raw r=0.471 once pct_65plus is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with uhc_service_coverage_index. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: uhc_service_coverage_index is modelled/derived; dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). Both variables are modelled: this coefficient compares two models, not two measurements. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-17", "level": "country", "kind": "exposure", "x": { "name": "hospital_beds_per_1000", "unit": "per 1,000 population", "year": "latest per country, 1970-2023", "source": "World Bank via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 191, "n_unit": "countries", "x_filled": 201, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.5364, "p": 1.27e-15, "ci95": [ 0.4269, 0.6304 ] }, "spearman": { "rho": 0.5384, "p": 9.56e-16 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": 0.1258, "p": 0.0828, "n": 191 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.0109, "p": 0.882, "n": 190 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.0465, "p": 0.524, "n": 190 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure", "GDP" ], "note": "A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 3.58e-15, "classification": "collapses", "verdict": "Raw strong positive correlation r=0.54 (95% CI 0.43 to 0.63), Spearman rho=0.54, n=191 countries, p=1.27e-15, BH q=3.6e-15. COLLAPSES: partial r=0.126 is under half the raw r=0.536 once pct_65plus is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with hospital_beds_per_1000. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: hospital_beds_per_1000 is administrative compilation (World Bank); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-18", "level": "country", "kind": "exposure", "x": { "name": "out_of_pocket_pct_health", "unit": "%", "year": "latest per country, 2021-2024", "source": "World Bank / WHO GHED via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 191, "n_unit": "countries", "x_filled": 193, "y_filled": 192, "frame_rows": 217, "pearson": { "r": -0.314, "p": 9.73e-06, "ci95": [ -0.4365, -0.18 ] }, "spearman": { "rho": -0.3164, "p": 8.2e-06 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": -0.1777, "p": 0.0139, "n": 191 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": 0.0082, "p": 0.911, "n": 191 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.0136, "p": 0.852, "n": 191 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure", "GDP" ], "note": "A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 1.32e-05, "classification": "collapses_extended", "verdict": "Raw moderate negative correlation r=-0.31 (95% CI -0.44 to -0.18), Spearman rho=-0.32, n=191 countries, p=9.73e-06, BH q=1.3e-05. COLLAPSES under the extended controls: survives the primary control (partial r=-0.178) but falls to 0.008 with the extended set pct_65plus+gdp_per_capita_usd. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: out_of_pocket_pct_health is administrative compilation (World Bank); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-19", "level": "country", "kind": "exposure", "x": { "name": "gini", "unit": "index, 0-100", "year": "latest per country, 1992-2025", "source": "World Bank (mostly PovcalNet/household survey based) via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 169, "n_unit": "countries", "x_filled": 171, "y_filled": 192, "frame_rows": 217, "pearson": { "r": -0.3329, "p": 9.76e-06, "ci95": [ -0.4607, -0.1916 ] }, "spearman": { "rho": -0.3613, "p": 1.39e-06 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": -0.0701, "p": 0.365, "n": 169 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.0312, "p": 0.687, "n": 169 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.0326, "p": 0.674, "n": 169 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure", "GDP", "survey vintage" ], "note": "Baseline J1. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 1.32e-05, "classification": "collapses", "verdict": "Raw moderate negative correlation r=-0.33 (95% CI -0.46 to -0.19), Spearman rho=-0.36, n=169 countries, p=9.76e-06, BH q=1.3e-05. COLLAPSES: partial r=-0.070 is under half the raw r=-0.333 once pct_65plus is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with gini. Baseline J1. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: gini is administrative compilation (World Bank); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-20", "level": "country", "kind": "exposure", "x": { "name": "fertility_rate", "unit": "births per woman", "year": "latest per country, mostly 2024", "source": "World Bank via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 192, "n_unit": "countries", "x_filled": 217, "y_filled": 192, "frame_rows": 217, "pearson": { "r": -0.3921, "p": 1.86e-08, "ci95": [ -0.5056, -0.2652 ] }, "spearman": { "rho": -0.6785, "p": 2.99e-27 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": 0.2485, "p": 0.000511, "n": 192 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": 0.307, "p": 1.56e-05, "n": 191 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": 0.2364, "p": 0.000993, "n": 191 }, "divergence_flag": true, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure (fertility drives it)" ], "note": "Fertility is the demographic engine of age structure; expect it to vanish once pct_65plus is controlled. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 3.6e-08, "classification": "sign_reverses", "verdict": "Raw moderate negative correlation r=-0.39 (95% CI -0.51 to -0.27), Spearman rho=-0.68, n=192 countries, p=1.86e-08, BH q=3.6e-08. Pearson and Spearman disagree by more than 0.25: non-linear or outlier-driven; neither number should be read alone. SIGN REVERSES under control: partial r=0.248 has the opposite sign to raw r=-0.392: the raw association was confounded in the other direction. Fertility is the demographic engine of age structure; expect it to vanish once pct_65plus is controlled. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: fertility_rate is administrative compilation (World Bank); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-21", "level": "country", "kind": "exposure", "x": { "name": "pct_safe_water_access", "unit": "% of population", "year": "latest per country, 2016-2024", "source": "World Bank (WHO/UNICEF Joint Monitoring Programme) via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 191, "n_unit": "countries", "x_filled": 214, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.3363, "p": 1.97e-06, "ci95": [ 0.2041, 0.4565 ] }, "spearman": { "rho": 0.5957, "p": 9.86e-20 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": -0.1302, "p": 0.0725, "n": 191 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.2264, "p": 0.00168, "n": 190 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.133, "p": 0.0673, "n": 190 }, "divergence_flag": true, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure", "GDP" ], "note": "A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 3.05e-06, "classification": "collapses", "verdict": "Raw moderate positive correlation r=0.34 (95% CI 0.20 to 0.46), Spearman rho=0.60, n=191 countries, p=1.97e-06, BH q=3.1e-06. Pearson and Spearman disagree by more than 0.25: non-linear or outlier-driven; neither number should be read alone. COLLAPSES: partial r=-0.130 is under half the raw r=0.336 once pct_65plus is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with pct_safe_water_access. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: pct_safe_water_access is administrative compilation (World Bank); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-22", "level": "country", "kind": "exposure", "x": { "name": "measles_immunization_pct", "unit": "%", "year": "latest per country, mostly 2023-2024", "source": "World Bank / WHO-UNICEF Estimates of National Immunization Coverage (WUENIC) via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 192, "n_unit": "countries", "x_filled": 193, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.2209, "p": 0.00208, "ci95": [ 0.0819, 0.3515 ] }, "spearman": { "rho": 0.3203, "p": 5.91e-06 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": -0.0114, "p": 0.875, "n": 192 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": -0.1248, "p": 0.0855, "n": 191 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": -0.0401, "p": 0.582, "n": 191 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure", "GDP" ], "note": "Baseline J5; a generic health-system proxy with no causal link to dementia. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 0.00248, "classification": "collapses", "verdict": "Raw weak positive correlation r=0.22 (95% CI 0.08 to 0.35), Spearman rho=0.32, n=192 countries, p=0.00208, BH q=0.0025. COLLAPSES: partial r=-0.011 is under half the raw r=0.221 once pct_65plus is controlled: the raw association mostly measures the confounder. Treat as an artefact of the control variable, not a relationship with measles_immunization_pct. Baseline J5; a generic health-system proxy with no causal link to dementia. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: measles_immunization_pct is administrative compilation (World Bank); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KP-23", "level": "country", "kind": "exposure", "x": { "name": "dementia_plan_exists", "unit": "1/0", "year": "2017 (WHO Global Dementia Observatory baseline survey wave)", "source": "WHO Global Dementia Observatory (GDO) via data/panel_country.csv", "tag": "measured/administrative" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 51, "n_unit": "countries", "x_filled": 51, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.6565, "p": 1.69e-07, "ci95": [ 0.465, 0.7893 ] }, "spearman": { "rho": 0.6475, "p": 2.81e-07 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": 0.3525, "p": 0.0112, "n": 51 }, "partial_extended": { "controls": [ "pct_65plus", "gdp_per_capita_usd" ], "r": 0.163, "p": 0.253, "n": 51 }, "partial_strict": { "controls": [ "pct_65plus", "gdp_per_capita_usd", "life_expectancy" ], "r": 0.164, "p": 0.25, "n": 51 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "age structure", "GDP", "only 51 respondents (self-selected)" ], "note": "Baseline I4. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds.", "fdr_q_within_level": 2.91e-07, "classification": "collapses_extended", "verdict": "Raw strong positive correlation r=0.66 (95% CI 0.47 to 0.79), Spearman rho=0.65, n=51 countries, p=1.69e-07, BH q=2.9e-07. COLLAPSES under the extended controls: survives the primary control (partial r=0.352) but falls to 0.163 with the extended set pct_65plus+gdp_per_capita_usd. Baseline I4. A linear partial on pct_65plus is NOT age-standardisation: the crude rate depends non-linearly on the 80+ share, which no country file holds. Provenance: dementia_plan_exists is measured/administrative; dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KU-1", "level": "country", "kind": "uninterpretable", "x": { "name": "life_expectancy", "unit": "years", "year": "latest per country, mostly 2024", "source": "World Bank via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_deaths_per_100k_age_std", "unit": "deaths per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 192, "n_unit": "countries", "x_filled": 217, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.563, "p": 1.9e-17, "ci95": [ 0.4579, 0.6526 ] }, "spearman": { "rho": 0.6755, "p": 6.22e-27 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": 0.0727, "p": 0.316, "n": 192 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "competing risks: you must survive other causes to die of dementia", "age structure" ], "note": "Dementia deaths rise with life expectancy because people who do not die of infection or heart disease at 60 live to die of dementia at 85. A correlation here is a competing-risks artefact and cannot be read as life expectancy 'causing' dementia.", "fdr_q_within_level": 5.89e-17, "classification": "uninterpretable", "verdict": "Raw strong positive correlation r=0.56 (95% CI 0.46 to 0.65), Spearman rho=0.68, n=192 countries, p=1.9e-17, BH q=5.9e-17. UNINTERPRETABLE: flagged a priori: competing risks: you must survive other causes to die of dementia. Documented so that nobody publishes it as a finding. Dementia deaths rise with life expectancy because people who do not die of infection or heart disease at 60 live to die of dementia at 85. A correlation here is a competing-risks artefact and cannot be read as life expectancy 'causing' dementia. Provenance: life_expectancy is administrative compilation (World Bank); dementia_deaths_per_100k_age_std is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KU-2", "level": "country", "kind": "uninterpretable", "x": { "name": "life_expectancy", "unit": "years", "year": "latest per country, mostly 2024", "source": "World Bank via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "dementia_daly_per_100k", "unit": "DALYs per 100,000 population", "year": "latest per country, 2000-2021 (most countries: 2021)", "source": "WHO Global Health Estimates (GHE) 2021 update, via Our World in Data indicator export via data/panel_country.csv", "tag": "modelled (WHO GHE cause-of-death ensemble)" }, "n": 192, "n_unit": "countries", "x_filled": 217, "y_filled": 192, "frame_rows": 217, "pearson": { "r": 0.6395, "p": 1.81e-23, "ci95": [ 0.5475, 0.7162 ] }, "spearman": { "rho": 0.7617, "p": 1.17e-37 }, "partial_primary": { "controls": [ "pct_65plus" ], "r": 0.0601, "p": 0.408, "n": 192 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "competing risks", "age structure" ], "note": "Same competing-risks structure with DALYs.", "fdr_q_within_level": 8.02e-23, "classification": "uninterpretable", "verdict": "Raw strong positive correlation r=0.64 (95% CI 0.55 to 0.72), Spearman rho=0.76, n=192 countries, p=1.81e-23, BH q=8e-23. UNINTERPRETABLE: flagged a priori: competing risks. Documented so that nobody publishes it as a finding. Same competing-risks structure with DALYs. Provenance: life_expectancy is administrative compilation (World Bank); dementia_daly_per_100k is modelled (WHO GHE cause-of-death ensemble). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KT-1", "level": "country", "kind": "artefact_test", "x": { "name": "ts_pop65_change_2000_2021_points", "unit": "change 2000-2021 in percent of population aged 65+, percentage points", "year": "2000-2021", "source": "World Bank via data/panel_country_timeseries.csv", "tag": "measured/administrative (derived change)" }, "y": { "name": "ts_crude_rate_change_2000_2021_pct", "unit": "percent change 2000-2021 in crude dementia death rate", "year": "2000-2021", "source": "computed from data/panel_country_timeseries.csv", "tag": "modelled numerator (derived change)" }, "n": 143, "n_unit": "countries", "x_filled": 143, "y_filled": 143, "frame_rows": 217, "pearson": { "r": 0.2972, "p": 0.000313, "ci95": [ 0.1398, 0.4399 ] }, "spearman": { "rho": 0.7525, "p": 2.38e-27 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": true, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "ageing drives the crude rate" ], "note": "Within-country change 2000-2021: does the death rate rise where the population aged?", "fdr_q_within_level": 0.000388, "classification": "artefact_documented", "verdict": "Raw weak positive correlation r=0.30 (95% CI 0.14 to 0.44), Spearman rho=0.75, n=143 countries, p=0.000313, BH q=0.00039. Pearson and Spearman disagree by more than 0.25: non-linear or outlier-driven; neither number should be read alone. STRUCTURAL DEPENDENCE DOCUMENTED: r=0.297, q=0.00039: documents the mechanical/structural dependence named. Within-country change 2000-2021: does the death rate rise where the population aged? Provenance: ts_pop65_change_2000_2021_points is measured/administrative (derived change); ts_crude_rate_change_2000_2021_pct is modelled numerator (derived change). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KT-2", "level": "country", "kind": "exposure", "x": { "name": "gdp_per_capita_usd", "unit": "current US$", "year": "latest per country, 2011-2025", "source": "World Bank via data/panel_country.csv", "tag": "administrative compilation (World Bank)" }, "y": { "name": "ts_crude_rate_change_2000_2021_pct", "unit": "percent change 2000-2021 in crude dementia death rate", "year": "2000-2021", "source": "computed from data/panel_country_timeseries.csv", "tag": "modelled numerator (derived change)" }, "n": 142, "n_unit": "countries", "x_filled": 214, "y_filled": 143, "frame_rows": 217, "pearson": { "r": 0.067, "p": 0.428, "ci95": [ -0.0988, 0.2292 ] }, "spearman": { "rho": 0.559, "p": 4.84e-13 }, "partial_primary": { "controls": [ "ts_pop65_change_2000_2021_points" ], "r": -0.1077, "p": 0.202, "n": 142 }, "partial_extended": null, "partial_strict": null, "divergence_flag": true, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "ageing", "improving death certification in developing countries" ], "note": "", "fdr_q_within_level": 0.458, "classification": "null", "verdict": "Raw negligible positive correlation r=0.07 (95% CI -0.10 to 0.23), Spearman rho=0.56, n=142 countries, p=0.428, BH q=0.46. Pearson and Spearman disagree by more than 0.25: non-linear or outlier-driven; neither number should be read alone. NULL: raw association not distinguishable from zero (BH q=0.46 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Provenance: gdp_per_capita_usd is administrative compilation (World Bank); ts_crude_rate_change_2000_2021_pct is modelled numerator (derived change). One variable is modelled and one measured: they are mixed in this coefficient. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "KT-3", "level": "country", "kind": "artefact_test", "x": { "name": "ts_age_std_rate_change_2000_2021_pct", "unit": "percent change 2000-2021 in the upstream 'deaths_per_100k' series", "year": "2000-2021", "source": "computed from data/panel_country_timeseries.csv", "tag": "modelled (derived change)" }, "y": { "name": "ts_crude_rate_change_2000_2021_pct", "unit": "percent change 2000-2021 in crude dementia death rate", "year": "2000-2021", "source": "computed from data/panel_country_timeseries.csv", "tag": "modelled numerator (derived change)" }, "n": 143, "n_unit": "countries", "x_filled": 143, "y_filled": 143, "frame_rows": 217, "pearson": { "r": 0.9984, "p": 6.91e-178, "ci95": [ 0.9978, 0.9989 ] }, "spearman": { "rho": 0.9967, "p": 1.12e-155 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "Twin check on the change series: if the 'age-standardised' change equals the crude change, standardisation never happened.", "fdr_q_within_level": 1.07e-176, "classification": "artefact_documented", "verdict": "Raw strong positive correlation r=1.00 (95% CI 1.00 to 1.00), Spearman rho=1.00, n=143 countries, p=6.91e-178, BH q=1.1e-176. STRUCTURAL DEPENDENCE DOCUMENTED: r=0.998, q=1.1e-176: documents the mechanical/structural dependence named. Twin check on the change series: if the 'age-standardised' change equals the crude change, standardisation never happened. Provenance: ts_age_std_rate_change_2000_2021_pct is modelled (derived change); ts_crude_rate_change_2000_2021_pct is modelled numerator (derived change). Both variables are modelled: this coefficient compares two models, not two measurements. Outcome is a MODELLED estimate, not a count; a correlation with a variable that feeds the model is partly circular. Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk.", "caveats": [ "Ecological (place-level) correlation: it describes counties/states/countries, not people, and cannot show that the exposure raises or lowers any individual's risk." ] }, { "id": "LS-1", "level": "study", "kind": "artefact_test", "x": { "name": "years_since_pub", "unit": "2026 minus publication year", "year": "1975-2026", "source": "data/research.json", "tag": "curated metadata (derived)" }, "y": { "name": "log10_icite_citations", "unit": "log10 of raw citation count (iCite)", "year": "as of 2026-09", "source": "data/citations_gaps.json", "tag": "measured count (bibliometric)" }, "n": 6915, "n_unit": "studies", "x_filled": 14775, "y_filled": 6915, "frame_rows": 14779, "pearson": { "r": 0.3207, "p": 3.35e-165, "ci95": [ 0.2994, 0.3417 ] }, "spearman": { "rho": 0.3538, "p": 5.51e-203 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "citation accumulation time" ], "note": "Raw citations grow with time since publication; this is the artefact RCR is designed to remove.", "fdr_q_within_level": 1.26e-164, "classification": "artefact_documented", "verdict": "Raw moderate positive correlation r=0.32 (95% CI 0.30 to 0.34), Spearman rho=0.35, n=6915 studies, p=3.35e-165, BH q=1.3e-164. STRUCTURAL DEPENDENCE DOCUMENTED: r=0.321, q=1.3e-164: documents the mechanical/structural dependence named. Raw citations grow with time since publication; this is the artefact RCR is designed to remove. Provenance: years_since_pub is curated metadata (derived); log10_icite_citations is measured count (bibliometric). Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LS-2", "level": "study", "kind": "artefact_test", "x": { "name": "years_since_pub", "unit": "2026 minus publication year", "year": "1975-2026", "source": "data/research.json", "tag": "curated metadata (derived)" }, "y": { "name": "log10_rcr", "unit": "log10 of NIH iCite Relative Citation Ratio (field- and year-normalised influence)", "year": "as of 2026-09", "source": "data/citations_gaps.json (iCite) joined by PMID/DOI", "tag": "modelled metric (iCite normalisation)" }, "n": 6828, "n_unit": "studies", "x_filled": 14775, "y_filled": 6828, "frame_rows": 14779, "pearson": { "r": -0.0541, "p": 7.78e-06, "ci95": [ -0.0777, -0.0304 ] }, "spearman": { "rho": -0.0288, "p": 0.0172 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "field/year normalisation" ], "note": "RCR is year-normalised, so near zero is the expected result; a residual r shows how the library's older papers were selected (survivorship).", "fdr_q_within_level": 1.06e-05, "classification": "artefact_documented", "verdict": "Raw negligible negative correlation r=-0.05 (95% CI -0.08 to -0.03), Spearman rho=-0.03, n=6828 studies, p=7.78e-06, BH q=1.1e-05. STRUCTURAL DEPENDENCE DOCUMENTED: r=-0.054, q=1.1e-05: documents the mechanical/structural dependence named. RCR is year-normalised, so near zero is the expected result; a residual r shows how the library's older papers were selected (survivorship). Provenance: years_since_pub is curated metadata (derived); log10_rcr is modelled metric (iCite normalisation). One variable is modelled and one measured: they are mixed in this coefficient. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LS-3", "level": "study", "kind": "exposure", "x": { "name": "log10_n_participants", "unit": "log10 of extracted participant count (numeric field only)", "year": "per study", "source": "data/research.json n_participants", "tag": "curated/extracted metadata" }, "y": { "name": "log10_rcr", "unit": "log10 of NIH iCite Relative Citation Ratio (field- and year-normalised influence)", "year": "as of 2026-09", "source": "data/citations_gaps.json (iCite) joined by PMID/DOI", "tag": "modelled metric (iCite normalisation)" }, "n": 2870, "n_unit": "studies", "x_filled": 5888, "y_filled": 6828, "frame_rows": 14779, "pearson": { "r": 0.1894, "p": 1.36e-24, "ci95": [ 0.1539, 0.2245 ] }, "spearman": { "rho": 0.1958, "p": 3.45e-26 }, "partial_primary": { "controls": [ "year" ], "r": 0.2063, "p": 5.95e-29, "n": 2870 }, "partial_extended": { "controls": [ "year", "is_review_or_meta" ], "r": 0.1163, "p": 2.56e-05, "n": 1303 }, "partial_strict": { "controls": [ "year", "journal_tier", "is_review_or_meta" ], "r": 0.1238, "p": 7.46e-06, "n": 1301 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "study design (meta-analyses have huge n and high RCR)", "field", "journal" ], "note": "Do bigger studies get more field-normalised influence?", "fdr_q_within_level": 2.91e-24, "classification": "survives", "verdict": "Raw weak positive correlation r=0.19 (95% CI 0.15 to 0.22), Spearman rho=0.20, n=2870 studies, p=1.36e-24, BH q=2.9e-24. SURVIVES the confounder test: partial r=0.206 (p=5.95e-29) controlling for year; extended partial r=0.1163; strict partial r=0.1238 controlling for year+journal_tier+is_review_or_meta. A place-level association not explained by the named control(s); candidate third things still unaddressed: study design (meta-analyses have huge n and high RCR), field, journal. Do bigger studies get more field-normalised influence? Provenance: log10_n_participants is curated/extracted metadata; log10_rcr is modelled metric (iCite normalisation). One variable is modelled and one measured: they are mixed in this coefficient. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LS-4", "level": "study", "kind": "exposure", "x": { "name": "year", "unit": "publication year", "year": "1975-2026", "source": "data/research.json", "tag": "curated metadata" }, "y": { "name": "log10_n_participants", "unit": "log10 of extracted participant count (numeric field only)", "year": "per study", "source": "data/research.json n_participants", "tag": "curated/extracted metadata" }, "n": 5887, "n_unit": "studies", "x_filled": 14775, "y_filled": 5888, "frame_rows": 14779, "pearson": { "r": 0.1861, "p": 4.92e-47, "ci95": [ 0.1613, 0.2107 ] }, "spearman": { "rho": 0.1605, "p": 2.79e-35 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "biobank era", "extraction bias (n only extracted for 4,014 records)" ], "note": "Are the library's studies getting bigger over time?", "fdr_q_within_level": 1.23e-46, "classification": "raw_only", "verdict": "Raw weak positive correlation r=0.19 (95% CI 0.16 to 0.21), Spearman rho=0.16, n=5887 studies, p=4.92e-47, BH q=1.2e-46. RAW ONLY: significant raw association but no confounder could be partialled (no control named or insufficient overlap). Are the library's studies getting bigger over time? Provenance: year is curated metadata; log10_n_participants is curated/extracted metadata. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LS-5", "level": "study", "kind": "exposure", "x": { "name": "evidence_ordinal", "unit": "site's evidence_strength label as ordinal: low/preliminary/early=1, moderate=2, high=3 (null-result excluded)", "year": "per study", "source": "data/research.json evidence_strength", "tag": "curated label (editorial judgement)" }, "y": { "name": "log10_rcr", "unit": "log10 of NIH iCite Relative Citation Ratio (field- and year-normalised influence)", "year": "as of 2026-09", "source": "data/citations_gaps.json (iCite) joined by PMID/DOI", "tag": "modelled metric (iCite normalisation)" }, "n": 6818, "n_unit": "studies", "x_filled": 14753, "y_filled": 6828, "frame_rows": 14779, "pearson": { "r": -0.0493, "p": 4.59e-05, "ci95": [ -0.073, -0.0256 ] }, "spearman": { "rho": -0.0283, "p": 0.0195 }, "partial_primary": { "controls": [ "year" ], "r": -0.0599, "p": 7.58e-07, "n": 6818 }, "partial_extended": null, "partial_strict": { "controls": [ "year", "journal_tier", "is_review_or_meta" ], "r": -0.1189, "p": 5.01e-08, "n": 2088 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "editorial judgement was made knowing the journal", "design" ], "note": "Does the site's evidence label track citation influence?", "fdr_q_within_level": 5.74e-05, "classification": "negligible", "verdict": "Raw negligible negative correlation r=-0.05 (95% CI -0.07 to -0.03), Spearman rho=-0.03, n=6818 studies, p=4.59e-05, BH q=5.7e-05. NEGLIGIBLE: raw r=-0.049 is statistically non-zero at n=6818 but below 0.10 in size; too small to interpret. Not a finding worth reporting as an association. Does the site's evidence label track citation influence? Provenance: evidence_ordinal is curated label (editorial judgement); log10_rcr is modelled metric (iCite normalisation). One variable is modelled and one measured: they are mixed in this coefficient. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LS-6", "level": "study", "kind": "exposure", "x": { "name": "evidence_ordinal", "unit": "site's evidence_strength label as ordinal: low/preliminary/early=1, moderate=2, high=3 (null-result excluded)", "year": "per study", "source": "data/research.json evidence_strength", "tag": "curated label (editorial judgement)" }, "y": { "name": "log10_n_participants", "unit": "log10 of extracted participant count (numeric field only)", "year": "per study", "source": "data/research.json n_participants", "tag": "curated/extracted metadata" }, "n": 5865, "n_unit": "studies", "x_filled": 14753, "y_filled": 5888, "frame_rows": 14779, "pearson": { "r": 0.1072, "p": 1.9e-16, "ci95": [ 0.0818, 0.1324 ] }, "spearman": { "rho": 0.1437, "p": 1.93e-28 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "editorial judgement uses n" ], "note": "", "fdr_q_within_level": 3.17e-16, "classification": "raw_only", "verdict": "Raw weak positive correlation r=0.11 (95% CI 0.08 to 0.13), Spearman rho=0.14, n=5865 studies, p=1.9e-16, BH q=3.2e-16. RAW ONLY: significant raw association but no confounder could be partialled (no control named or insufficient overlap). Provenance: evidence_ordinal is curated label (editorial judgement); log10_n_participants is curated/extracted metadata. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LS-7", "level": "study", "kind": "exposure", "x": { "name": "design_ordinal", "unit": "crude evidence-hierarchy rank of study_design text: 6 meta-analysis/systematic review ... 1 case report/narrative review; unrankable = blank", "year": "per study", "source": "data/research.json study_design", "tag": "curated label (editorial judgement)" }, "y": { "name": "log10_rcr", "unit": "log10 of NIH iCite Relative Citation Ratio (field- and year-normalised influence)", "year": "as of 2026-09", "source": "data/citations_gaps.json (iCite) joined by PMID/DOI", "tag": "modelled metric (iCite normalisation)" }, "n": 2103, "n_unit": "studies", "x_filled": 4234, "y_filled": 6828, "frame_rows": 14779, "pearson": { "r": 0.0792, "p": 0.000279, "ci95": [ 0.0365, 0.1215 ] }, "spearman": { "rho": 0.0887, "p": 4.65e-05 }, "partial_primary": { "controls": [ "year" ], "r": 0.0832, "p": 0.000133, "n": 2103 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "field", "journal" ], "note": "Evidence-hierarchy rank of the design vs RCR.", "fdr_q_within_level": 0.000299, "classification": "negligible", "verdict": "Raw negligible positive correlation r=0.08 (95% CI 0.04 to 0.12), Spearman rho=0.09, n=2103 studies, p=0.000279, BH q=0.0003. NEGLIGIBLE: raw r=0.079 is statistically non-zero at n=2103 but below 0.10 in size; too small to interpret. Not a finding worth reporting as an association. Evidence-hierarchy rank of the design vs RCR. Provenance: design_ordinal is curated label (editorial judgement); log10_rcr is modelled metric (iCite normalisation). One variable is modelled and one measured: they are mixed in this coefficient. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LS-8", "level": "study", "kind": "exposure", "x": { "name": "open_access", "unit": "1 = open access per the fetch pipeline, 0 = not", "year": "per study", "source": "data/research.json open_access", "tag": "measured/administrative (PubMed flag)" }, "y": { "name": "log10_rcr", "unit": "log10 of NIH iCite Relative Citation Ratio (field- and year-normalised influence)", "year": "as of 2026-09", "source": "data/citations_gaps.json (iCite) joined by PMID/DOI", "tag": "modelled metric (iCite normalisation)" }, "n": 6663, "n_unit": "studies", "x_filled": 14415, "y_filled": 6828, "frame_rows": 14779, "pearson": { "r": 0.0001, "p": 0.995, "ci95": [ -0.0239, 0.0241 ] }, "spearman": { "rho": -0.0059, "p": 0.632 }, "partial_primary": { "controls": [ "year" ], "r": -0.0268, "p": 0.0284, "n": 6663 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "funder mandates", "journal", "year" ], "note": "Open-access citation advantage within this library.", "fdr_q_within_level": 0.995, "classification": "null", "verdict": "Raw negligible positive correlation r=0.00 (95% CI -0.02 to 0.02), Spearman rho=-0.01, n=6663 studies, p=0.995, BH q=0.99. NULL: raw association not distinguishable from zero (BH q=0.99 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Open-access citation advantage within this library. Provenance: open_access is measured/administrative (PubMed flag); log10_rcr is modelled metric (iCite normalisation). One variable is modelled and one measured: they are mixed in this coefficient. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LS-9", "level": "study", "kind": "exposure", "x": { "name": "journal_tier", "unit": "site's journal tier, 1 = top tier ... 4 = lowest", "year": "per study", "source": "data/research_score.json", "tag": "curated label (site rubric)" }, "y": { "name": "log10_rcr", "unit": "log10 of NIH iCite Relative Citation Ratio (field- and year-normalised influence)", "year": "as of 2026-09", "source": "data/citations_gaps.json (iCite) joined by PMID/DOI", "tag": "modelled metric (iCite normalisation)" }, "n": 6818, "n_unit": "studies", "x_filled": 10099, "y_filled": 6828, "frame_rows": 14779, "pearson": { "r": -0.2603, "p": 4.94e-106, "ci95": [ -0.2823, -0.2381 ] }, "spearman": { "rho": -0.2295, "p": 3.53e-82 }, "partial_primary": { "controls": [ "year" ], "r": -0.2579, "p": 5.11e-104, "n": 6818 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "tier was assigned partly by reputation, which is citation-driven" ], "note": "Tier 1 = best, so a NEGATIVE r means top-tier journals have higher RCR.", "fdr_q_within_level": 1.48e-105, "classification": "survives", "verdict": "Raw weak negative correlation r=-0.26 (95% CI -0.28 to -0.24), Spearman rho=-0.23, n=6818 studies, p=4.94e-106, BH q=1.5e-105. SURVIVES the confounder test: partial r=-0.258 (p=5.11e-104) controlling for year. A place-level association not explained by the named control(s); candidate third things still unaddressed: tier was assigned partly by reputation, which is citation-driven. Tier 1 = best, so a NEGATIVE r means top-tier journals have higher RCR. Provenance: journal_tier is curated label (site rubric); log10_rcr is modelled metric (iCite normalisation). One variable is modelled and one measured: they are mixed in this coefficient. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LS-10", "level": "study", "kind": "exposure", "x": { "name": "has_effect_size", "unit": "1 = an effect_size string was extracted", "year": "per study", "source": "data/research.json", "tag": "curated metadata" }, "y": { "name": "log10_rcr", "unit": "log10 of NIH iCite Relative Citation Ratio (field- and year-normalised influence)", "year": "as of 2026-09", "source": "data/citations_gaps.json (iCite) joined by PMID/DOI", "tag": "modelled metric (iCite normalisation)" }, "n": 6828, "n_unit": "studies", "x_filled": 14779, "y_filled": 6828, "frame_rows": 14779, "pearson": { "r": 0.0441, "p": 0.000264, "ci95": [ 0.0204, 0.0678 ] }, "spearman": { "rho": 0.0344, "p": 0.00447 }, "partial_primary": { "controls": [ "year" ], "r": 0.0369, "p": 0.00231, "n": 6828 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "extraction pipeline generation" ], "note": "", "fdr_q_within_level": 0.000299, "classification": "negligible", "verdict": "Raw negligible positive correlation r=0.04 (95% CI 0.02 to 0.07), Spearman rho=0.03, n=6828 studies, p=0.000264, BH q=0.0003. NEGLIGIBLE: raw r=0.044 is statistically non-zero at n=6828 but below 0.10 in size; too small to interpret. Not a finding worth reporting as an association. Provenance: has_effect_size is curated metadata; log10_rcr is modelled metric (iCite normalisation). One variable is modelled and one measured: they are mixed in this coefficient. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LS-11", "level": "study", "kind": "exposure", "x": { "name": "is_review_or_meta", "unit": "1 = meta-analysis, systematic or narrative review; 0 = other ranked design; blank = design unknown", "year": "per study", "source": "data/research.json study_design", "tag": "curated label" }, "y": { "name": "log10_rcr", "unit": "log10 of NIH iCite Relative Citation Ratio (field- and year-normalised influence)", "year": "as of 2026-09", "source": "data/citations_gaps.json (iCite) joined by PMID/DOI", "tag": "modelled metric (iCite normalisation)" }, "n": 2103, "n_unit": "studies", "x_filled": 4244, "y_filled": 6828, "frame_rows": 14779, "pearson": { "r": 0.1861, "p": 7.8e-18, "ci95": [ 0.1445, 0.227 ] }, "spearman": { "rho": 0.2028, "p": 5.91e-21 }, "partial_primary": { "controls": [ "year" ], "r": 0.1906, "p": 1.2e-18, "n": 2103 }, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "reviews are cited more in every field" ], "note": "", "fdr_q_within_level": 1.46e-17, "classification": "survives", "verdict": "Raw weak positive correlation r=0.19 (95% CI 0.14 to 0.23), Spearman rho=0.20, n=2103 studies, p=7.8e-18, BH q=1.5e-17. SURVIVES the confounder test: partial r=0.191 (p=1.2e-18) controlling for year. A place-level association not explained by the named control(s); candidate third things still unaddressed: reviews are cited more in every field. Provenance: is_review_or_meta is curated label; log10_rcr is modelled metric (iCite normalisation). One variable is modelled and one measured: they are mixed in this coefficient. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LU-1", "level": "study", "kind": "uninterpretable", "x": { "name": "evidence_ordinal", "unit": "site's evidence_strength label as ordinal: low/preliminary/early=1, moderate=2, high=3 (null-result excluded)", "year": "per study", "source": "data/research.json evidence_strength", "tag": "curated label (editorial judgement)" }, "y": { "name": "site_score", "unit": "alzhub's own 0-100 study usefulness score (journal 25, design 25, evidence 20, validation 15, recency 10, size 5)", "year": "per study", "source": "data/research_score.json", "tag": "derived index (site rubric; built FROM evidence/design/journal/year/size fields)" }, "n": 10073, "n_unit": "studies", "x_filled": 14753, "y_filled": 10099, "frame_rows": 14779, "pearson": { "r": 0.8227, "p": 0.0, "ci95": [ 0.8163, 0.8289 ] }, "spearman": { "rho": 0.7935, "p": 0.0 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "the score is computed FROM the evidence label (20 of 100 points)" ], "note": "Circular by construction: not a finding.", "fdr_q_within_level": 0.0, "classification": "uninterpretable", "verdict": "Raw strong positive correlation r=0.82 (95% CI 0.82 to 0.83), Spearman rho=0.79, n=10073 studies, p=0.0, BH q=0. UNINTERPRETABLE: flagged a priori: the score is computed FROM the evidence label (20 of 100 points). Documented so that nobody publishes it as a finding. Circular by construction: not a finding. Provenance: evidence_ordinal is curated label (editorial judgement); site_score is derived index (site rubric; built FROM evidence/design/journal/year/size fields). Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LU-2", "level": "study", "kind": "exposure", "x": { "name": "site_score", "unit": "alzhub's own 0-100 study usefulness score (journal 25, design 25, evidence 20, validation 15, recency 10, size 5)", "year": "per study", "source": "data/research_score.json", "tag": "derived index (site rubric; built FROM evidence/design/journal/year/size fields)" }, "y": { "name": "log10_rcr", "unit": "log10 of NIH iCite Relative Citation Ratio (field- and year-normalised influence)", "year": "as of 2026-09", "source": "data/citations_gaps.json (iCite) joined by PMID/DOI", "tag": "modelled metric (iCite normalisation)" }, "n": 6818, "n_unit": "studies", "x_filled": 10099, "y_filled": 6828, "frame_rows": 14779, "pearson": { "r": 0.0842, "p": 3.33e-12, "ci95": [ 0.0606, 0.1077 ] }, "spearman": { "rho": 0.1015, "p": 4.58e-17 }, "partial_primary": { "controls": [ "year" ], "r": 0.0705, "p": 5.49e-09, "n": 6818 }, "partial_extended": null, "partial_strict": { "controls": [ "year", "journal_tier", "is_review_or_meta" ], "r": -0.1112, "p": 3.31e-07, "n": 2098 }, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "recency is 10 points of the score and RCR is year-normalised", "journal tier" ], "note": "Does the site's own rubric track external influence?", "fdr_q_within_level": 4.99e-12, "classification": "negligible", "verdict": "Raw negligible positive correlation r=0.08 (95% CI 0.06 to 0.11), Spearman rho=0.10, n=6818 studies, p=3.33e-12, BH q=5e-12. NEGLIGIBLE: raw r=0.084 is statistically non-zero at n=6818 but below 0.10 in size; too small to interpret. Not a finding worth reporting as an association. Does the site's own rubric track external influence? Provenance: site_score is derived index (site rubric; built FROM evidence/design/journal/year/size fields); log10_rcr is modelled metric (iCite normalisation). One variable is modelled and one measured: they are mixed in this coefficient. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LM-1", "level": "study", "kind": "measured_vs_modelled", "x": { "name": "log10_icite_citations", "unit": "log10 of raw citation count (iCite)", "year": "as of 2026-09", "source": "data/citations_gaps.json", "tag": "measured count (bibliometric)" }, "y": { "name": "log10_openalex_citations", "unit": "log10 of raw citation count (OpenAlex)", "year": "as of 2026-09", "source": "data/citations_gaps.json", "tag": "measured count (bibliometric)" }, "n": 6899, "n_unit": "studies", "x_filled": 6915, "y_filled": 6921, "frame_rows": 14779, "pearson": { "r": 0.9748, "p": 0.0, "ci95": [ 0.9736, 0.9759 ] }, "spearman": { "rho": 0.9825, "p": 0.0 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "Twin check between two citation databases.", "fdr_q_within_level": 0.0, "classification": "twin", "verdict": "Raw strong positive correlation r=0.97 (95% CI 0.97 to 0.98), Spearman rho=0.98, n=6899 studies, p=0.0, BH q=0. TWIN: |r|=0.975 >= 0.95: the two columns are the same quantity twice; never use both in one model. Twin check between two citation databases. Provenance: log10_icite_citations is measured count (bibliometric); log10_openalex_citations is measured count (bibliometric). Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LM-2", "level": "study", "kind": "measured_vs_modelled", "x": { "name": "log10_rcr", "unit": "log10 of NIH iCite Relative Citation Ratio (field- and year-normalised influence)", "year": "as of 2026-09", "source": "data/citations_gaps.json (iCite) joined by PMID/DOI", "tag": "modelled metric (iCite normalisation)" }, "y": { "name": "icite_percentile", "unit": "NIH iCite citation percentile (0-100) within field and year", "year": "as of 2026-09", "source": "data/citations_gaps.json (iCite)", "tag": "modelled metric (iCite normalisation)" }, "n": 6636, "n_unit": "studies", "x_filled": 6828, "y_filled": 6646, "frame_rows": 14779, "pearson": { "r": 0.8754, "p": 0.0, "ci95": [ 0.8696, 0.8809 ] }, "spearman": { "rho": 1.0, "p": 0.0 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [], "note": "RCR vs iCite percentile: same underlying metric, different scale.", "fdr_q_within_level": 0.0, "classification": "consistency_check", "verdict": "Raw strong positive correlation r=0.88 (95% CI 0.87 to 0.88), Spearman rho=1.00, n=6636 studies, p=0.0, BH q=0. CONSISTENCY CHECK: r=0.875: agreement level between two versions of the outcome / measurement. RCR vs iCite percentile: same underlying metric, different scale. Provenance: log10_rcr is modelled metric (iCite normalisation); icite_percentile is modelled metric (iCite normalisation). Both variables are modelled: this coefficient compares two models, not two measurements. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LG-1", "level": "genetics", "kind": "exposure", "x": { "name": "log10_allele_frequency", "unit": "log10 of risk-allele frequency", "year": "per record", "source": "data/genetics.json allele_frequency (gnomAD/Ensembl)", "tag": "measured (population genetics database)" }, "y": { "name": "abs_log_or", "unit": "absolute natural log of the reported odds ratio (effect magnitude regardless of direction)", "year": "per record", "source": "data/genetics.json odds_ratio", "tag": "curated from published studies" }, "n": 90, "n_unit": "curated variant records", "x_filled": 97, "y_filled": 107, "frame_rows": 135, "pearson": { "r": -0.0917, "p": 0.39, "ci95": [ -0.2932, 0.1176 ] }, "spearman": { "rho": -0.4257, "p": 2.88e-05 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": true, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "negative", "why": "rare variants carry larger effects" }, "confounders_named": [ "study power (rare variants are only reported when their effect is large)", "winner's curse", "mixed effect-size types (per-allele vs genotype)" ], "note": "Classic population-genetics pattern: rarer variants have larger reported effects. Partly biology (selection), partly detection bias.", "fdr_q_within_level": 0.39, "classification": "null", "verdict": "Raw negligible negative correlation r=-0.09 (95% CI -0.29 to 0.12), Spearman rho=-0.43, n=90 curated variant records, p=0.39, BH q=0.39. Pearson and Spearman disagree by more than 0.25: non-linear or outlier-driven; neither number should be read alone. NULL: raw association not distinguishable from zero (BH q=0.39 within level). Absence of a place-level correlation is not evidence of no individual-level effect. Classic population-genetics pattern: rarer variants have larger reported effects. Partly biology (selection), partly detection bias. Provenance: log10_allele_frequency is measured (population genetics database); abs_log_or is curated from published studies. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LG-2", "level": "genetics", "kind": "exposure", "x": { "name": "log10_n_cases", "unit": "log10 of number of cases in the source study", "year": "per record", "source": "data/genetics.json n_cases", "tag": "curated from published studies" }, "y": { "name": "abs_log_or", "unit": "absolute natural log of the reported odds ratio (effect magnitude regardless of direction)", "year": "per record", "source": "data/genetics.json odds_ratio", "tag": "curated from published studies" }, "n": 77, "n_unit": "curated variant records", "x_filled": 93, "y_filled": 107, "frame_rows": 135, "pearson": { "r": -0.5, "p": 3.66e-06, "ci95": [ -0.6511, -0.3108 ] }, "spearman": { "rho": -0.6757, "p": 1.57e-11 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "negative", "why": "smaller studies overestimate effects" }, "confounders_named": [ "small-study effects / winner's curse", "APOE is both large-n and large-effect" ], "note": "Small-study effect check.", "fdr_q_within_level": 7.32e-06, "classification": "raw_only", "verdict": "Raw strong negative correlation r=-0.50 (95% CI -0.65 to -0.31), Spearman rho=-0.68, n=77 curated variant records, p=3.66e-06, BH q=7.3e-06. RAW ONLY: significant raw association but no confounder could be partialled (no control named or insufficient overlap). Small-study effect check. Provenance: log10_n_cases is curated from published studies; abs_log_or is curated from published studies. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LG-3", "level": "genetics", "kind": "consistency", "x": { "name": "log10_n_cases", "unit": "log10 of number of cases in the source study", "year": "per record", "source": "data/genetics.json n_cases", "tag": "curated from published studies" }, "y": { "name": "ci_log_width", "unit": "ln(upper CI) - ln(lower CI) of the odds ratio: precision of the estimate", "year": "per record", "source": "data/genetics.json ci_lower/ci_upper", "tag": "curated from published studies" }, "n": 73, "n_unit": "curated variant records", "x_filled": 93, "y_filled": 101, "frame_rows": 135, "pearson": { "r": -0.5597, "p": 2.63e-07, "ci95": [ -0.6997, -0.3784 ] }, "spearman": { "rho": -0.741, "p": 6.64e-14 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "negative", "why": "larger studies give narrower CIs" }, "confounders_named": [], "note": "Precision should improve with n: a sanity check on the curation.", "fdr_q_within_level": 1.05e-06, "classification": "consistency_check", "verdict": "Raw strong negative correlation r=-0.56 (95% CI -0.70 to -0.38), Spearman rho=-0.74, n=73 curated variant records, p=2.63e-07, BH q=1e-06. CONSISTENCY CHECK: r=-0.560: agreement level between two versions of the outcome / measurement. Precision should improve with n: a sanity check on the curation. Provenance: log10_n_cases is curated from published studies; ci_log_width is curated from published studies. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LG-4", "level": "genetics", "kind": "exposure", "x": { "name": "year", "unit": "publication year", "year": "1975-2026", "source": "data/research.json", "tag": "curated metadata" }, "y": { "name": "abs_log_or", "unit": "absolute natural log of the reported odds ratio (effect magnitude regardless of direction)", "year": "per record", "source": "data/genetics.json odds_ratio", "tag": "curated from published studies" }, "n": 107, "n_unit": "curated variant records", "x_filled": 135, "y_filled": 107, "frame_rows": 135, "pearson": { "r": -0.2801, "p": 0.00347, "ci95": [ -0.4463, -0.0954 ] }, "spearman": { "rho": -0.1478, "p": 0.129 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": { "direction": "negative", "why": "effect sizes shrink as sample sizes grow" }, "confounders_named": [ "later GWAS find smaller common-variant effects", "early candidate-gene studies overestimated" ], "note": "", "fdr_q_within_level": 0.00463, "classification": "raw_only", "verdict": "Raw weak negative correlation r=-0.28 (95% CI -0.45 to -0.10), Spearman rho=-0.15, n=107 curated variant records, p=0.00347, BH q=0.0046. RAW ONLY: significant raw association but no confounder could be partialled (no control named or insufficient overlap). Provenance: year is curated metadata; abs_log_or is curated from published studies. Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] }, { "id": "LD-1", "level": "drugs", "kind": "uninterpretable", "x": { "name": "total_claims", "unit": "Medicare Part D/Part B claims in the latest available year for the drug-program pair", "year": "2022-2023", "source": "CMS via data/partd_ad_drugs.csv", "tag": "measured/administrative" }, "y": { "name": "faers_reports_generic_name", "unit": "cumulative FAERS adverse-event reports naming the drug (generic_name field)", "year": "all years to 2026-09-27", "source": "openFDA FAERS via data/faers_ad_drugs.json", "tag": "measured count (spontaneous reports, no denominator)" }, "n": 9, "n_unit": "drug-program pairs", "x_filled": 9, "y_filled": 9, "frame_rows": 9, "pearson": { "r": 0.8475, "p": 0.00391, "ci95": [ 0.4196, 0.9672 ] }, "spearman": { "rho": 0.8488, "p": 0.00381 }, "partial_primary": null, "partial_extended": null, "partial_strict": null, "divergence_flag": false, "sign_flip_vs_literature": false, "literature_expectation": null, "confounders_named": [ "utilisation drives reports (no denominator)", "stimulated reporting (litigation, media)", "cumulative reports vs single-year claims", "n is the number of drug-program pairs, not people" ], "note": "Reported only to document that it cannot be interpreted: FAERS counts have no exposure denominator and n is nine drug-program rows.", "fdr_q_within_level": 0.00391, "classification": "uninterpretable", "verdict": "Raw strong positive correlation r=0.85 (95% CI 0.42 to 0.97), Spearman rho=0.85, n=9 drug-program pairs, p=0.00391, BH q=0.0039. UNINTERPRETABLE: flagged a priori: utilisation drives reports (no denominator). Documented so that nobody publishes it as a finding. Reported only to document that it cannot be interpreted: FAERS counts have no exposure denominator and n is nine drug-program rows. Provenance: total_claims is measured/administrative; faers_reports_generic_name is measured count (spontaneous reports, no denominator). Study-level correlation: describes the composition of alzhub's library / curated records, not any patient population.", "caveats": [ "study-level composition only" ] } ] }