You can call any of these for free. You do not need a diagnosis, an appointment, or a crisis to call — questions and plain exhaustion are reasons enough.
Alzheimer's Association 24/7 Helpline800-272-390024 hours a day, 365 days a yearFree, confidential support and information for people with dementia, caregivers, families and the public. Master's-level clinicians. Also Spanish.
988 Suicide & Crisis Lifeline98824/7Call or text 988, or chat at 988lifeline.org. For anyone in emotional distress or a mental-health crisis, including caregivers.
Alzheimer's Foundation of America (AFA) Helpline866-232-84849:00 am - 9:00 pm ET, seven days a weekLicensed social workers providing support, info and referrals. Also text message and web chat.Text: 646-586-5283
In a medical emergency, call 911.
More numbers, and what each line does, on the help page.
What the evidence says: vaccines, drugs, prevention
Alzhub's library holds 27,380 studies, and almost all of them are about individual
people — a person takes a drug, a person receives a vaccine, an outcome is measured in that
person. This page leads with that evidence, organised by topic. The older catalogue of
place-level (state/county/country) correlations is still here, real and unedited — just moved
to the bottom, where a weaker class of evidence belongs.
The previous version of this page was a catalogue of ecological questions only
— do states, counties or countries with more of an exposure also show more of an outcome. That
is a correlation between places, and this site's own study library is overwhelmingly a
record of people: a vaccination status, a prescription, a lifestyle exposure, each tied to
one person's own outcome. Vaccines, drugs and modifiable risk factors now lead the page, each
section reporting what the individual-level studies themselves found — not a new statistic
computed here. The place-level catalogue is kept, unchanged, at the bottom.
Every count, study design, effect size, participant number, title, and
link below is read directly out of our study library (or, for the geography section,
the same public panels as before) at build time. Where a topic has no matching individual-level
study, that is reported as a gap — not filled with a plausible-sounding number.
43 place-level (ecological) questions, kept at the bottom
1 report a stated gap rather than a number
The geography catalogue still has more individual rows
than the topic sections above it — each of its 43 questions was already a single
exposure/outcome pair before this redesign, and re-merging them would have discarded coefficients
already computed independently for every one of them. What changed is prominence, not row count:
the individual-level evidence now leads the page, and the geography section is boxed, captioned as
weaker, and reads last.
Vaccines
3 questions
Vaccination status is recorded in an individual's own medical record, and dementia diagnosis is recorded in that same person -- an individual-level exposure and an individual-level outcome, the kind of evidence this site's library is actually built from. The strongest cards below are natural experiments; the weakest is a stated gap.
Herpes zoster (shingles) vaccination and dementia risk
13 studies
The strongest individual-level evidence this library holds on any prevention question. Three of the five studies below are natural experiments: an age or birth-date eligibility cut-off for the free shingles vaccine gave two otherwise-similar groups different vaccine access, which is the closest this evidence gets to a randomised comparison without actually randomising anyone. All five are still observational cohorts, not trials -- the biological mechanism (viral reactivation and neuroinflammation is one hypothesis) is not established, and this describes the study groups, not a guarantee for any one person.
13 studies: Meta-analysis 1, RCT / clinical trial 1, Cohort 9, Other / not reported 2.
This chart is wider than your screen — drag it sideways.
Source: each study's own reported effect size. Bars show the absolute drop in new dementia diagnoses for the vaccine-eligible group; whiskers are each study's own 95% confidence interval.
Herpes zoster (shingles) vaccination and dementia risk — the best-evidence studies on this topic, from our study library.
Influenza, pneumococcal and Tdap vaccination and dementia risk
3 studies
Pooled observational evidence, from the same systematic review that covers the shingles vaccine above. The direction is consistent across every vaccine type studied, but these are not natural experiments -- a healthier person is generally more likely to get any vaccine, which can inflate an observational estimate like these.
3 studies: Meta-analysis 1, Cohort 2.
This chart is wider than your screen — drag it sideways.
Source: key findings reported in the meta-analysis itself. Bars run from the reported relative risk to 1.0 (no association); lower is fewer dementia diagnoses among the vaccinated.
Influenza, pneumococcal and Tdap vaccination and dementia risk — the best-evidence studies on this topic, from our study library.
A gap, stated plainly rather than filled with a plausible-sounding number: this library has no individual-level study measuring dementia risk by COVID-19 vaccination status. The COVID-vaccine studies it does hold measure death, hospitalisation or breakthrough infection -- real outcomes, just not this one.
Gap, stated plainly: No study in our library measures dementia risk by COVID-19 vaccination status.
Drugs
12 questions
Two FDA-approved anti-amyloid antibodies have randomised-trial evidence of slowing decline; four older drugs are approved for symptom management only; the rest are repurposed candidates under investigation, not approved for Alzheimer's. Each card says which is which before it shows a single number.
Study designs behind these topics
This chart is wider than your screen — drag it sideways.
Source: the study design each matching record states. Darker segments are higher up the evidence pyramid.
Approved on the strength of randomised trials showing a modest slowing of decline versus placebo in early Alzheimer's disease. Carries a boxed safety warning for brain swelling or bleeding (ARIA), monitored by MRI during treatment.
CDR-SB mean difference -0.18 (95% CI -0.34 to -0.03), below the MCID; ARIA-oedema NNH 9
Showing the 5
best-evidence studies of 111 in our library matching this topic, ranked by
whether the record reports an effect size, then by study design, then by sample size.
Approved on the same basis as lecanemab: randomised trials showing a modest slowing of decline in early Alzheimer's disease, with the same ARIA monitoring requirement.
74 studies: Systematic review 9, Meta-analysis 3, RCT / clinical trial 10, Cohort 4, Other / not reported 48.
Donanemab and cognitive/functional decline — the best-evidence studies on this topic, from our study library.
CDR-SB mean difference -0.18 (95% CI -0.34 to -0.03), below the MCID; ARIA-oedema NNH 9
Showing the 5
best-evidence studies of 74 in our library matching this topic, ranked by
whether the record reports an effect size, then by study design, then by sample size.
Donepezil and symptom progression
96 studies
Approved (FDA, 1996) — cholinesterase inhibitor
Long-approved for symptomatic treatment: it can improve or stabilise symptoms for a period, but it does not modify the underlying disease course.
Showing the 5
best-evidence studies of 96 in our library matching this topic, ranked by
whether the record reports an effect size, then by study design, then by sample size.
Memantine and symptom progression
94 studies
Approved (FDA, 2003) — NMDA-receptor antagonist
Approved for moderate-to-severe Alzheimer's, usually alongside a cholinesterase inhibitor. Symptomatic, not disease-modifying.
94 studies: Systematic review 5, Meta-analysis 6, RCT / clinical trial 8, Cohort 7, Other / not reported 68.
Memantine and symptom progression — the best-evidence studies on this topic, from our study library.
Showing the 5
best-evidence studies of 94 in our library matching this topic, ranked by
whether the record reports an effect size, then by study design, then by sample size.
Twelve of these fourteen topics are the modifiable risk factors named in the Lancet Commission's dementia-prevention framework (education, hearing loss, traumatic brain injury, hypertension, alcohol, obesity, smoking, depression, social isolation, physical inactivity, diabetes, air pollution), plus diet and sleep. Every count below is an individual-level study -- a person's own exposure, a person's own outcome -- not a place-level average.
Study designs behind these topics
This chart is wider than your screen — drag it sideways.
Source: the study design each matching record states, for records filed under prevention or risk factors.
RR 0.80 (95% CI 0.75-0.86) for all-cause dementia with high vs low physical activity; leisure-time RR 0.76 (0.65-0.88); occupational RR 1.20 (1.01-1.42)
Pooled HR 0.72 (95% CI 0.65-0.80) overall; HR 0.85 (95% CI 0.79-0.91) after maximal confounder adjustment; high-intensity activity HR 0.56 (95% CI 0.45-0.68); moderate-intensity HR 0.85 (95% CI 0.80-0.93); low-intensity HR 0.94 (95% CI 0.77-1.15, not significant)
Showing the 5
best-evidence studies of 134 in our library matching this topic, ranked by
whether the record reports an effect size, then by study design, then by sample size.
Showing the 5
best-evidence studies of 122 in our library matching this topic, ranked by
whether the record reports an effect size, then by study design, then by sample size.
Sleep
82 studies
82 studies: Systematic review 2, Meta-analysis 6, Cohort 11, Case-control 1, Cross-sectional 11, Other / not reported 51.
Sleep — the best-evidence studies on this topic, from our study library.
HR ~1.22 (age 50) and ~1.37 (age 60) for short vs. 7-hour sleep; persistent short sleep across ages 50/60/70 ~30% higher risk
Showing the 5
best-evidence studies of 82 in our library matching this topic, ranked by
whether the record reports an effect size, then by study design, then by sample size.
Showing the 5
best-evidence studies of 89 in our library matching this topic, ranked by
whether the record reports an effect size, then by study design, then by sample size.
Everything below this line compares PLACES, not people: do states, counties or countries with more of an exposure also show more of an outcome. A correlation between places is not a statement about a person -- it cannot tell you that living somewhere changes your own risk, only that two averages move together. It is also frequently a correlation about measurement rather than disease: under-diagnosis and genuinely lower disease look identical in this kind of data, because both produce a lower recorded rate. These 43 questions are kept, unchanged, at the bottom of this page because the panels behind them are real and the arithmetic is real -- but they are a weaker form of evidence than anything above, and should be read that way.
A correlation between places is not a statement about a
person. It cannot tell you that living somewhere changes your own risk — only
that two averages move together, for reasons the data below cannot separate. It is also
frequently a correlation about measurement: under-diagnosis and genuinely lower disease
look identical in this kind of data, because both produce a lower recorded rate.
S1 Do states with higher fine-particulate air pollution carry higher Alzheimer's mortality rates?
Ready
Level: state
Units: 51
Analysed: 51
Pearson r: -0.0 (p = 0.999)
Spearman ρ: 0.072 (p = 0.613)
Partial r: -0.097 (controlling for median household income, primary-care physicians per 100,000)
PM2.5 is a live candidate environmental risk factor. State-level pollution averages vary enough to give the comparison range, and the mortality outcome is complete at this level.
Design. Ecological cross-section, 51 states and DC.
What it might actually be measuring. Recorded-cause mortality tracks diagnostic and coding capacity, so Test 1 matters most here: correlate the outcome against income and physician supply first. If mortality tracks those hard, the pollution correlation is partly measuring health-system intensity.
Verdict. Raw negligible negative association (r=-0.00, n=51, p=0.999, not distinguishable from zero at p<0.05). Because the raw association was not significant to begin with, the partial r=-0.10 (p=0.497) controlling for median household income + primary-care physicians per 100,000 has no real baseline effect to explain away -- reported for completeness, not as a finding. Ecological correlation only: does not show the exposure causes or prevents Alzheimer's at the individual level.
S1 — the two columns, and where each comes from.
Role
Column
What it is
Filled
Source
Exposure
pm25_avg_ugm3
Average annual fine particulate matter (PM2.5) concentration
Reported as a 2022-2024 multi-year average by the source; year recorded as the midpoint (2023).
Unadjusted (not age-adjusted) annual Alzheimer's death rate per 100,000 population
Report explicitly presents UNADJUSTED rates (not age-adjusted) so that larger/older-population states aren't penalized in the comparison; do not treat as age-standardized.
S2 Is state-level hearing-loss prevalence associated with Alzheimer's prevalence?
Ready
Level: state
Units: 51
Analysed: 51
Pearson r: -0.611 (p = 0.0)
Spearman ρ: -0.595 (p = 0.0)
Partial r: -0.587 (controlling for the share aged 65+)
Hearing loss is one of the best-supported modifiable risk factors in the Lancet Commission work, so it is the strongest prior for a positive finding - which is exactly why a null would be informative.
Design. Ecological cross-section, 51 states and DC.
What it might actually be measuring. Prevalence is modelled from an age structure, so it will track the share aged 65+ almost by construction. Adjust for age structure before reading the exposure; if the association vanishes, it was demography.
Verdict. Raw strong negative association (r=-0.61, n=51, p=1.93e-06). Controlling for the share aged 65+, the partial r is -0.59 (p=5.89e-06) and the association SURVIVES -- not explained away by the share aged 65+ alone, though this remains an ecological (group-level) association, not evidence of an individual-level effect. SIGN FLIP vs. established literature (hearing loss is a well-established (Lancet Commission) modifiable risk factor; more hearing loss is expected to track with more dementia, not less) -- read this as a measurement/diagnostic-capacity finding about the data, not a discovery about risk. Ecological correlation only: does not show the exposure causes or prevents Alzheimer's at the individual level.
S2 — the two columns, and where each comes from.
Role
Column
What it is
Filled
Source
Exposure
pct_hearing_loss
Percent of adults with any hearing loss
Oldest data-year in the risk-factor block (2014-2016 pooled); no newer full 51-state BRFSS hearing-loss table was locatable in this session.
Estimated percentage of the population age 65+ living with Alzheimer's dementia
Modeled estimate based on 2020 age/sex/race population distributions, not a direct case count or registry; published in the 2025 F&F report.
S3 Is state-level diabetes prevalence associated with Alzheimer's prevalence?
Ready
Level: state
Units: 51
Analysed: 50
Pearson r: 0.134 (p = 0.352)
Spearman ρ: 0.399 (p = 0.004)
Partial r: 0.237 (controlling for the obesity rate, the hypertension rate, the physical inactivity rate)
Midlife diabetes is an established risk factor. Both columns are complete, making this the cleanest first test of the whole panel.
Design. Ecological cross-section, 51 states and DC.
What it might actually be measuring. Diabetes, obesity, inactivity and hypertension are mutually correlated, so a single-exposure model will not isolate any one of them. Run the exposures together and report which survive; a lone 'significant' predictor among four collinear ones is not a finding.
Flagged: Pearson and Spearman
disagree materially on this pair, which usually means a non-linear relationship or a few outlying units.
Neither figure should be read on its own.
Verdict. Raw weak positive association (r=0.13, n=50, p=0.352, not distinguishable from zero at p<0.05). Pearson (r=0.13) and Spearman (rho=0.40) disagree materially -- likely non-linear or outlier-driven; neither number should be read alone. Because the raw association was not significant to begin with, the partial r=0.24 (p=0.105) controlling for the obesity rate + the hypertension rate + the physical inactivity rate has no real baseline effect to explain away -- reported for completeness, not as a finding. Ecological correlation only: does not show the exposure causes or prevents Alzheimer's at the individual level.
S3 — the two columns, and where each comes from.
Role
Column
What it is
Filled
Source
Exposure
pct_diabetes
Percent of adults with diagnosed diabetes
Tennessee blank: TN did not submit sufficient 2024 BRFSS sample for KFF to report a value.
Estimated percentage of the population age 65+ living with Alzheimer's dementia
Modeled estimate based on 2020 age/sex/race population distributions, not a direct case count or registry; published in the 2025 F&F report.
S4 Do more-educated states record higher Alzheimer's mortality?
Ready
Level: state
Units: 51
Analysed: 51
Pearson r: -0.58 (p = 0.0)
Spearman ρ: -0.427 (p = 0.002)
Partial r: -0.255 (controlling for the share aged 65+, median household income)
A sign flip is the highest-value signal in an ecological analysis. Education is protective at the individual level, so if it correlates POSITIVELY with recorded mortality at state level that is almost certainly a diagnostic-capacity artifact - and demonstrating it is a real result.
Design. Ecological cross-section, 51 states and DC.
What it might actually be measuring. This is the designed artifact test. Expect a positive sign. Run the full four-step triage: Test 1 against wealth, Test 2 stratified into income quartiles, Test 3 against the modelled prevalence outcome, Test 4 partial on age and income together. Publishing the triage, not the coefficient, is the deliverable.
Verdict. Raw strong negative association (r=-0.58, n=51, p=8.14e-06). Controlling for the share aged 65+ + median household income, the partial r falls to -0.25 (p=0.0711) -- well under half the raw value, so the raw association largely COLLAPSES and is consistent with it measuring the share aged 65+ + median household income rather than the share with a bachelor's degree itself. Ecological correlation only: does not show the exposure causes or prevents Alzheimer's at the individual level.
S4 — the two columns, and where each comes from.
Role
Column
What it is
Filled
Source
Exposure
pct_bachelors_plus
Percent of population 25+ with a bachelor's degree or higher
Unadjusted (not age-adjusted) annual Alzheimer's death rate per 100,000 population
Report explicitly presents UNADJUSTED rates (not age-adjusted) so that larger/older-population states aren't penalized in the comparison; do not treat as age-standardized.
S5 Do states with more neurologists per capita identify more Alzheimer's cases?
Ready
Level: state
Units: 51
Analysed: 51
Pearson r: 0.01 (p = 0.947)
Spearman ρ: 0.207 (p = 0.145)
Partial r: 0.119 (controlling for total population)
A direct measure of diagnostic supply against a case count. It quantifies how much of the geography of Alzheimer's is a geography of who gets diagnosed.
Design. Ecological cross-section, 51 states and DC.
What it might actually be measuring. Counts are denominators-inflated: a large state has more cases for reasons unrelated to supply. Model the RATE not the count, or include population as an offset - otherwise the correlation restates population.
Verdict. Raw negligible positive association (r=0.01, n=51, p=0.947, not distinguishable from zero at p<0.05). Because the raw association was not significant to begin with, the partial r=0.12 (p=0.406) controlling for total population has no real baseline effect to explain away -- reported for completeness, not as a finding. Ecological correlation only: does not show the exposure causes or prevents Alzheimer's at the individual level.
S5 — the two columns, and where each comes from.
Role
Column
What it is
Filled
Source
Exposure
neurologists_per_100k
Neurologists per 100,000 population
Based on AAN membership records, not a full practitioner census; likely a modest, roughly uniform undercount of true neurologist supply.
Estimated number of people age 65+ living with Alzheimer's dementia
Report gives this in thousands (1 decimal); value here is that figure x1000, rounded to nearest integer. Same modeled-estimate caveat as alz_prevalence_65plus_pct.
S6 Is geriatrician supply associated with Alzheimer's mortality at state level?
Ready
Level: state
Units: 51
Analysed: 51
Pearson r: -0.285 (p = 0.043)
Spearman ρ: -0.271 (p = 0.055)
Partial r: -0.409 (controlling for total population)
Tests whether specialist workforce tracks recorded outcomes, the same confound structure as S5 but with a different workforce series, so the two can be compared.
Design. Ecological cross-section, 51 states and DC.
What it might actually be measuring. Geriatrician counts are raw counts, so they track state size. Rate-adjust first. If S5 and S6 disagree in sign, that disagreement is itself the finding about how unstable workforce measures are at this level.
Verdict. Raw weak negative association (r=-0.28, n=51, p=0.0428). Controlling for total population, the partial r is -0.41 (p=0.00286) and the association SURVIVES -- not explained away by total population alone, though this remains an ecological (group-level) association, not evidence of an individual-level effect. Ecological correlation only: does not show the exposure causes or prevents Alzheimer's at the individual level.
S6 — the two columns, and where each comes from.
Role
Column
What it is
Filled
Source
Exposure
geriatricians_count
Number of board-certified geriatricians practicing in the state (bonus column, not in required spec)
2021 is the most recent year available in the source report; included as a bonus workforce-supply column alongside primary_care_physicians_per_100k and neurologists_per_100k.
Unadjusted (not age-adjusted) annual Alzheimer's death rate per 100,000 population
Report explicitly presents UNADJUSTED rates (not age-adjusted) so that larger/older-population states aren't penalized in the comparison; do not treat as age-standardized.
S11 Do states with more unpaid caregiving hours have more Alzheimer's cases?
Ready
Level: state
Units: 51
Analysed: 51
Pearson r: 0.944 (p = 0.0)
Spearman ρ: 0.967 (p = 0.0)
Partial r: -0.046 (controlling for total population)
Both are burden measures. Useful as a consistency check between an independent caregiver-burden series and the case count.
Design. Ecological cross-section, 51 states and DC.
What it might actually be measuring. Both columns scale with state population. Normalise both per capita or the correlation is a population correlation wearing a caregiving label.
Verdict. Raw strong positive association (r=0.94, n=51, p=2.61e-25). Controlling for total population, the partial r falls to -0.05 (p=0.75) -- well under half the raw value, so the raw association largely COLLAPSES and is consistent with it measuring total population rather than unpaid caregiving hours itself. Ecological correlation only: does not show the exposure causes or prevents Alzheimer's at the individual level.
S11 — the two columns, and where each comes from.
Role
Column
What it is
Filled
Source
Exposure
unpaid_caregiver_hours_millions
Hours of unpaid care provided by family/other caregivers
US total 19,161 million hours (~19.2 billion).
Estimated number of people age 65+ living with Alzheimer's dementia
Report gives this in thousands (1 decimal); value here is that figure x1000, rounded to nearest integer. Same modeled-estimate caveat as alz_prevalence_65plus_pct.
Study counts and designs come straight from the study design each matching
record in our library states. Reported effect is whatever
the study itself reported — a relative risk, a percentage-point difference, an odds ratio —
left blank when the record does not carry one. Nothing on the vaccine, drug or prevention sections is a
coefficient computed by this site; that work belongs to the studies cited.
Pearson r and Spearman ρ, in the geography section only, run
from −1 to +1 and measure how tightly two columns move together (Spearman on ranks, so it
survives a few extreme values that would distort r). Partial r re-runs that
correlation after removing the influence of a named confounder.
Observational means the study watched what happened
without assigning who got the exposure; randomised (RCT) means it did. An
observational association, however large, is not proof that the exposure caused the outcome —
people who choose an exposure (a vaccine, a drug, an exercise habit) differ from people who do not in
ways a study can rarely fully account for. None of this is medical advice.
The full study library is browsable on
Research, and the panels behind the geography section are on
Data & downloads. Found a wrong number? Tell us.