Skip to content

Risk reduction and prevention

TL;DR — The observational case for prevention is large and the trial case is small, and the gap between them is the subject of this page. Meta-analysis of population attributable fractions across 74 studies gives weighted PAFs of 9.3% for low education (95% CI 6.9–11.7), 7.3% for physical inactivity (3.9–11.2), 7.2% for hearing loss (5.2–9.7), 7.1% for hypertension (5.4–8.8) and 5.3% for obesity (3.2–7.4), with a seven-factor combined weighted PAF of 32.0% (26.6–37.5) — and higher PAFs in low- and middle-income countries than in high-income countries (Stephan 2024, PMID 38824956). Randomised multidomain trials have produced effects that are statistically real and clinically tiny where positive, and null where the population was unselected: FINGER's between-group difference in annual neuropsychological test battery change was 0.022 z-points (95% CI 0.002–0.042, P=0.030); US POINTER's structured versus self-guided difference was 0.029 SD per year (0.008–0.050, P=0.008) in a trial where both arms improved; MAPT and preDIVA were null (Ngandu 2015, PMID 25771249; Baker 2025, PMID 40720610; Andrieu 2017, PMID 28359749; Moll van Charante 2016, PMID 27474376). Single-factor trials divide the same way: SPRINT MIND reduced MCI (HR 0.81, 95% CI 0.69–0.95) but not probable dementia (0.83, 0.67–1.04), while ACHIEVE's hearing intervention was null overall (difference 0.002, 95% CI −0.077 to 0.081) with a prespecified interaction suggesting benefit only in the higher-risk ARIC subcohort (p_interaction=0.010) (SPRINT MIND 2019, PMID 30688979; Lin 2023, PMID 37478886). The most striking recent causal claim comes from outside the lifestyle paradigm: a birth-date regression-discontinuity design in Wales found herpes zoster vaccination reduced 7-year dementia diagnosis by 3.5 percentage points (95% CI 0.6–7.1, P=0.019), a 20.0% relative reduction (Eyting 2025, PMID 40175543).

Two kinds of evidence that are routinely conflated

Observational PAF Randomised trial
Question answered If exposure were eliminated across the life course, what fraction of dementia might not occur? If we intervene on exposure now, in this population, for this duration, what happens?
Time horizon Decades, often from midlife 2–6 years, usually starting after 60
Assumption Association is causal and reversible None (randomisation handles confounding)
Failure mode Residual confounding, reverse causation, overlapping risk factors summed Wrong population, wrong window, too short, contaminated control arm

The Lancet standing Commission on dementia prevention, intervention and care sets out the modifiable-risk agenda in its 2024 report (Livingston 2024, PMID 39096926). This build did not retrieve numeric content from that document — the PubMed record carries no abstract — so the quantitative PAF figures on this page are taken from Stephan's independent meta-analysis rather than attributed to the Commission.

Population attributable fractions

Stephan's systematic review identified PAF values for 61 modifiable risk factors across 74 studies, with sufficient data to meta-analyse 12 (48 studies). Unweighted PAFs assume each factor acts alone; weighted PAFs adjust for communality (overlap) between factors and are the appropriate figure for policy (Stephan 2024, PMID 38824956):

Risk factor Unweighted PAF (95% CI) Weighted PAF (95% CI)
Low education 17.2% (14.4–20.0) 9.3% (6.9–11.7)
Hypertension 15.8% (14.7–17.1) 7.1% (5.4–8.8)
Hearing loss 15.6% (10.3–20.9) 7.2% (5.2–9.7)
Physical inactivity 15.2% (12.8–17.7) 7.3% (3.9–11.2)
Obesity 9.4% (7.3–11.7) 5.3% (3.2–7.4)
Seven-factor model (low education, midlife hypertension, midlife obesity, smoking, physical inactivity, depression, diabetes; 9 studies) 55.0% (46.5–63.5) 32.0% (26.6–37.5)

Two features of this table are frequently lost in summary. First, the unweighted and weighted figures differ by roughly a factor of two, because risk factors co-occur; quoting unweighted PAFs — or summing individual factors — nearly doubles the apparent preventable fraction. Second, pooled PAFs for most individual factors were higher in low- and middle-income countries than in high-income countries, which is where the projected burden growth is concentrated (epidemiology and burden) but where only 7% of sites in the audited global AD trial sample were located (Cummings 2025, PMID 40555627).

That communality correction has a lineage. Barnes and Yaffe estimated that up to half of worldwide AD cases (17.2 million of 33.9 million) were potentially attributable to seven factors, with 10–25% reductions preventing 1.1–3.0 million cases (Barnes 2011, PMID 21775213). Norton then used English survey data to adjust for non-independence and cut the combined worldwide PAR from 49.4% (25.7–68.4) to 28.2% (14.2–41.5) — 9.6 million of 33.9 million cases — with a 10%-per-decade relative reduction in each factor projected to cut 2050 prevalence by 8.3% (Norton 2014, PMID 25030513). Stephan's 32.0% (26.6–37.5) is the current pooled estimate of that same seven-factor construct. The number has been stable at "about one third after overlap" for a decade; what has not been stable is any randomised demonstration that moving those factors moves dementia incidence.

Multidomain trials

Trial Population Intervention Result
FINGER (Ngandu 2015, PMID 25771249) 1,260 Finns aged 60–77, CAIDE risk score ≥6, cognition at or slightly below age-expected 2 years of diet, exercise, cognitive training and vascular risk monitoring vs general health advice NTB total z change 0.20 (intervention) vs 0.16 (control); between-group difference per year 0.022 (95% CI 0.002–0.042), P=0.030. Dropout 12%. Adverse events 7% vs 1%, mostly musculoskeletal pain (5% vs 0%)
US POINTER (Baker 2025, PMID 40720610) 2,111 US adults 60–79 with sedentary lifestyle, suboptimal diet and ≥2 additional risk criteria; 68.9% female; 5 sites 2 years, structured higher-intensity vs self-guided lifestyle intervention (both encouraged activity, cognition, diet, social engagement and cardiovascular monitoring) Global cognitive composite increased in both arms: 0.243 SD/y (95% CI 0.227–0.258) structured vs 0.213 SD/y (0.198–0.229) self-guided; difference 0.029 SD/y (0.008–0.050), P=0.008. Benefit consistent across APOE ε4 status (p=0.95) and greater in those with lower baseline cognition (p=0.02). Fewer adverse events in the structured arm
MAPT (Andrieu 2017, PMID 28359749) 1,680 French/Monegasque adults ≥70 with memory complaint, IADL limitation or slow gait 3 years: multidomain (43 group sessions) + omega-3, multidomain + placebo, omega-3 alone, or placebo No significant difference in 3-year composite cognitive change for any arm vs placebo: combined 0.093 (95% CI 0.001–0.184, adjusted P=0.142); multidomain+placebo 0.079 (−0.012 to 0.170, P=0.179); omega-3 alone 0.011 (−0.081 to 0.103, P=0.812)
preDIVA (Moll van Charante 2016, PMID 27474376) 3,526 Dutch adults aged 70–78 from 116 general practices, unselected 6 years of nurse-led multidomain cardiovascular care vs usual care Dementia in 121/1,853 (7%) vs 112/1,601 (7%); HR 0.92 (95% CI 0.71–1.19), P=0.54. No difference in disability, mortality (HR 0.98) or incident cardiovascular disease (HR 1.06). Median follow-up 6.7 years

Reading these four together:

  • Selection matters. FINGER and US POINTER enriched for risk and were positive; preDIVA enrolled an unselected population with modest baseline cardiovascular risk against a high standard of usual care and was null — the authors attribute the absence of effect to exactly that.
  • Effect sizes are very small. FINGER's 0.022 z-points per year and POINTER's 0.029 SD per year are group-mean differences on composite cognitive scores with no established clinical anchor. Neither trial measured dementia incidence as a primary outcome; preDIVA, which did, was null.
  • The comparator matters. US POINTER compared two active interventions, and both arms' cognition improved over 2 years — a practice-effect and healthy-volunteer pattern. The trial therefore establishes that structure adds something to self-guided lifestyle change, not that lifestyle change prevents dementia.
  • Duration is short relative to the exposure window. Midlife vascular risk predicted late-life amyloid two decades later, whereas late-life vascular risk did not (Gottesman 2017, PMID 28399252) — the trials intervene at the wrong end of that timeline.

Single-factor trials

Factor Trial Result
Blood pressure SPRINT MIND: 9,361 hypertensive adults ≥50 without diabetes or prior stroke, SBP <120 vs <140 mm Hg, median 3.34 y of intervention, 5.11 y follow-up Probable dementia HR 0.83 (95% CI 0.67–1.04); MCI HR 0.81 (0.69–0.95); MCI-or-dementia HR 0.85 (0.74–0.97). Underpowered for dementia after early termination (PMID 30688979)
Hearing ACHIEVE: 977 adults aged 70–84 with untreated hearing loss and no substantial cognitive impairment, hearing aids + counselling vs health education, 3 years Primary analysis null: 3-year cognitive change −0.200 (95% CI −0.256 to −0.144) intervention vs −0.202 (−0.258 to −0.145) control; difference 0.002 (−0.077 to 0.081), P=0.96. Prespecified sensitivity analysis: significant difference in effect between the ARIC subcohort (older, more risk factors, lower baseline cognition) and de-novo volunteers, p_interaction=0.010 (PMID 37478886)
Omega-3 supplementation MAPT arm No effect (0.011, 95% CI −0.081 to 0.103, P=0.812) (PMID 28359749)
Herpes zoster vaccination Natural experiment: eligibility in Wales determined by exact date of birth (born before vs on/after 2 Sept 1933); vaccination uptake jumped from 0.01% to 47.2% across a one-week birth-date threshold; regression-discontinuity design on electronic health records Dementia diagnosis reduced by 3.5 percentage points over 7 years (95% CI 0.6–7.1, P=0.019), a 20.0% relative reduction (6.5–33.4); stronger in women; confirmed in England and Wales combined using death-certificate dementia (PMID 40175543)

ACHIEVE is the clearest illustration of the population problem: a null primary result with a prespecified interaction indicating that the intervention may work in people at higher risk of decline and not in low-risk volunteers. That is a hypothesis-generating subgroup finding, and the authors present it as such — but it is the same pattern as FINGER-versus-preDIVA, from a different direction.

The zoster-vaccine result is methodologically the strongest causal design on this page: a discontinuity in eligibility that is arbitrary with respect to everything except vaccination, replicated in a second population with a different outcome definition. It does not establish mechanism — off-target immune effects and suppression of neurotropic herpesvirus reactivation are both candidates — and it is a single natural experiment.

Age-specific dementia incidence fell 13% per calendar decade across seven European and US cohorts over 27 years (95% CI 7–19%) (Wolters 2020, PMID 32611641). Framingham, which used consistent diagnostic criteria across four epochs in 5,205 people aged ≥60, is the longest single-cohort demonstration: 5-year cumulative hazard 3.6 → 2.8 → 2.2 → 2.0 per 100 persons, a 22%, 38% and 44% decline relative to the late-1970s epoch, observed only among people with at least a high-school diploma (HR 0.77, 95% CI 0.67–0.88) and not fully explained by the falling prevalence of vascular risk factors (Satizabal 2016, PMID 26863354). In Gothenburg 85-year-olds the fall was concentrated in vascular dementia, explained mainly by rising education and by a weaker association between stroke and dementia in later-born cohorts (OR for dementia given stroke fell from 4.3 to 1.8, interaction P=0.008) (Skoog 2017, PMID 28733627). Stroke itself confers pooled HR 1.69 (95% CI 1.49–1.92) for prevalent and RR 2.18 (1.90–2.50) for incident stroke (Kuźma 2018, PMID 30177276).

The midlife exposure window is not an inference from trials that failed late. In eastern Finland, after a mean 21-year follow-up of 1,449 people, midlife systolic blood pressure ≥160 mm Hg carried OR 2.3 (95% CI 1.0–5.5) for later AD and midlife cholesterol ≥6.5 mmol/L carried OR 2.1 (1.0–4.4); both together carried OR 3.5 (1.6–7.9); midlife diastolic pressure did not (Kivipelto 2001, PMID 11408299). In 10,276 Kaiser Permanente members examined at ages 40–45, midlife obesity (BMI ≥30) carried HR 1.74 (1.34–2.26) for later dementia and overweight HR 1.35 (1.14–1.60), independent of vascular comorbidity, with a parallel gradient on skinfold thickness (Whitmer 2005, PMID 15863436). These are the observational foundations of the PAF arithmetic; they are also the reason a trial that starts after age 70 is a test of the wrong decade.

That is an observed, population-scale, decades-long reduction in risk — achieved by education systems and cardiovascular medicine, not by a dementia-prevention programme. It is the best existing evidence that dementia risk is modifiable and simultaneously evidence that the effective interventions operate over decades and outside the health system's dementia services.

Against this, GBD attributes essentially all of the projected 2019→2050 growth in dementia cases (57.4 million → 152.8 million) to population growth and ageing, with age-standardised prevalence flat (global change 0.1%, 95% UI −7.5 to 10.8) (GBD 2019 Dementia Forecasting Collaborators 2022, PMID 34998485), and GBD 2016 attributed 6.4 million (95% UI 3.4–10.5) of 28.8 million dementia DALYs to its four modelled risks — high BMI, high fasting plasma glucose, smoking and sugar-sweetened beverages (PMID 30497964). Both statements are true: risk per person is falling in some populations while the number of people at risk rises faster.

Recent trials make the intervention contrast sharper

EXERT randomised 296 sedentary adults with amnestic MCI to moderate/high-intensity aerobic training or lower-intensity stretching/balance for 18 months. Attendance was 81% and 87%, but the primary cognitive trajectory did not differ (regression −0.078, SE 0.074; P=0.3); both groups were cognitively stable at 12 months and lost a mean 0.51% hippocampal volume (Baker 2025, PMID 40271888). The trial does not show that exercise is ineffective; it shows that intensity did not separate from an active, supported movement control.

A much smaller 20-week trial of 51 people with MCI or early AD reported between-group differences favouring an intensive diet/exercise/stress-management/social-support programme on CGIC (P=0.001), CDR-SB (P=0.032) and global CDR (P=0.037), with ADAS-Cog borderline (P=0.053) (Ornish 2024, PMID 38849944). Its size, short duration, wait-list control and bundled intervention make component attribution and durability uncertain. These two trials point in opposite directions partly because they ask different questions: supported exercise intensity versus an intensive multidomain package against usual care.

Biomarker subgrouping has not yet rescued equivocal prevention trials. In MAPT, apparent multidomain-plus-omega-3 benefits among 161 plasma-amyloid-positive participants lost significance after multiplicity adjustment in intention-to-treat analyses, though the corresponding per-protocol comparison (n=154) remained significant at 12 and 36 months after adjustment; no effect survived at 60 months in either analysis, and the authors call the finding a trend requiring confirmation (Delrieu 2023, PMID 37872582). In 527 MAPT participants, baseline p-tau181 predicted decline but neither changed with intervention nor identified a cognitive responder subgroup; certainty is further limited by the unexpected fall in p-tau181 among those with high baseline values (Coley 2024, PMID 38310892). A biomarker is not validated as a treatment-effect surrogate merely because it is prognostic.

How to state the prevention claim honestly

  1. Association-based potential: about one-third of dementia is statistically attributable to seven modifiable factors after adjusting for overlap (32.0%, 95% CI 26.6–37.5) (PMID 38824956).
  2. Trial-demonstrated effect: small cognitive-composite differences in enriched populations over 2 years (FINGER, US POINTER); no demonstrated reduction in dementia incidence from any multidomain trial; one positive single-factor trial for MCI (SPRINT MIND) and one positive natural experiment (zoster vaccine).
  3. The gap is not evidence that prevention fails. The trials test the wrong-aged population for too short a period against increasingly good usual care, and they are powered for cognitive composites rather than dementia.
  4. Nor is it evidence that prevention works. No randomised trial has yet reduced dementia incidence through lifestyle modification.

Both statements have to be held at once; this knowledge base does not collapse them.

Open questions

  • Would a multidomain intervention starting in midlife — the window in which vascular risk predicts amyloid (Gottesman 2017, PMID 28399252) — reduce dementia incidence, and can such a trial be run at all?
  • Is US POINTER's 0.029 SD/year difference clinically meaningful, and would it accumulate or plateau over 10 years (Baker 2025, PMID 40720610)?
  • Does hearing intervention work in higher-risk populations, as the ACHIEVE interaction suggests, and what is the mechanism — sensory input, social engagement, or shared vascular cause (Lin 2023, PMID 37478886)?
  • Why is the zoster-vaccine effect stronger in women, and does it reflect herpesvirus suppression or non-specific immune effects (Eyting 2025, PMID 40175543)?
  • Do prevention trials in low- and middle-income countries, where PAFs are higher, produce larger effects (Stephan 2024, PMID 38824956)?
  • How much of the 13%-per-decade incidence decline is education, how much cardiovascular treatment, and is any of it still available to capture (Wolters 2020, PMID 32611641; Skoog 2017, PMID 28733627)?
  • Can biomarker outcomes (plasma p-tau217, amyloid PET) shorten prevention trials enough to make midlife intervention testable, and would biomarker change be an acceptable surrogate (see fluid biomarkers)?
  • Should blood-pressure targets for cognitive outcomes differ from cardiovascular targets, given SPRINT MIND's MCI benefit and the greater hippocampal volume loss in the intensive arm (PMID 30688979; see vascular and metabolic contributions)?

References

  1. Livingston G, et al. Dementia prevention, intervention, and care: 2024 report of the Lancet standing Commission. Lancet. 2024;404:572-628. PMID 39096926.
  2. Stephan BCM, et al. Population attributable fractions of modifiable risk factors for dementia: a systematic review and meta-analysis. Lancet Healthy Longev. 2024;5:e406-e421. PMID 38824956.
  3. Ngandu T, et al. A 2 year multidomain intervention of diet, exercise, cognitive training, and vascular risk monitoring versus control to prevent cognitive decline in at-risk elderly people (FINGER): a randomised controlled trial. Lancet. 2015;385:2255-63. PMID 25771249.
  4. Baker LD, et al. Structured vs self-guided multidomain lifestyle interventions for global cognitive function: the US POINTER randomized clinical trial. JAMA. 2025;334:681-691. PMID 40720610.
  5. Andrieu S, et al. Effect of long-term omega 3 polyunsaturated fatty acid supplementation with or without multidomain intervention on cognitive function in elderly adults with memory complaints (MAPT): a randomised, placebo-controlled trial. Lancet Neurol. 2017;16:377-389. PMID 28359749.
  6. Moll van Charante EP, et al. Effectiveness of a 6-year multidomain vascular care intervention to prevent dementia (preDIVA): a cluster-randomised controlled trial. Lancet. 2016;388:797-805. PMID 27474376.
  7. SPRINT MIND Investigators for the SPRINT Research Group. Effect of intensive vs standard blood pressure control on probable dementia: a randomized clinical trial. JAMA. 2019;321:553-561. PMID 30688979.
  8. Lin FR, et al. Hearing intervention versus health education control to reduce cognitive decline in older adults with hearing loss in the USA (ACHIEVE): a multicentre, randomised controlled trial. Lancet. 2023;402:786-797. PMID 37478886.
  9. Eyting M, et al. A natural experiment on the effect of herpes zoster vaccination on dementia. Nature. 2025;641:438-446. PMID 40175543.
  10. Kuźma E, et al. Stroke and dementia risk: a systematic review and meta-analysis. Alzheimers Dement. 2018;14:1416-1426. PMID 30177276.
  11. Gottesman RF, et al. Association between midlife vascular risk factors and estimated brain amyloid deposition. JAMA. 2017;317:1443-1450. PMID 28399252.
  12. Wolters FJ, et al. Twenty-seven-year time trends in dementia incidence in Europe and the United States: the Alzheimer Cohorts Consortium. Neurology. 2020;95:e519-e531. PMID 32611641.
  13. Skoog I, et al. Decreasing prevalence of dementia in 85-year olds examined 22 years apart. Sci Rep. 2017;7:6136. PMID 28733627.
  14. GBD 2019 Dementia Forecasting Collaborators. Estimation of the global prevalence of dementia in 2019 and forecasted prevalence in 2050. Lancet Public Health. 2022;7:e105-e125. PMID 34998485.
  15. GBD 2016 Dementia Collaborators. Global, regional, and national burden of Alzheimer's disease and other dementias, 1990-2016. Lancet Neurol. 2019;18:88-106. PMID 30497964.
  16. Cummings JL, et al. Globalization of Alzheimer's disease clinical trials: current characteristics and future goals. Int Psychogeriatr. 2025;37:100108. PMID 40555627.
  17. Barnes DE, et al. The projected effect of risk factor reduction on Alzheimer's disease prevalence. Lancet Neurol. 2011;10:819-28. PMID 21775213.
  18. Norton S, et al. Potential for primary prevention of Alzheimer's disease: an analysis of population-based data. Lancet Neurol. 2014;13:788-94. PMID 25030513.
  19. Satizabal CL, et al. Incidence of dementia over three decades in the Framingham Heart Study. N Engl J Med. 2016;374:523-32. PMID 26863354.
  20. Kivipelto M, et al. Midlife vascular risk factors and Alzheimer's disease in later life: longitudinal, population based study. BMJ. 2001;322:1447-51. PMID 11408299.
  21. Whitmer RA, et al. Obesity in middle age and future risk of dementia: a 27 year longitudinal population based study. BMJ. 2005;330:1360. PMID 15863436.
  22. Baker LD, et al. Effects of exercise on cognition and Alzheimer's biomarkers in a randomized controlled trial of adults with mild cognitive impairment: the EXERT study. Alzheimers Dement. 2025;21:e14586. PMID 40271888.
  23. Ornish D, et al. Effects of intensive lifestyle changes on the progression of mild cognitive impairment or early dementia due to Alzheimer's disease. Alzheimers Res Ther. 2024;16:122. PMID 38849944.
  24. Delrieu J, et al. Cognitive impact of multidomain intervention and omega 3 according to blood Aβ42/40 ratio: a subgroup analysis from the randomized MAPT trial. Alzheimers Res Ther. 2023;15:183. PMID 37872582.
  25. Coley N, et al. Plasma p-tau181 as an outcome and predictor of multidomain intervention effects. Lancet Healthy Longev. 2024;5:e120-e130. PMID 38310892.