Prevention & risk

• One copy (heterozygous, e3/e4): OR 3.2 (95% CI 2.8-3.8); two copies (homozygous, e4/e4): OR 14.9 (95% CI 10.8-20.6), both vs. e3/e3 reference
• Evidence: High
• Genetic

Carrying one copy of the APOE4 allele roughly triples Alzheimer's odds; carrying two copies (homozygous) raises odds nearly 15-fold relative to the common e3/e3 genotype, a clear gene-dosage effect. The e2/e3 genotype was mildly protective (OR 0.6, 95% CI 0.5-0.8).

Landmark, still most-cited meta-analysis; not included in the Lancet Commission's modifiable-risk PAF model since APOE4 is non-modifiable. Effect size varies substantially by ancestry: stronger in Japanese populations (e4/e4 OR 33.1) and weaker in African American/Hispanic populations in the same meta-analysis.

Source: Farrer LA, Cupples LA, Haines JL, et al., JAMA 1997 (pooled meta-analysis, 40 research teams, 5,930 AD cases, 8,607 controls) · 1997

• RR 1.73 (95% CI 1.59-1.87) for 1+ affected first-degree relative; RR 3.98 (95% CI 3.26-4.82) for 2+ affected first-degree relatives
• Evidence: High
• Genetic

Risk climbs steeply with the number of affected close relatives; risk elevation extends even to second- and third-degree relatives, indicating a broad polygenic/inherited contribution beyond APOE4 alone.

Utah-specific founder population (largely European ancestry), which may limit generalizability. The reported RR for 3+ affected first-degree relatives (2.48, 95% CI 1.07-4.89) is numerically lower than for 2+ relatives, an instability the study authors themselves note as likely due to small subgroup size.

Source: Cannon-Albright LA, et al., Neurology 2019 (Utah Population Database, 270,818 individuals, 4,436 AD deaths) · 2019

• HR 1.12 (95% CI 1.02-1.23), meta-analysis of 21 cohorts (n=29,850); lifetime risk at age 65: women 24.6% vs. men 15.5% (all-cause dementia)
• Evidence: Moderate
• Demographic

Women show meaningfully higher lifetime dementia risk than men in Western cohort data, but the Lancet Commission attributes this substantially to women's greater life expectancy and historically lower educational attainment rather than to a clearly established sex-specific biological mechanism. The Commission's own text notes findings on sex differences are inconsistent across countries.

For Alzheimer's disease specifically (Framingham data), lifetime risk at 65 was 21.1% for women vs. 11.6% for men (p<0.001). Evidence is heterogeneous across countries/study designs.

Source: Lancet Commission 2024 report; Chene G, et al., Alzheimer's & Dementia 2015 (Framingham Heart Study) · 2015

• Symptomatic Alzheimer's prevalence reaches 90-100% by the seventh decade of life; lifetime risk of AD >90%
• Evidence: High
• Genetic

Trisomy 21 causes near-universal Alzheimer's brain pathology and very high rates of clinical dementia by late-middle-age, because the extra copy of chromosome 21 causes lifelong overexpression of the amyloid precursor protein (APP) gene. Among a cohort of 388 adults with Down syndrome, median age of Alzheimer's-dementia diagnosis was 53.7 years and median age of prodromal-AD diagnosis was 50.2 years.

The strongest known genetic risk factor for early-onset Alzheimer's in people who survive to middle age. Figures reflect people with Down syndrome who survive into adulthood; onset ages are typically decades earlier than late-onset Alzheimer's in the general population.

Source: Fortea J, Vilaplana E, Carmona-Iragui M, et al., The Lancet 2020 · 2020

• HR 0.89 (95% CI 0.83-0.95) for dementia (10 studies); HR 0.70 (95% CI 0.60-0.82) for Alzheimer's disease specifically (9 cohorts)
• Evidence: Moderate
• Lifestyle

Higher adherence to a Mediterranean-style diet is associated with an 11% lower risk of dementia overall and a 30% lower risk of Alzheimer's disease specifically in pooled cohort data spanning 2000-2024.

Purely observational; no large randomized controlled trial has confirmed causality for dementia as a primary endpoint. Residual confounding by socioeconomic status, education, and general healthy-lifestyle clustering is likely.

Source: Meta-analysis of 23 studies, 2000-2024 · 2024

• HR 0.75 (95% CI 0.58-0.97) for maintaining an individually optimal activity level vs. persistent inactivity
• Evidence: Moderate
• Lifestyle

Sustained physical activity across adulthood is associated with roughly a quarter reduction in dementia risk compared with persistent inactivity, in a cohort of 29,826 people followed a median 24.5 years with activity assessed twice a decade apart. This is the protective mirror of the 'physical-inactivity' risk-factor entry.

Same reverse-causation concern as inactivity: declining activity can itself be an early, pre-diagnostic dementia symptom, which may inflate the apparent protective effect of staying active.

Source: Lancet Commission 2024 report (cited cohort study) · 2024

• Probable dementia HR 0.83 (95% CI 0.67-1.04), 7.2 vs 8.6 cases per 1,000 person-years (not significant); MCI HR 0.81 (0.69-0.95), 14.6 vs 18.3 per 1,000 person-years; MCI or probable dementia HR 0.85 (0.74-0.97)
• Evidence: High
• Cardiovascular

The only large randomised trial of blood-pressure targets with adjudicated dementia as the primary cognitive outcome found no significant reduction in probable dementia with intensive control, although mild cognitive impairment and the combined MCI-or-dementia outcome were significantly reduced. The trial was stopped early for cardiovascular benefit, so it accrued fewer dementia cases than planned and may have been underpowered for the dementia endpoint.

PMID 30688979; PMC6439590; NCT01206062. 9,361 adults aged >=50 with hypertension, no diabetes or prior stroke, 102 US/Puerto Rico sites; intensive n=4,678, standard n=4,683; mean age 67.9; 35.6% women; 91.5% had >=1 follow-up cognitive assessment; median intervention 3.34 years, median follow-up 5.11 years; 149 vs 176 probable-dementia cases. Recorded as a null primary result with a significant secondary (MCI) result. Complements the observational 'hypertension' record.

Source: SPRINT MIND Investigators for the SPRINT Research Group; Williamson JD, Pajewski NM, Auchus AP, et al. Effect of Intensive vs Standard Blood Pressure Control on Probable Dementia: A Randomized Clinical Trial. JAMA 2019;321(6):553-561 · 2019

• All dementia triggers HR 1.03 (95% CI 0.91-1.17); probable AD HR 0.96 (0.74-1.24); MCI HR 1.12 (0.92-1.37); cognitive change over time similar in both arms
• Evidence: High
• Other

In a large placebo-controlled trial of community-dwelling adults aged 70+ (US minorities 65+) free of cardiovascular disease and dementia, daily low-dose aspirin did not reduce incident dementia, probable Alzheimer's disease, MCI or cognitive decline over a median 4.7 years. Aspirin is therefore not a dementia-prevention strategy in healthy older people.

PMID 32213642; PMC7455352; NCT01038583. 19,114 participants, US and Australia; 964 triggered dementia assessment; 575 adjudicated dementia cases (DSM-IV), 41% clinically probable AD. Class II evidence per the journal's classification. Deliberately recorded as a null finding.

Source: Ryan J, Storey E, Murray AM, et al. Randomized placebo-controlled trial of the effects of aspirin on dementia and cognitive decline. Neurology 2020;95(3):e320-e331 · 2020

• All-cause dementia HR 1.12 (95% CI 0.94-1.33; P=.21); Alzheimer's disease HR 1.16 (0.97-1.39; P=.11); progression from MCI to dementia HR 1.13 (0.85-1.50; P=.39); dementia rate 3.3 vs 2.9 per 100 person-years (ginkgo vs placebo)
• Evidence: High
• Other

The largest placebo-controlled trial of Ginkgo biloba, in 3,069 community volunteers aged 75+ with normal cognition or MCI followed for a median 6.1 years, found no reduction in all-cause dementia or Alzheimer's disease; point estimates were, if anything, slightly unfavourable. Ginkgo remains widely sold for memory despite this null result.

PMID 19017911; PMC2823569; NCT00010803. 5 US academic centres, 2000-2008; ginkgo n=1,545, placebo n=1,524; 2,587 normal cognition and 482 MCI at entry; 523 incident dementia cases (277 ginkgo, 246 placebo), 92% possible/probable AD or AD with vascular disease; dropout/loss to follow-up 6.3%; similar adverse-event profiles. Null finding recorded deliberately.

Source: DeKosky ST, Williamson JD, Fitzpatrick AL, et al. Ginkgo biloba for prevention of dementia: a randomized controlled trial. JAMA 2008;300(19):2253-2262 · 2008

• Adjusted HR vs placebo: vitamin E 0.88 (95% CI 0.64-1.20); selenium 0.83 (0.60-1.13); combination 1.00 (0.75-1.35); dementia incidence 4.4% overall (325/7,338), not different across arms
• Evidence: Moderate
• Lifestyle

Neither vitamin E nor selenium, alone or together, prevented dementia in 7,540 men aged 60+ enrolled through the SELECT prostate-cancer prevention trial. The randomised phase (2002-2009) converted to an observational cohort after SELECT closed for futility, which weakens the design, but the result is consistently null.

PMID 28319243; PMC5506489. Mean baseline age 67.5 (SD 5.3); 52.2% college-educated; 10.0% Black; 6.7% Hispanic; 130 SELECT sites; 3,786 continued into the cohort phase (Sep 2009-May 2015); modified intent-to-treat Cox models. Men only; dementia ascertained by two-stage screen plus record review, not universal clinical evaluation. Null finding recorded deliberately.

Source: Kryscio RJ, Abner EL, Caban-Holt A, et al. Association of Antioxidant Supplement Use and Dementia in the Prevention of Alzheimer's Disease by Vitamin E and Selenium Trial (PREADViSE). JAMA Neurol 2017;74(5):567-573 · 2017

• No significant association between MHT use and incident MCI or dementia in random-effects meta-analysis of 10 studies (1 RCT, 9 observational; 1,016,055 participants); subgroups by timing, duration and type also non-significant (pooled point estimate not given in abstract)
• Evidence: Moderate
• Other

A 2025 systematic review and meta-analysis found no evidence that menopause hormone therapy either raises or lowers dementia risk, and no effect of timing, duration or formulation. The authors conclude MHT should be prescribed for its other benefits and risks, not for dementia prevention. Certainty of evidence ranged from moderate to very low, and no study addressed premature ovarian insufficiency or testosterone.

PMID 41448220; PROSPERO CRD42025639384; primary studies searched 1 Jan 2000-20 Oct 2025; risk of bias via RoB 2 / ROBINS-E; GRADE. Funded by the Public Health Agency of Canada. The abstract reports the null result qualitatively; the numeric pooled RR was not in the source text we have, so the effect-size entry states the finding without inventing a point estimate.

Source: Melville M, He L, Desai R, et al. Menopause hormone therapy and risk of mild cognitive impairment or dementia: a systematic review and meta-analysis. Lancet Healthy Longev 2025;6(12):100803 · 2025

• Pooled OR 0.80 (95% CI 0.75-0.86) for dementia (36 studies) and OR 0.68 (0.56-0.81) for Alzheimer's disease (21 studies) in statin users vs non-users; men OR 0.86 (0.81-0.92) and women OR 0.86 (0.81-0.92); high-potency statins 20% vs low-potency 16% dementia risk reduction (P=0.05 for heterogeneity)
• Evidence: Low
• Cardiovascular

Pooled observational evidence associates statin use with about 20% lower odds of dementia and 32% lower odds of Alzheimer's disease, with no difference between lipophilic and hydrophilic statins. Because this is observational (healthy-user and indication bias are likely) and no adequately designed randomised trial exists, the authors frame it as absence of neurocognitive harm plus a possible favourable effect, not proof of prevention.

PMID 34871380. Search to January 2021; restricted maximum-likelihood random-effects model. Evidence strength set to 'low' because all included studies are observational and the authors call for ad hoc RCTs. Contrast with the 'high-ldl-cholesterol' record (LDL as a risk factor) - statin benefit does not follow automatically from that association.

Source: Olmastroni E, Molari G, De Beni N, et al. Statin use and risk of dementia or Alzheimer's disease: a systematic review and meta-analysis of observational studies. Eur J Prev Cardiol 2022;29(5):804-814 · 2022

• Dementia HR 1.21 (95% CI 1.07-1.38; 23 longitudinal studies); cognitive decline OR 1.23 (1.05-1.44; 24 studies); tooth loss alone: dementia HR 1.13 (1.04-1.23), cognitive decline OR 1.23 (1.09-1.39); complete tooth loss dementia HR 1.23 (1.05-1.45)
• Evidence: Low
• Other

Longitudinal studies link poor periodontal health and tooth loss to roughly 20% higher dementia risk, but the review authors rate the evidence low quality, highly heterogeneous, and 'at least partly due to reverse causality' (early cognitive decline worsens oral care). Not among the Lancet Commission's 14 factors.

PMID 36073186; PMC9826143. Five databases searched to April 2022; 47 studies included (24 cognitive decline, 23 dementia); random-effects pooling. Partial tooth loss OR 1.50 (1.02-2.23) for cognitive decline. Recorded with the authors' caution about reverse causality.

Source: Asher S, Stephen R, Mantyla P, Suominen AL, Solomon A. Periodontal health, cognitive decline, and dementia: A systematic review and meta-analysis of longitudinal studies. J Am Geriatr Soc 2022;70(9):2695-2709 · 2022

• Dementia: RR 0.92 (95% CI 0.88-0.96) at 1-17.5 g/day and RR 1.23 (1.09-1.35) at >17.5 g/day vs 0 g/day (J-shaped; lowest risk ~15 g/day); cognitive dysfunction: RR 0.97 (0.95-0.99) at 1-30.5 g/day and RR 1.07 (1.01-1.15) at >30.5 g/day (lowest risk ~30 g/day)
• Evidence: Low
• Lifestyle

A 2024 dose-response meta-analysis of 17 longitudinal studies finds a J-shaped curve: light drinking (about one standard drink a day or less) is associated with slightly lower dementia risk than abstention, while heavier intake raises risk by about a quarter. The apparent protective effect of light drinking is contested because abstainer groups often include former heavy drinkers and people who stopped for health reasons (sick-quitter bias); this record should be read with the 'excessive-alcohol' record.

PMID 39038743. Search to April 2023; restricted cubic spline dose-response; 80,680 persons with 4,929 dementia cases and 13,530 persons with 1,579 cognitive-dysfunction cases. J-shape persisted in subgroups by age (<=65, >65) and study duration for dementia. Evidence strength 'low' reflects observational design and abstainer-bias concerns raised in the summary (the abstract itself does not discuss sick-quitter bias).

Source: Zarezadeh M, Mahmoudinezhad M, Faghfouri AH, et al. Alcohol consumption in relation to cognitive dysfunction and dementia: A systematic review and dose-response meta-analysis of comparative longitudinal studies. Ageing Res Rev 2024;100:102419 · 2024

• Overall population: HR 0.93 (95% CI 0.86-1.01), not significant; high-risk populations (CKD, COPD, vascular disease): 2-3 doses HR 0.84 (0.76-0.92), >=4 doses HR 0.43 (0.38-0.48)
• Evidence: Low
• Other

Across eight cohort studies with nearly 10 million people, influenza vaccination was not significantly associated with lower dementia incidence in the general population, although a dose-response association appeared in people with chronic kidney disease, COPD or vascular disease. Repeated vaccination is strongly confounded by health-seeking behaviour, so the subgroup finding should not be read as causal; contrast with the quasi-experimental zoster-vaccine record.

PMID 40601364. Search to 6 April 2025; 8 cohort studies, 9,938,696 subjects; Newcastle-Ottawa risk of bias low in all but one study; random-effects model. Direction recorded as no-significant-effect for the overall population (the primary comparison); the authors' own conclusion emphasises the dose-response finding.

Source: Yang WK, Shao SC, Liu CC, Chi CC. Influenza vaccination and risk of dementia: a systematic review and meta-analysis. Age Ageing 2025;54(7):afaf169 · 2025

• Pooled survival time ratio 1.26 (95% CI 1.09-1.47) for dementia-free survival with high-school completion or above vs less than high school (total effect); natural direct effect 1.18 (1.01-1.36); indirect effect via occupational complexity 1.06 (1.02-1.09); 28% of the education effect mediated by occupational complexity
• Evidence: Moderate
• Other

Across 10,195 older adults in nine cohorts from six countries, completing high school lengthened dementia-free survival by about a quarter, and adult occupational complexity independently added protection, explaining roughly a quarter of the education effect. A threshold pattern was seen: finishing high school mattered, with little extra gain from further education. Adds the adult-life occupation pathway to the 'less-education' record.

PMID 34776437; PMC8748312 (full text loaded). Median baseline age 74.1 (range 58-103); accelerated failure time models per cohort then meta-analysed; cumulative dementia incidence across cohorts 3%-22%. Pooled values quoted are from the mediation-analysis table (Table 5) of the full text; occupational complexity estimates in Table 4 varied by cohort (e.g., EAS Black and LEILA75+ protective) and the pooled Table 3/4 rows were in figures not captured as text.

Source: Hyun J, Hall CB, Katz MJ, et al. Education, Occupational Complexity, and Incident Dementia: A COSMIC Collaborative Cohort Study. J Alzheimers Dis 2022;85(1):179-196 · 2022

What was tried: 2-year culturally adapted systematic multidomain lifestyle intervention (SLI: structured, supervised, monitored) vs flexible lifestyle intervention (FLI: health advice)

• 1,065 people
• Ages 60-77 (mean 67.5)
• For 2 years
• p = <0.0001

What was measured: co-primary: feasibility (RE-AIM Reach/Implementation/Maintenance) and change in global cognitive composite over 2 years

What was found: Mean annual change 0.31 SD (95% CI 0.28-0.34) SLI vs 0.20 SD (0.17-0.23) FLI; between-group difference 0.11 SD per year (95% CI 0.06-0.15)

Who was studied: At-risk adults 60-77 (CAIDE score >=6 and suboptimal cognitive performance) in 11 Latin American countries; 75% women; 59% Mestizo, 27% White

Limitations: Comparator was an active health-advice arm, not usual care; both groups improved (practice effects likely); participants and staff unmasked; analytic sample 1065 of 1719 assessed; higher dropout in FLI (20.2% vs 15.2%); adverse events far more frequent in SLI (412 vs 66, mostly musculoskeletal). NCT06492967.

Replication: replicated

Source: Lancet 2026;408(10553):417-429 · 2026 · PubMed 42442374 · doi:10.1016/S0140-6736(26)01278-X

What was tried: 48 months of supervised moderate-to-high-intensity multimodal training (aerobic, resistance, stretching) vs usual habits

• 882 people
• Ages older adults (ages not given in abstract)
• For 48 months
• p = 0.008 (healthy); 0.052 (SMC)

What was measured: 48-month change in MMSE

What was found: Healthy participants: exercise group higher MMSE than control (between-group effect p=0.008, partial eta-squared 0.007; time x group interaction partial eta-squared 0.025); SMC group between-group p=0.052 (partial eta-squared 0.004; interaction partial eta-squared 0.031); better memory, attention and reasoning in healthy and SMC exercisers

Who was studied: 882 community-dwelling older adults classed as healthy, subjective memory complaints (SMC) or MCI; 674 in final analysis

Limitations: Single-blind; 24% not in final analysis (674/882); MCI subgroup too small for conclusive analysis (authors); very small effect sizes; MMSE has ceiling effects in healthy adults.

Replication: not-replicated

Source: Iuliano E et al. Front Aging 2026;7:1847451 · 2026 · PubMed 42639433 · doi:10.3389/fragi.2026.1847451

What was tried: 6-month WeChat-delivered digital multidomain cognitive intervention vs control

• 166 people
• Ages >=60
• For 6 months
• p = 0.037 (MoCA)

What was measured: general cognitive function (MoCA)

What was found: MoCA improved vs control (t=2.106); ROCFT immediate and long-delay recall improved; UCLA loneliness reduced (Z=-2.641); no other significant differences

Who was studied: 166 adults >=60 with modified dementia risk score >79 at Fujian Provincial Hospital and 4 community health centres (April-December 2024); 154 (92.8%) completed

Limitations: Abstract does not enumerate limitations; small single-region sample (166), 6-month duration, no long-term follow-up, multiple secondary outcomes with borderline p-values.

Replication: not-replicated

Source: Yan Y et al. J Clin Psychiatry 2026;87(1):25m15860 · 2026 · PubMed 41499182 · doi:10.4088/JCP.25m15860

What was tried: Same as J-MINT 18-month multidomain intervention; secondary analysis by presence of untreated/uncontrolled vascular risk factors (BP, HbA1c, HDL, non-HDL cholesterol)

• 531 people
• Ages 65-85
• For 18 months
• p = 0.032 (interaction)

What was measured: interaction of intervention x VRF status on change in average neuropsychological Z score over 18 months

What was found: In participants with VRFs: Z score difference 0.11 (95% CI 0.02-0.20); no benefit in those without VRFs; interaction p=0.032

Who was studied: J-MINT participants (MCI, 65-85); 298 had >=1 untreated or uncontrolled VRF

Limitations: Secondary/subgroup analysis of a trial that was null on its primary outcome; hypothesis-generating. UMIN000038671.

Replication: not-replicated

Source: Alzheimers Dement 2025 (J-MINT investigators) · 2025 · PubMed 41222037 · doi:10.1002/alz.70822

What was tried: Non-physician community healthcare provider-led stepped-care antihypertensive titration to systolic <130 / diastolic <80 mm Hg (with physician supervision, home BP monitoring, health coaching) vs usual care; village-level cluster randomisation

• 33,995 people
• Ages 40+
• For 48 months
• p = 0.0035

What was measured: all-cause dementia (blinded endpoint adjudication)

What was found: All-cause dementia RR 0.85 (95% CI 0.76-0.95); net SBP reduction 22.0 mm Hg (95% CI 20.6-23.4), DBP reduction 9.3 mm Hg (95% CI 8.7-10.0); serious adverse events RR 0.94 (95% CI 0.91-0.98)

Who was studied: Adults >=40 with uncontrolled hypertension in 326 rural Chinese villages (163 intervention, 163 usual care)

Limitations: Open-label cluster design; rural China with very large achievable BP reduction (22 mm Hg) and low baseline treatment, so generalisability to well-treated populations is uncertain; only a 4-year window. NCT03527719.

Replication: not-replicated

Source: He J et al. Nat Med 2025;31(6):2054-2061 · 2025 · PubMed 40258956 · doi:10.1038/s41591-025-03616-8

What was tried: 3-year online, personalised multidomain coaching in 2-4 modules targeting physical activity, nutrition, cognitive activity (computerised brain training) and depression/anxiety, vs control receiving module-eligible information only

• 6,104 people
• Ages 55-77
• For 3 years
• p = <0.001

What was measured: 3-year change in global cognitive composite

What was found: Change in z score 0.28 (95% CI 0.25-0.32) intervention vs 0.10 (95% CI 0.07-0.13) control; between-group difference 0.18 (95% CI 0.13-0.23)

Who was studied: Dementia-free community-dwelling Australians aged 55-77 with modifiable dementia risk factors (45 and Up Study cohort); 64% female

Limitations: Online-only cognitive testing (potential overlap between brain-training module practice and outcome tests); single-blind; control received information only; younger and healthier than FINGER-type samples; 'social' tag reflects the depression/anxiety module (closest schema category). ACTRN12618000851268.

Replication: not-replicated

Source: Brodaty H et al. Nat Med 2025;31(2):565-573 · 2025 · PubMed 39875685 · doi:10.1038/s41591-024-03351-6

What was tried: Hearing intervention (audiological counselling + hearing aids) vs health education control

• 977 people
• Ages 70-84
• For 3 years

What was measured: interaction between predicted risk of cognitive decline and randomisation on 3-year cognitive decline

What was found: Among participants in the top quartile of predicted risk, 3-year cognitive decline was 61.6% (95% CI 33.7%-94.1%) slower with hearing intervention than control

Who was studied: 977 ACHIEVE participants; predicted-risk model developed in 2692 dementia-free ARIC participants

Limitations: Secondary/subgroup analysis; primary ACHIEVE result was null; effect-modification analyses are hypothesis-generating. NCT03243422.

Replication: not-replicated

Source: Pike JR et al. Alzheimers Dement 2025;21(5):e70156 · 2025 · PubMed 40369891 · doi:10.1002/alz.70156

What was tried: Best-practice hearing intervention (4 audiologist sessions, hearing aids, counselling, education) vs health-education control, 3 years

• 977 people
• Ages 70-84
• For 3 years

What was measured: exploratory: social isolation (Cohen Social Network Index) and loneliness (UCLA Loneliness Scale) over 3 years

What was found: Hearing intervention vs control: social network size +1.05 (95% CI 0.01-2.09), diversity +0.19 (95% CI 0.02-0.36), embeddedness +0.27 (95% CI 0.09-0.44); loneliness -0.94 (95% CI -1.78 to -0.11)

Who was studied: 977 adults 70-84 with untreated hearing loss without substantial cognitive impairment at 4 US sites (238 from ARIC, 739 de novo); mean age 76.3; mean better-ear PTA 39.4 dB

Limitations: Authors: exploratory outcomes; clinical meaningfulness of retaining ~1 additional social contact is unknown and loneliness changes were not clinically meaningful.

Replication: not-replicated

Source: Reed NS et al. JAMA Intern Med 2025;185(7):797-806 · 2025 · PubMed 40354063 · doi:10.1001/jamainternmed.2025.1140

What was tried: 2-year personalised multidomain risk-reduction: participant-chosen goals across 8 targeted risk factors with health coaching and nurse visits vs health education control; shifted to remote delivery during COVID-19

• 172 people
• Ages 70-89
• For 2 years
• p = 0.02

What was measured: change in composite modified Neuropsychological Test Battery

What was found: Composite cognition ATE 0.14 SD (95% CI 0.03-0.25), described as 74% improvement vs control change; composite risk-factor score ATE 0.11 SD (95% CI 0.01-0.20, P=0.03); QOL ATE 0.81 points (95% CI -0.21 to 1.84, P=0.12)

Who was studied: Adults 70-89 at elevated dementia risk (>=2 of 8 targeted risk factors) from Kaiser Permanente Washington primary care; 62.8% women; mean age 75.7

Limitations: Small pilot-scale trial (n=172); COVID-19 forced switch to remote delivery; musculoskeletal pain 14 vs 0. NCT03683394.

Replication: not-replicated

Source: Yaffe K et al. JAMA Intern Med 2024;184(1):54-62 (online Nov 2023) · 2024 · PubMed 38010725 · doi:10.1001/jamainternmed.2023.6279

What was tried: Four-arm cognitive training RCT: memory training, reasoning training, speed-of-processing training (with optional booster sessions), or no-contact control; linked to Medicare claims 1999-2019

• 2,021 people
• Ages 65+
• For 20 years follow-up

What was measured: claims-based diagnosed Alzheimer's disease and related dementias (Chronic Conditions Warehouse algorithm)

What was found: Speed training with >=1 booster session HR 0.75 (95% CI 0.59-0.95); speed training without booster HR 1.01 (95% CI 0.81-1.27); no main effect of memory or reasoning training

Who was studied: ACTIVE participants enrolled in traditional Medicare at baseline (older adults 65+, diverse US sample)

Limitations: Claims-based (not adjudicated) dementia; booster completion is post-randomisation (self-selected), so the HR 0.75 is not a pure ITT effect; original ACTIVE trial had no primary dementia outcome; sample restricted to traditional Medicare enrolees.

Replication: not-replicated

Source: Alzheimers Dement (N Y) 2026;12(1):e70197 · 2026 · PubMed 41669119 · doi:10.1002/trc2.70197

What was tried: Hearing intervention vs health education control

• 164 people
• Ages 70-84 (mean 78.1)
• For 3 years
• p = 0.019 (GFAP); 0.098 (NfL)

What was measured: 3-year change in plasma GFAP (neuroinflammation) and NfL (neurodegeneration)

What was found: GFAP change: intervention -0.026 (95% CI -0.144 to 0.093) vs control 0.149 (0.039 to 0.260); difference -0.175 (95% CI -0.322 to -0.028). NfL: 0.171 vs 0.330; difference -0.159 (95% CI -0.348 to 0.029)

Who was studied: ACHIEVE participants recruited from the ARIC cohort with plasma at baseline and 3 years (n=164; mean age 78.1)

Limitations: Small ARIC-only subsample (n=164); non-specific biomarkers; secondary analysis. NCT03243422.

Replication: not-replicated

Source: Pike JR et al. EClinicalMedicine 2026;99:104164 · 2026 · PubMed 42733779 · doi:10.1016/j.eclinm.2026.104164

What was tried: Structured high-intensity multidomain lifestyle (physical + cognitive activity, MIND-style healthy diet, social engagement, cardiovascular health monitoring; team-based with coaching and accountability) vs self-guided lower-intensity version of the same domains. NOTE: no no-intervention/usual-care control arm.

• 2,111 people
• Ages 60-79
• For 2 years
• p = 0.008

What was measured: difference between arms in annual rate of change in global cognitive function

What was found: Annual rate of increase 0.243 SD (95% CI 0.227-0.258) structured vs 0.213 SD (95% CI 0.198-0.229) self-guided; between-group difference 0.029 SD/year (95% CI 0.008-0.050)

Who was studied: US older adults at risk of cognitive decline: sedentary lifestyle + suboptimal diet + >=2 of family history of memory impairment, cardiometabolic risk, race/ethnicity, older age, sex; 5 US sites

Limitations: Both arms improved substantially (0.24 and 0.21 SD/yr, likely practice effects + self-guided arm itself being an intervention); no true control arm so absolute effect of lifestyle change vs nothing cannot be estimated; small between-group difference (0.029 SD/yr); clinical relevance and durability unresolved per authors; single-blind. NCT03688126.

Replication: replicated

Source: Baker LD et al. JAMA 2025;334(8):681-691 · 2025 · PubMed 40720610 · doi:10.1001/jama.2025.12923

What was tried: 12 months of continuous positive airway pressure (CPAP) plus best supportive care vs best supportive care alone

• 148 people
• Ages middle-aged (not specified in abstract)
• For 12 months (primary endpoint at 6 months)
• p = 0.91 (primary)

What was measured: Montreal Cognitive Assessment (MoCA) score at 6 months

What was found: MoCA difference -0.04 (95% CI -0.72 to 0.65), P=0.91 [null]; DMN functional connectivity difference -13.73 (95% CI -23.40 to -4.06), P=0.01; cortical thickness difference -0.06 mm (95% CI -0.10 to -0.01), P=0.02

Who was studied: Middle-aged adults with obstructive sleep apnea (AHI >=15/h) and normal cognition; 5 hospitals, China, 2017-2021

Limitations: Small; cognitively normal participants with ceiling on MoCA; short; not an older/at-risk-for-dementia population; 'sensory' tag used as closest schema category for a sleep/breathing intervention.

Replication: not-replicated

Source: Am J Respir Crit Care Med 2025;211(4):628-636 · 2025 · PubMed 39998447 · doi:10.1164/rccm.202406-1170OC

What was tried: Intensive (SBP <120) vs standard (SBP <140) treatment for 3.3 years; cognitive status re-ascertained by telephone from Dec 2019

• 9,361 people
• Ages 50+
• For median 7 years follow-up

What was measured: probable dementia (prespecified secondary analysis with extended follow-up)

What was found: Probable dementia HR 0.86 (95% CI 0.72-1.02), 8.5 vs 10.2 per 1000 person-years (248 vs 293 cases); MCI HR 0.87 (95% CI 0.76-1.00); MCI or probable dementia HR 0.89 (95% CI 0.79-0.99)

Who was studied: SPRINT participants; 7221 (77%) eligible for recontact, cognitive status ascertained for 4232 (59%); mean age 67, 36% female

Limitations: Still under-accrued for dementia (216 of 326 target cases); only 59% of eligible re-contacted; telephone ascertainment; Class II evidence.

Replication: not-replicated

Source: Neurology 2025;104(3):e213334 (SPRINT Research Group) · 2025 · PubMed 39819096 · doi:10.1212/WNL.0000000000213334

What was tried: High-dose subcutaneous gantenerumab up to 1500 mg every 2 weeks, open-label, for up to 3 years after the double-blind DIAN-TU-001 period

• 73 people
• Ages adults at risk for dominantly inherited AD (ages not given in abstract)
• For 3 years (stopped early after interim analysis; 13 completed 3 years)
• p = <0.0001 (PiB-PET SUVR); HRs not significant

What was measured: final analysis: 11C-PiB-PET SUVR change at 3 years (mITT n=55); prespecified interim: CDR-SB clinical decline in asymptomatic carriers vs external controls

What was found: Interim HR for CDR-SB decline in asymptomatic carriers 0.79 (95% CI 0.47-1.32; n=53) for any gantenerumab exposure and 0.53 (95% CI 0.27-1.03; n=22) for the longest-treated group; PiB-PET SUVR change from OLE baseline to year 3: -0.71 (95% CI -0.88 to -0.53)

Who was studied: Dominantly inherited AD mutation carriers who had completed DIAN-TU-001 and knew their mutation status; 18 sites in 8 countries; 74 recruited, 73 treated

Limitations: Authors: conclusions limited by open-label design and use of external controls; study stopped early by sponsor (no regulatory path for gantenerumab) - 47 of 73 stopped dosing for that reason and only 13 completed 3 years; small numbers (HR estimates on n=53 and n=22).

Replication: not-replicated

Source: Bateman RJ et al. Lancet Neurol 2025;24(4):316-330 · 2025 · PubMed 40120616 · doi:10.1016/S1474-4422(25)00024-9

What was tried: Hearing intervention: audiological counselling + provision and fitting of hearing aids vs health education control (individual sessions with health educator on chronic disease prevention); follow-up every 6 months

• 977 people
• Ages 70-84 (mean 76.8)
• For 3 years
• p = 0.96 (primary); 0.010 (cohort interaction); 0.027 (ARIC subgroup)

What was measured: 3-year change in global cognition standardised factor score from comprehensive neurocognitive battery

What was found: Total cohort: 3-year change -0.200 (95% CI -0.256 to -0.144) hearing vs -0.202 (-0.258 to -0.145) control; difference 0.002 (95% CI -0.077 to 0.081), p=0.96 [blank]. Prespecified sensitivity analysis by cohort (p interaction=0.010): ARIC cohort difference 0.191 (95% CI 0.022 to 0.360), p=0.027, described as a 48% reduction in 3-year cognitive change; de novo cohort difference -0.061 (95% CI -0.151 to 0.028), p=0.18

Who was studied: US adults 70-84 with untreated hearing loss, without substantial cognitive impairment; 4 community sites; two recruitment sources: ARIC cohort (n=238, older, more risk factors, lower baseline cognition) and healthy de novo volunteers (n=739); 88% White

Limitations: Primary analysis null; the widely cited 48% figure is a prespecified sensitivity/subgroup analysis in the higher-risk ARIC cohort (24% of sample) and needs replication; unmasked; 88% White; healthy-volunteer de novo cohort declined less than expected. NCT03243422.

Replication: not-replicated

Source: Lin FR et al. Lancet 2023;402(10404):786-797 · 2023 · PubMed 37478886 · doi:10.1016/S0140-6736(23)01406-X

What was tried: 2x2 factorial: cocoa extract (500 mg/day flavanols) vs placebo and daily multivitamin-mineral (Centrum Silver) vs placebo, 3 years; telephone cognitive assessment

• 2,262 people
• Ages 65+ (women), 60+ (men) per COSMOS design; mean 73
• For 3 years
• p = 0.28 (cocoa); 0.007 (multivitamin)

What was measured: change in global cognition composite with cocoa extract (primary); with multivitamin (prespecified secondary)

What was found: Cocoa extract: mean z 0.03 (95% CI -0.02 to 0.08), P=0.28 [null]; multivitamin: mean z 0.07 (95% CI 0.02-0.12), P=0.007 [positive]; multivitamin effect with CVD history 0.14 (95% CI -0.02 to 0.31) vs no history 0.06 (0.01-0.11), interaction nominal P=0.01

Who was studied: US older adults (mean age 73; 60% women; 89% non-Hispanic White); ancillary to COSMOS

Limitations: Multivitamin was the secondary comparison; telephone-only testing; 89% non-Hispanic White; incident MCI/dementia not reduced (see cosmos-mind-2023-incidence row).

Replication: replicated

Source: Baker LD et al. Alzheimers Dement 2023;19(4):1308-1319 · 2023 · PubMed 36102337 · doi:10.1002/alz.12767

What was tried: CPAP treatment for obstructive sleep apnea vs control (sham/no CPAP), pooled from randomised trials

• 1,926 people
• For varied
• p = 0.025 (severe OSA attention/speed)

What was measured: cognitive domains: attention/processing speed, executive function, memory; subjective sleepiness

What was found: Attention/processing speed improved only in severe OSA: SMD 0.17 (95% CI 0.02-0.31), I2=0%; no significant effects in other domains or overall

Who was studied: Adults with OSA in 14 RCTs (pooled n=1926)

Limitations: POOLED estimate from heterogeneous short trials, not a single trial; mostly not dementia-risk populations; sleep-apnea treatment tagged 'sensory' as nearest schema category.

Replication: not-replicated

Source: J Alzheimers Dis 2020;75(3):705-715 · 2020 · PubMed 32310179 · doi:10.3233/JAD-200088

What was tried: Dulaglutide 1.5 mg subcutaneous weekly (GLP-1 receptor agonist) vs placebo; cognition assessed with MoCA and DSST at baseline and follow-up

• 9,901 people
• Ages >=50
• For median 5.4 years (IQR 5.1-5.9)
• p = 0.11 (unadjusted); 0.0018 (post-hoc adjusted)

What was measured: exploratory: first occurrence of substantive cognitive impairment (MoCA or DSST score >=1.5 SD below country baseline mean)

What was found: 4.05 vs 4.35 events per 100 patient-years; HR 0.93 (95% CI 0.85-1.02); after post-hoc adjustment for individual standardised baseline scores HR 0.86 (95% CI 0.79-0.95)

Who was studied: Adults >=50 with type 2 diabetes and additional cardiovascular risk factors at 371 sites in 24 countries; 9901 randomized (4949 dulaglutide, 4952 placebo); 8828 with baseline plus follow-up cognitive scores

Limitations: Exploratory analysis of a cardiovascular outcomes trial (cognition was not a primary endpoint); the prespecified unadjusted comparison was not significant and benefit appeared only after post-hoc adjustment; diabetes population only.

Replication: not-replicated

Source: Cukierman-Yaffe T et al. Lancet Neurol 2020;19(7):582-590 · 2020 · PubMed 32562683 · doi:10.1016/S1474-4422(20)30173-3

What was tried: Intensive antihypertensive treatment to systolic BP goal <120 mm Hg vs standard goal <140 mm Hg

• 9,361 people
• Ages 50+ (mean 67.9)
• For median intervention 3.34 years; median cognitive follow-up 5.11 years

What was measured: adjudicated probable dementia

What was found: Probable dementia HR 0.83 (95% CI 0.67-1.04), 7.2 vs 8.6 per 1000 person-years (149 vs 176 cases) -- not significant; MCI HR 0.81 (95% CI 0.69-0.95), 14.6 vs 18.3 per 1000 PY; MCI or probable dementia HR 0.85 (95% CI 0.74-0.97), 20.2 vs 24.1 per 1000 PY

Who was studied: US/Puerto Rico ambulatory adults >=50 with hypertension and elevated cardiovascular risk, without diabetes or prior stroke; 102 sites

Limitations: Primary endpoint (probable dementia) not significant; trial stopped early for CV benefit (Aug 2015) so fewer dementia cases than planned -- authors state likely underpowered for dementia; excluded diabetics and stroke; 91.5% had >=1 follow-up cognitive assessment. NCT01206062.

Replication: not-replicated

Source: SPRINT MIND Investigators (Williamson JD et al.) JAMA 2019;321(6):553-561 · 2019 · PubMed 30688979 · doi:10.1001/jama.2018.21442

What was tried: Antihypertensive treatment to a 24-hour ambulatory systolic BP goal of <=130 mmHg (intensive) vs <=145 mmHg (standard) for 3 years

• 199 people
• Ages >=75
• For 3 years
• p = 0.03 (WMH); 0.91 (gait); 0.01 (MACE); cognition not significant (value not in abstract)

What was measured: co-primary: change in mobility (gait speed) and accrual of white-matter hyperintensity volume at 3 years; cognitive function (executive processing) secondary

What was found: Achieved 24-h SBP 127.7 vs 144.0 mmHg (difference 16.3); WMH volume accrual 0.29% intensive vs 0.48% standard; gait-speed change 0.40 vs 0.42 s (no difference); cognitive outcomes not different; major adverse cardiovascular events 4 vs 17

Who was studied: Adults >=75 with systolic hypertension and MRI white-matter hyperintensity lesions; mean age 80.5, 54% women; mean 24-h systolic BP 149 mmHg at baseline

Limitations: Small (199 randomized) single-centre trial with a 3-year horizon; cognition was a secondary outcome; authors note benefit on function may require periods >3 years.

Replication: replicated

Source: White WB et al. Circulation 2019;140(20):1626-1635 · 2019 · PubMed 31607143 · doi:10.1161/CIRCULATIONAHA.119.041603

What was tried: 2x2 factorial: 6 months supervised aerobic exercise and/or DASH diet nutritional counselling vs health education control

• 160 people
• Ages >55
• For 6 months
• p = 0.046 (exercise main effect); 0.012 (combined arm vs control)

What was measured: prespecified composite of executive function

What was found: Aerobic exercise d=0.32 (p=0.046); DASH diet d=0.30 (p=0.059, not significant); combined exercise+DASH vs health education d=0.40 (p=0.012); no improvement in memory or language/verbal fluency

Who was studied: 160 sedentary men and women >55 with cognitive impairment no dementia (CIND) and cardiovascular risk factors

Limitations: Authors describe findings as preliminary: small (n=160), 6-month intervention, single site; participants already had CIND (so this is delay-of-progression rather than primary prevention); DASH main effect not significant.

Replication: not-replicated

Source: Blumenthal JA et al. Neurology 2019;92(3):e212-e223 (epub 2018-12-19) · 2019 · PubMed 30568005 · doi:10.1212/WNL.0000000000006784

What was tried: 10 sessions of memory, reasoning or speed-of-processing training plus 4 booster sessions at 11 and 35 months, vs no-contact control

• 2,832 people
• Ages 65+ (mean 73.6 at baseline)
• For 10 years
• p = <0.05 (IADL comparison)

What was measured: targeted cognitive abilities and self-reported/performance-based everyday function (IADL)

What was found: Targeted cognition at 10 y: reasoning ES 0.23 (99% CI 0.09-0.38); speed ES 0.66 (99% CI 0.43-0.88); memory training effect not maintained. IADL difficulty: memory ES 0.48 (99% CI 0.12-0.84), reasoning 0.38 (0.02-0.74), speed 0.36 (0.01-0.72). Booster: reasoning ES 0.21 (0.01-0.41), speed ES 0.62 (0.31-0.93)

Who was studied: Independently living older adults in 6 US cities; mean baseline age 73.6; 26% African American

Limitations: Effects limited to trained abilities; substantial attrition over 10 years; IADL self-reported; no dementia endpoint in this paper (see active-20yr-claims-dementia-2026).

Replication: not-replicated

Source: Rebok GW et al. J Am Geriatr Soc 2014;62(1):16-24 · 2014 · PubMed 24417410 · doi:10.1111/jgs.12607

What was tried: Same as US POINTER: structured vs self-guided multidomain lifestyle intervention; ancillary MRI/PET substudy

• 983 people
• Ages 60-79
• For 2 years

What was measured: 4 primary imaging outcomes: global beta-amyloid burden (PET), entorhinal cortex tau burden (PET), hippocampal volume, white matter hyperintensity volume

What was found: No intervention-group differences in longitudinal cognitive or imaging outcomes in the imaging subsample; lower baseline hippocampal volume associated with greater cognitive benefit of structured arm (lower HC: 0.077 SD, 95% CI 0.022-0.132; higher HC: 0.002 SD, 95% CI -0.037 to 0.041; interaction P=0.03); ERC tau-cognition association difference 0.289 (95% CI 0.029-0.550; interaction P=0.03)

Who was studied: US POINTER participants eligible for and consenting to neuroimaging (983 of 1943 assessed; 516 structured, 467 self-guided; 61.5% female; mean age 68.4)

Limitations: Subsample of parent trial (about half); 2-year window may be too short for biomarker change; no usual-care control.

Replication: ongoing

Source: Harrison TM et al. JAMA Neurol 2026;83(6):521-529 · 2026 · PubMed 42008251 · doi:10.1001/jamaneurol.2026.0832

What was tried: Oral semaglutide (GLP-1 receptor agonist) up to 14 mg once daily (flexible dose) vs placebo for up to 156 weeks

• 3,808 people
• Ages 55-85 (mean 72.2)
• For 104 weeks primary endpoint (treatment up to 156 weeks); trials discontinued for negative outcome
• p = 0.57 (evoke); 0.46 (evoke+)

What was measured: change in Clinical Dementia Rating-Sum of Boxes (CDR-SB) from baseline to week 104

What was found: Evoke: CDR-SB change 2.3 (SE 0.1) semaglutide vs 2.3 (0.1) placebo, estimated difference -0.08 (95% CI -0.35 to 0.20); evoke+: 2.2 vs 2.1, difference 0.10 (95% CI -0.17 to 0.38)

Who was studied: Amyloid-confirmed early symptomatic Alzheimer's disease (MCI or mild dementia due to AD), aged 55-85; evoke+ additionally allowed significant small-vessel pathology (54 participants, 2.8%); 566 sites in 40 countries

Limitations: Treatment trial in symptomatic AD, not a primary-prevention trial; observational GLP-1/dementia associations were not confirmed; TEAEs 91.2% semaglutide vs 84.8% placebo. NCT04777396, NCT04777409. Both trials discontinued.

Replication: not-replicated

Source: Lancet 2026;407(10544):2167-2179 (evoke and evoke+ investigators) · 2026 · PubMed 41865758 · doi:10.1016/S0140-6736(26)00459-9

What was tried: Daily subcutaneous liraglutide (GLP-1 receptor agonist) vs placebo for 52 weeks

• 204 people
• For 52 weeks
• p = 0.14 (primary)

What was measured: change in cerebral glucose metabolic rate (FDG-PET)

What was found: Primary FDG-PET difference -0.17 (95% CI -0.39 to 0.06), P=0.14; ADAS-Exec better with liraglutide 0.15 (95% CI 0.03-0.28), unadjusted P=0.01; ADCS-ADL -0.58 (95% CI -3.13 to 1.97), P=0.65; CDR-SB -0.06 (95% CI -0.57 to 0.44), P=0.81

Who was studied: Mild to moderate Alzheimer's disease syndrome, non-diabetic; multicentre UK phase 2b

Limitations: Phase 2b, small n, symptomatic AD (treatment, not prevention); secondary cognitive signal unadjusted for multiplicity. NCT01843075.

Replication: not-replicated

Source: Edison P et al. Nat Med 2026;32(1):353-361 (epub Dec 2025) · 2026 · PubMed 41326666 · doi:10.1038/s41591-025-04106-7

What was tried: Cognitive behavioural therapy for insomnia (CBT-I) vs control; assessments at baseline, 6 weeks and 1 year

• 200 people
• Ages older adults (not specified in abstract)
• For 1 year
• p = 0.78; 0.40; 0.77

What was measured: change in cognitive performance (processing speed, executive function, memory) and amyloid-beta burden (subsample n=50)

What was found: Processing speed mean difference 0.017 (95% CI -0.1036 to 0.1376), p=0.78; executive function -0.0881 (95% CI -0.2945 to 0.1182), p=0.40; memory 0.4068 (95% CI -2.3965 to 3.2101), p=0.77; no group difference in amyloid deposition

Who was studied: Cognitively normal older adults with insomnia symptoms

Limitations: 1-year follow-up too short for AD-risk outcomes per authors; small amyloid subsample; 'cognitive-training' tag used for a behavioural (CBT) intervention as closest schema category.

Replication: not-replicated

Source: Alzheimers Dement 2026;22(6):e71591 · 2026 · PubMed 42298279 · doi:10.1002/alz.71591

What was tried: Crenezumab (anti-amyloid-beta antibody) subcutaneous 300 mg then 720 mg every 2 weeks, later optional 60 mg/kg IV every 4 weeks, vs placebo, for 5-8 years in cognitively unimpaired mutation carriers; mutation non-carriers received placebo as a masked kindred control

• 252 people
• Ages 30-60
• For 5-8 years (common close); enrolled Dec 2013-Feb 2017, final data March 2022
• p = 0.43 (composite); 0.16 (FCSRT-CI)

What was measured: dual primary: annualised rate of change in API preclinical ADAD composite total score and in FCSRT Cueing Index

What was found: API ADAD composite annualised change -1.10 (SE 0.29) crenezumab vs -1.43 (0.29) placebo, difference 0.33 (95% CI -0.48 to 1.13); FCSRT-CI -0.03 vs -0.04, difference 0.01 (95% CI 0.00 to 0.02)

Who was studied: Cognitively unimpaired PSEN1 Glu280Ala (E280A) kindred members aged 30-60 in Medellin, Colombia; 85 crenezumab carriers, 84 placebo carriers, 83 placebo non-carriers; 63% women

Limitations: Single-centre, single-kindred phase 2 with 169 mutation carriers; crenezumab did not remove fibrillar amyloid, so the trial does not test plaque clearance; authors conclude robust fibrillar amyloid removal appears necessary for clinical efficacy.

Replication: not-replicated

Source: Tariot PN et al. Lancet Neurol 2026;25(2):147-159 · 2026 · PubMed 41579901 · doi:10.1016/S1474-4422(25)00426-0

What was tried: 18 months of supervised (first 12 months) moderate-high intensity aerobic training vs low-intensity stretching/balance/range-of-motion

• 296 people
• Ages 65-89 (per trial design; abstract states older adults)
• For 18 months
• p = 0.3

What was measured: global cognitive composite (ADAS-Cog-Exec) at 12 months

What was found: Aerobic vs stretching regression coefficient -0.078 (SE 0.074); no 12-month cognitive decline in either group; mean 12-month hippocampal volume loss 0.51% in both groups

Who was studied: Sedentary older adults with amnestic MCI; multicentre US phase 3

Limitations: No no-exercise control (both arms exercised, so absolute exercise effect unmeasured); the companion ADNI comparison (PMID 40271887) is non-randomised; age range stated from trial design rather than abstract.

Replication: not-replicated

Source: Baker LD et al. Alzheimers Dement 2025;21(4):e14586 · 2025 · PubMed 40271888 · doi:10.1002/alz.14586

What was tried: FINGER-based multimodal lifestyle intervention vs control

• 56 people
• Ages 60-77 (mean 68.8)
• For ~28 months between scans

What was measured: hippocampal volume, entorhinal cortex thickness, free water fraction, PSMD, WMH volume, grey-matter cerebral blood flow

What was found: No conclusive intervention effect on any marker; entorhinal thickness and free water declined MORE in intervention group in preregistered models but not after adjusting for baseline differences; exploratory: cerebral blood flow increased more in intervention group, associated with SBP reductions (no numeric estimates in abstract)

Who was studied: Leipzig AgeWell.de participants with 3T MRI before/after intervention (56 baseline, 41 follow-up; mean age 68.8)

Limitations: Very small (41 with follow-up), single site, smaller than planned; baseline imbalance.

Replication: not-replicated

Source: Beyer F et al. Neuroimage Clin 2025;46:103796 · 2025 · PubMed 40378769 · doi:10.1016/j.nicl.2025.103796

What was tried: 18-month multidomain intervention: management of vascular risk factors + exercise + nutritional counselling + cognitive training vs control

• 531 people
• Ages 65-85
• For 18 months

What was measured: change in cognitive composite score over 18 months

What was found: Between-group difference in composite change 0.047 (95% CI -0.029 to 0.124)

Who was studied: Japanese community-dwelling older adults 65-85 with mild cognitive impairment

Limitations: MCI population (later stage than FINGER); 18 months; 24% attrition; subgroup findings exploratory.

Replication: not-replicated

Source: Sakurai T et al. Alzheimers Dement 2024;20(6):3918-3930 · 2024 · PubMed 38646854 · doi:10.1002/alz.13838

What was tried: 4 years of oral conjugated equine estrogens 0.45 mg/d or transdermal estradiol 50 ug/d (each with cyclic micronized progesterone) vs placebo in the KEEPS trial (2005-2008), re-assessed ~10 years after trial completion (2017-2022)

• 275 people
• Ages recently postmenopausal at randomization (ages not given in abstract)
• For 4-year treatment + ~10-year post-trial follow-up (~14 years from randomization)

What was measured: cognitive performance (4 cognitive factor scores and global score) at ~10 years after trial end by original randomization arm

What was found: MHT allocation had no significant effect on cognitive slopes during KEEPS or across all years of follow-up in any domain; cross-sectional performance similar in oCEE, tE2 and placebo groups

Who was studied: Recently postmenopausal women (within 3 years of final menstrual period at randomization); 299 of 727 original participants enrolled at 7 of 9 sites, 275 with cognitive data at both time points

Limitations: Observational follow-up of 41% of the original randomized cohort (authors report similar baseline characteristics for returners vs non-returners); 7 of 9 sites; healthy population.

Replication: replicated

Source: Gleason CE et al. PLoS Med 2024;21(11):e1004435 · 2024 · PubMed 39570992 · doi:10.1371/journal.pmed.1004435

What was tried: MIND diet (Mediterranean-DASH hybrid) with mild caloric restriction vs control diet with mild caloric restriction; both arms received dietary counselling and weight-loss support; 3 years

• 604 people
• Ages 65+
• For 3 years
• p = 0.23

What was measured: change from baseline in global cognition score and four domain scores (12-test battery)

What was found: Global cognition improved in both groups: +0.205 SD MIND vs +0.170 SD control; mean difference 0.035 SD (95% CI -0.022 to 0.092)

Who was studied: Cognitively unimpaired older adults with family history of dementia, BMI >25 and suboptimal diet; two US sites (Chicago, Boston)

Limitations: Control diet also involved counselling and caloric restriction (active comparator); both groups improved (practice effects/weight loss); age range not stated in abstract (65+ per trial design, from PMC). NCT02817074.

Replication: not-replicated

Source: Barnes LL et al. N Engl J Med 2023;389(7):602-611 · 2023 · PubMed 37466280 · doi:10.1056/NEJMoa2302368

What was tried: Umibecestat (oral BACE1 inhibitor) 50 mg or 15 mg daily vs placebo in cognitively unimpaired APOE e4 homozygotes, and heterozygotes with elevated brain amyloid; studies terminated early after ~7 months average exposure

• 1,556 people
• Ages 60-75
• For 7 months treatment on average (terminated early); median 4 (IQR 3-6) months washout follow-up

What was measured: planned: time to diagnosis of MCI/dementia due to AD and API preclinical composite; this report: reversibility of cognitive worsening after washout

What was found: Small, non-progressive but statistically significant decline vs placebo on RBANS and API Preclinical Composite (not CDR-SB) during treatment; RBANS differences no longer significant at end of washout follow-up (numeric effect sizes not in abstract)

Who was studied: Cognitively unimpaired 60-75 year-old APOE e4 homozygotes and heterozygotes with elevated amyloid

Limitations: Studies terminated early (July 2019) so no efficacy readout; reversibility assessed over a median of only 4 months off drug; abstract reports no effect sizes.

Replication: replicated

Source: Tariot PN et al. Alzheimers Dement 2024;20(11):7745-7761 · 2024 · PubMed 39320017 · doi:10.1002/alz.14237

What was tried: Intensive vs standard SBP control; brain MRI at baseline and ~4 years

• 673 people
• Ages 50+ (mean 67.3)
• For median 3.98 years
• p = 0.03

What was measured: change in hippocampal volume, AD regional atrophy pattern, posterior cingulate cerebral blood flow, cingulum fractional anisotropy

What was found: Hippocampal volume decreased 7.45 to 7.39 cm3 (intensive) vs 7.48 to 7.46 cm3 (standard); between-group difference in change -0.033 cm3 (95% CI -0.062 to -0.003), P=0.03 (greater loss with intensive treatment); no differences in AD regional atrophy, CBF or FA

Who was studied: SPRINT MRI substudy: 673 with baseline MRI (356 intensive, 317 standard), 454 with follow-up MRI; mean age 67.3, 40.3% women

Limitations: Small substudy; hippocampal difference tiny and of uncertain clinical meaning (authors relate it to greater total brain volume decrease with intensive treatment); no cognitive consequences shown.

Replication: not-replicated

Source: Nasrallah IM et al. JAMA Neurol 2021;78(5):568-577 · 2021 · PubMed 33683313 · doi:10.1001/jamaneurol.2021.0178

What was tried: Atabecestat (oral BACE inhibitor) 5 mg or 25 mg daily vs placebo (1:1:1) in amyloid-positive cognitively normal adults

• 557 people
• Ages 60-85
• For planned 4.5 years; dosing stopped May 2018 (trial ran Nov 2015-Dec 2018) with 6 months off-treatment follow-up
• p = <0.001

What was measured: change from baseline in Preclinical Alzheimer Cognitive Composite (PACC)

What was found: Atabecestat 25 mg vs placebo: PACC least-squares mean difference -1.09 (95% CI -1.66 to -0.53) at month 6 and -1.62 (95% CI -2.49 to -0.76) at month 12; RBANS -3.70 (95% CI -5.76 to -1.63) at month 3; worsening reversed after 6 months off treatment

Who was studied: Amyloid-positive, cognitively normal (CDR 0) adults aged 60-85 at 143 centres in 14 countries; 557 randomized of 4464 screened (34% of planned 1650); 61% women; mean age 70.4

Limitations: Truncated trial: enrollment stopped at 34% of plan and treatment stopped early for hepatic adverse events, so the planned efficacy question could not be answered; follow-up off-treatment limited to 6 months.

Replication: replicated

Source: Sperling R et al. JAMA Neurol 2021;78(3):293-301 · 2021 · PubMed 33464300 · doi:10.1001/jamaneurol.2020.4857

What was tried: Conjugated equine estrogen 0.625 mg + medroxyprogesterone acetate 2.5 mg daily vs placebo

• 4,532 people
• Ages 65+
• For mean 4.05 years (SD 1.19) to last 3MSE
• p = 0.01 (dementia); 0.72 (MCI)

What was measured: incidence of probable dementia

What was found: Probable dementia HR 2.05 (95% CI 1.21-3.48); 45 vs 22 per 10,000 person-years (40 vs 21 cases); 23 additional dementia cases per 10,000 women per year. MCI HR 1.07 (95% CI 0.74-1.55), 63 vs 59 per 10,000 PY

Who was studied: Postmenopausal women >=65 years free of probable dementia, from 39 WHI clinical centres (USA)

Limitations: Small number of events (61); women >=65 (late initiation), so does not test the 'critical window' hypothesis; parent WHI E+P trial stopped July 2002 for other harms.

Replication: not-replicated

Source: Shumaker SA et al. JAMA 2003;289(20):2651-62 · 2003 · PubMed 12771112 · doi:10.1001/jama.289.20.2651

What was tried: Multidomain lifestyle intervention (FINGER-adapted), 2 years

• 1,008 people
• Ages 60-79
• For 2 years

Who was studied: Cognitively healthy Dutch adults 60-79 with dementia prevention potential; 1008 included via Dutch Brain Research Registry campaign (13,795 registrants, 3475 eligible)

Limitations: No primary outcome publication located as of 2026-09-22; n reflects inclusions via the registry campaign only (total enrolment may differ).

Replication: ongoing

Source: Waterink L et al. Alzheimers Dement (N Y) 2025;11(1):e70017 (recruitment paper) · 2025 · PubMed 39790861 · doi:10.1002/trc2.70017

What was tried: Donanemab IV monthly x9 infusions vs placebo, then clinical assessments every 6 months (decentralized design, plasma p-tau217 eligibility)

• 2,196 people
• Ages 55-80
• For 9 monthly infusions then event-driven follow-up

What was measured: time to clinical progression (time-to-event)

Who was studied: Adults 55-80 in the US and Japan with AD pathology detected by plasma p-tau217; 63,124 screened, 2196 enrolled (CDR-Global 0 n=1202; 0.5 n=664); mean baseline amyloid 63.2 centiloids (CDR 0) and 70.7 (CDR 0.5)

Limitations: Baseline/screening paper only; no efficacy results. Plasma p-tau217 eligibility rates differed by age, race and ethnicity (authors).

Replication: ongoing

Source: Yaari R et al. Alzheimers Dement 2025;21(9):e70662 (screening and baseline paper) · 2025 · PubMed 40955720 · doi:10.1002/alz.70662

What was tried: Multimodal lifestyle (aerobic + resistance exercise, MIND diet advice, BrainHQ cognitive training, medical monitoring/health education) vs health education and coaching, 2 years

• 600 people
• Ages 55-79
• For 2 years

What was measured: change in global composite cognitive score

Who was studied: Australian adults 55-79

Limitations: Protocol only (planned n=600); no results located as of 2026-09-22.

Replication: ongoing

Source: Gardener SL et al. Alzheimers Dement (N Y) 2024;10(2):e12466 (protocol) · 2024 · PubMed 38596483 · doi:10.1002/trc2.12466

What was tried: FINGER 2.0 multidomain lifestyle (diet, physical activity, cognitive training, vascular/metabolic monitoring, social interaction) +/- metformin 1000-2000 mg/day (trial-within-trial, double-blind vs placebo) vs regular health advice

• 600 people
• Ages 60-79
• For 2 years

What was measured: change in global cognition (NTB)

Who was studied: APOE e4-enriched adults 60-79 at increased dementia risk; UK, Finland, Sweden

Limitations: Phase IIb proof-of-concept protocol; planned n=600; no results located as of 2026-09-22. NCT05109169.

Replication: ongoing

Source: Barbera M et al. Alzheimers Res Ther 2024;16(1):23 (protocol) · 2024 · PubMed 38297399 · doi:10.1186/s13195-023-01355-x

What was tried: Lecanemab (BAN2401) IV, dose regimens tailored to baseline amyloid: A3 (intermediate amyloid ~20-40 centiloids, phase 2, PET endpoints) and A45 (elevated amyloid >40 centiloids, phase 3, cognitive composite primary) vs placebo, 4-year treatment

• 1,400 people
• Ages 55-80
• For 4 years treatment

What was measured: A45: change in cognitive composite (PACC5); A3: amyloid PET change

Who was studied: Cognitively unimpaired adults 55-80 with intermediate or elevated brain amyloid; plasma Abeta42/40 used to screen before PET; recruitment via TRC-PAD, ADNeT and J-TRC registries

Limitations: Design paper only; no results. ClinicalTrials.gov NCT04468659: phase 3, ACTIVE_NOT_RECRUITING, enrollment 1400 (ESTIMATED), start 2020-07-14, primary completion 2028-12-21 (registry loaded 2026-09-27).

Replication: ongoing

Source: Rafii MS et al. Alzheimers Dement 2023;19(4):1227-1233 (design paper; epub Aug 2022) · 2023 · PubMed 35971310 · doi:10.1002/alz.12748

What was tried: Structured multidomain lifestyle (diet training, supervised then unsupervised exercise, cognitive training, vascular risk management) vs self-guided intervention, 2 years

• 1,200 people
• Ages 60-77
• For 2 years

What was measured: global cognition (modified Neuropsychological Test Battery)

Who was studied: Singapore adults 60-77 with CAIDE score >=6, >=1 LIBRA lifestyle risk, MoCA 18-27

Limitations: Protocol only; planned n=1200; no results located in PubMed as of 2026-09-22.

Replication: ongoing

Source: Xu X et al. J Prev Alzheimers Dis 2022;9(1):40-48 (protocol) · 2022 · PubMed 35098972 · doi:10.14283/jpad.2022.5

What was tried: Multidomain lifestyle intervention (FINGER-adapted) in rural communities; intervention arm details not in the baseline paper

• 5,765 people
• Ages 60+ (mean 70.9)

Who was studied: Rural residents >=60 years, western Shandong province, China; 5765 completed baseline assessment (mean age 70.9; 57.2% women; 40.6% illiterate)

Limitations: No intervention outcome publication located in PubMed as of 2026-09-22 (search 'MIND-China[tiab]' returned baseline/cross-sectional papers only). the participant count is the baseline cohort, not necessarily the randomised sample.

Replication: ongoing

Source: Wang Y et al. Alzheimers Dement (N Y) 2022;8(1):e12254 (baseline paper) · 2022 · PubMed 35441085 · doi:10.1002/trc2.12254