Brain Health

Around 45% of dementia cases are tied to things that can be changed, and the surprise is how few of them are about the brain: hearing, cholesterol, blood pressure, vision, sleep and the company you keep. Treat the heart and the senses, and the brain mostly follows.

Get your hearing and eyes checked and fix what you find, treat blood pressure and cholesterol from midlife, avoid type 2 diabetes, move every day, sleep regularly and get checked for sleep apnea, stay socially connected, and keep learning things that are hard for you. Skip brain-training apps and memory supplements. Most of this evidence is observational: the 2024 Lancet Commission's map of 14 modifiable risk factors is a synthesis of cohort studies,[1] and only one large randomised trial built around dementia as its main outcome has shown fewer diagnoses — blood-pressure lowering, in a village-level trial in rural China.[2]

Nearly half of dementia traces to 14 risks, mostly outside the brain

Ordered by their share of cases, the Commission's 14 are hearing loss and high LDL ("bad") cholesterol at 7% each; less education and social isolation at 5%; depression, traumatic brain injury and air pollution at 3%; physical inactivity, smoking, type 2 diabetes, hypertension and untreated visual loss at 2%; and obesity and excessive alcohol at 1%. Cholesterol and vision loss were added in 2024.[3] Each share is the proportion of dementia that would not occur if that factor vanished across the whole population, assuming every link is causal. It is not a personal risk estimate. Dementia prevention carries the full table, with each factor's risk multiplier and how common it is.

The ranking follows how common a factor is, not how dangerous it is. Hearing loss and high cholesterol tie at the top on ordinary risk multipliers, 1.4× and 1.3×, because 59% and 77% of the relevant population have them. Depression carries by far the largest individual risk (2.2×) but affects 7% of people, so it contributes 3%. Hypertension's individual effect is modest (1.2×), but a small share does not mean a small payoff from treatment, as the next section shows.

Ten of the 14 are midlife factors, one early-life and three late-life. The window that matters most is roughly ages 45 to 65.

The estimates may be generous. Re-tested inside a single cohort of 5,526 Americans, only three of the 14 survived — and midlife hypertension came out protective, itself a reason to distrust the re-analysis. That work is a conference abstract, not a peer-reviewed paper: a caution flag, not a refutation.[4]

One confound runs through almost all the observational evidence on this page: the silent early phase of dementia depresses engagement for years before diagnosis. In 28-year follow-up of 10,308 British civil servants, physical activity began declining up to nine years before diagnosis, and across the full follow-up there was no association between activity and dementia at all; the authors concluded that earlier positive findings likely reflected reverse causation.[5]

The brain is downstream of the heart

The biology behind heart attacks, strokes and type 2 diabetes also drives much of cognitive decline. The cochlea runs on a dense capillary bed that needs continuous blood flow, the usual explanation for why hearing loss travels with cardiovascular disease and diabetes.[6] The retina is the only place blood vessels can be photographed without cutting anyone open. So the highest-leverage brain interventions are mostly not brain-specific, and the brain-specific ones — hearing, vision, demanding learning — stack on top of cardiovascular care rather than replacing it.

Blood pressure is the one lever with a positive dementia trial. In rural China, 33,995 adults aged 40 and over with uncontrolled high blood pressure were randomised by village to a programme in which community health workers started and adjusted drugs towards below 130/80 mmHg, or to usual care. Over four years systolic pressure ended 22 mmHg lower and dementia about 15% lower, with fewer serious adverse events, not more. It was open-label, in a single population, with a bigger blood-pressure gap than most real-world care achieves.[7] Earlier, SPRINT-MIND tested a systolic target below 120 against below 140 mmHg: it cut mild cognitive impairment by about a fifth and the combined impairment-or-dementia endpoint by about 15%, while dementia alone moved the right way but stayed compatible with no effect, in a trial probably underpowered for it.[8] See Blood pressure.

Cholesterol and blood sugar are consistent risk links, but no trial has yet shown that treating them prevents dementia. High LDL ties hearing loss for the largest share of cases. Well-controlled diabetes does much less damage than uncontrolled, though intensive control is no better for dementia than standard control, and driving blood sugar near-normal with insulin or sulfonylureas is risky. Take a statin when your heart needs one, not late in life to prevent dementia. See Lipid management.

Exercise improves thinking skills in trials but has not been shown to prevent dementia. One small trial found aerobic exercise enlarged the hippocampus, the brain's main memory structure,[9] but pooled trials have not confirmed it;[10] for resistance training, one 12-month trial in older women improved a single executive-function measure.[11] In observational data, prolonged sitting tracks brain atrophy whether or not you meet the exercise guidelines, and about 7,000 steps a day goes with 38% lower dementia incidence than 2,000.[12] See Exercise and Sitting.

Bundling the levers helps a little. FINGER, a Finnish trial combining diet, exercise, cognitive training and vascular-risk monitoring in 1,260 at-risk older adults, improved cognition over two years by a small margin.[13] Its American successor, U.S. POINTER, pitted a structured against a self-guided version in 2,111 adults: both arms improved, the structured one slightly more on global cognition, a gain that came from executive function, with memory showing essentially nothing. There was no do-nothing arm, and the Alzheimer's Association funded and helped run it.[14] The benefit was the same in carriers and non-carriers of the APOE ε4 risk gene, which is the practical answer to a family history: the levers look the same either way.

Hearing: the biggest single share, an unproven fix

Hearing loss ties high LDL cholesterol as the largest modifiable midlife risk factor, at a rounded 7% share,[15] but nobody has shown that treating it protects the brain. ACHIEVE, the only randomised trial with cognitive decline as its primary endpoint, enrolled 977 older adults and missed it: hearing intervention did not slow cognitive decline overall. The much-quoted 48% reduction comes from a prespecified subgroup of 238 older, faster-declining participants inside a null trial — a lead, not a result.[16]

The encouraging data are observational. In 2,777 Australians aged 70 and over with self-reported hearing problems, analysed to imitate a trial but not one, dementia over seven years occurred in 5.0% of those prescribed hearing aids against 7.5% of those not — roughly 25 fewer diagnoses per 1,000 people — with risk tending to fall the more the aids were worn, an imprecise contrast. The authors cannot rule out confounding, and the benefit showed in diagnoses rather than in sharper day-to-day thinking.[17] Treat hearing loss for conversation, work and social life; any brain benefit comes on top.

The link with dying sooner is weaker than headlines suggest. Pooling 36 cohorts, adults with hearing loss died at about 21% higher rates, but the cohorts disagree sharply and across the five European ones the link is flat.[18] Among 1,863 US adults with measured hearing loss followed a median 10.4 years, regular hearing-aid users died at about three-quarters the rate of never-users and occasional users showed no benefit — observational, and healthier people may simply be likelier to wear aids.[19]

A normal audiogram can miss the problem. The standard test stops at 8 kHz; extended high-frequency audiometry tests above that ceiling. In a single 263-adult analysis from a Swedish tinnitus cohort, still a non-peer-reviewed preprint, thresholds above 8 kHz explained about a third of the variance in speech-in-noise performance against roughly a fifth for the standard test — an apparent advantage, not a decisive one.[20] If you struggle in restaurants but pass a standard test, ask for it; the 2024 American Academy of Otolaryngology–Head and Neck Surgery Foundation (AAO-HNSF) guideline doesn't mention it.[21]

Noise damage is dose-dependent, and the scale is logarithmic: safe weekly exposure runs from about 40 hours at 80 decibels (a doorbell) to roughly 20 minutes at 100 decibels (a hair dryer). Filtered earplugs around loud noise are a cheap, sensible upgrade. Hearing also covers the diabetes link nobody screens for, ear-damaging chemicals, why no supplement protects hearing, and why gene therapy will not reach age-related hearing loss any time soon.

Vision: fix the deficit first, admire the frontier second

Untreated vision loss joined the modifiable list in 2024 as one of the smaller levers, at about 2% of cases.[22] What carries the dementia signal is how badly you see, not the eye diagnosis: in US health records, blindness raised five-year Alzheimer's risk by about 29%, while dry age-related macular degeneration without severe impairment did not — and shared disease, sensory deprivation and reverse causation all remain plausible explanations.[23] No trial has yet tested whether correcting vision protects thinking.

The unglamorous conditions carry the strongest evidence. Cataract is the leading cause of blindness worldwide, and in a randomised trial first-eye surgery in women over 70 cut the rate of falling by about a third.[24] A smaller second-eye trial saw a similar drop that missed statistical significance, which its authors read as uncertain rather than absent. The dementia link is weak: one cohort found roughly 29% lower dementia risk after cataract surgery, but healthy-patient selection remains possible and later meta-analyses disagree.[25] Glaucoma is silent until advanced, and the vision it takes does not come back. The LiGHT trial of 718 people established an outpatient laser as a reasonable first-line alternative to daily drops,[26] and at six years about 70% of laser-treated eyes were still at target pressure with no drops and no surgery.[27] For dry eye, 3,000 mg a day of fish oil did no better than placebo in the large DREAM trial.[28] All of this needs a dilated eye exam, not a frontier scan.

The frontier is interesting and not ready. AI models read photographs of the back of the eye to estimate retinal age. In 35,913 UK Biobank adults, each year by which retinal age exceeded real age went with about 2% higher all-cause mortality,[29] a small signal whose plausible range reaches down to no effect; a related model, RetiAGE, scores the chance a person is over 65 and is also used to rank mortality risk.[30] The models degrade sharply outside the population they were trained on,[31] no professional body uses them for risk stratification, and no eye exam will print you one.

Lutein, red light and blue light. Macular pigment — lutein and zeaxanthin, from leafy greens and egg yolks — filters short-wavelength light and goes with better contrast and glare recovery, though the supplement trials showing a gain are small and often industry-run; the food route has no downside, but in the largest test, more than 3,000 older adults with macular degeneration, the supplement did not improve cognition, and nor did the trial's omega-3 arm.[32] Red light therapy depends on the device. Home 670 nm panels are unproven: an uncontrolled 12-month pilot found no structural benefit.[33] Clinic-delivered multiwavelength treatment for dry macular degeneration gained about six letters of acuity and slowed progression to geographic atrophy in a double-masked, sham-controlled trial, but both its trials have industry ties and no independent long-term replication: a question for a retina specialist, not a settled therapy.[34] Clear blue-light glasses probably make little or no difference to sharpness of vision and may not ease eye strain, by a Cochrane review of 17 randomised trials; it could not tell whether they affect sleep and found no randomised evidence on macular health.[35]

Vision covers cataract, glaucoma, dry eye, myopia and diabetic eye disease in full, how often to get examined, and who actually needs an eye supplement. Vision decline with age covers the changes nearly everyone notices after 40, why eye exercises don't reverse them, and what does help: reading glasses, eye drops, lighting and night-driving habits.

Sleep: treat apnea, and don't chase deep sleep

Obstructive sleep apnea is linked to about 40% higher dementia risk, though few studies accounted for obesity, and it is heavily underdiagnosed. Treat it for its proven benefit on daytime sleepiness; trials have not yet shown that CPAP (continuous positive airway pressure) slows cognitive decline.

Deep (slow-wave) sleep is said to be when the brain's waste clearance works hardest at removing beta-amyloid, and a Framingham follow-up linked each percentage-point yearly decline in deep sleep to about 27% higher dementia risk.[36] Both claims are weaker than they sound. The main human causal evidence is a single 17-person crossover study that moved beta-amyloid but not tau, the other core Alzheimer's protein,[37] and carrying APOE ε4 predicted faster loss of deep sleep — so early disease may be eroding the sleep rather than the reverse. Total sleep on a regular schedule is a better-supported target than deep sleep. See Sleep architecture and Sleep.

Diet and supplements: the famous numbers shrank

The "half the Alzheimer's incidence" figure for the MIND diet came from a cohort of 923 older adults;[38] a cohort of about 130,000 people followed for decades puts the association at about 14% lower dementia risk,[39] and the only randomised trial of the pattern, in 604 at-risk older adults over three years, was a clean null on cognition and brain imaging.[40] Follow it for the heart; the dementia-specific claim is not established.

Ultra-processed food points the other way. A 2026 systematic review found only five observational studies on it and cognition, three of which linked it to worse cognitive outcomes; whether that is the processing itself or the poor overall diet these foods come with is still open, and the reviewers call for future studies to measure overall diet quality.[41]

Cognitive supplements — omega-3, B vitamins, ginkgo, curcumin — are null or inconsistent on cognitive endpoints, and ginkgo failed two large prevention trials. The DO-HEALTH trial is sometimes cited the other way, which gets it backwards: it missed all six of its pre-specified primary endpoints, cognition among them; what moved was a post-hoc analysis of methylation aging clocks in 777 of its 2,157 participants — surrogate markers, not cognition and not dementia.[42] See Omega-3. No supplement combination has any dementia-endpoint evidence at all.

Cognitive engagement: build reserve, and don't write off people

Education is the one cognitive item the Commission lists, and the idea behind it is cognitive reserve: autopsy studies routinely find extensive Alzheimer's-type damage in people who had no symptoms in life, and little in people who clearly had dementia. The brain holds spare capacity between damage and observable decline, and lifelong intellectual challenge is thought to build it. A recent, unsettled coinage for what it protects is brainspan, the years during which the brain stays efficient enough to support agency and coherent self-regulation.[43]

A meta-analysis of 27 long-term studies puts more early-life education at about 18% lower dementia risk, complex midlife work at about 9%, and late-life engagement at about 19%,[44] and about 28% of education's protective effect runs through the more complex jobs it makes possible.[45] All of it is observational, and the late-life figure is the most exposed to the reverse causation described above. Reserve buys time, not immunity: once the spare capacity runs out, decline tends to be faster in people who had more of it, because more damage built up first.

Don't write off socialising. Splitting that late-life category, social connection came out strongest — roughly 30% lower dementia risk, against about 9% for cognitive activity.[46] Small randomised trials, nearly all at high risk of bias, suggest social programmes improve thinking scores, partly in people already impaired.[47] One short trial found that learning demanding new skills improved episodic memory where socialising about familiar topics did not, so the two are not substitutes; combining them — group classes, ensemble music, collaborative crafts, conversation circles — has support from both directions.

Brain-training apps make you better at the app. In 2016 the makers of Lumosity agreed to pay $2 million to settle US Federal Trade Commission charges that they advertised, without scientific backing, that the app would improve work and school performance and protect against dementia.[48] More than 70 cognitive scientists convened by the Stanford Center on Longevity and the Max Planck Institute concluded that the literature does not support claims that brain games improve everyday cognition or prevent brain disease;[49] a rival letter drew 133 signatures, several with commercial ties. Cochrane favours the sceptical reading: computerised training gives small immediate benefits of uncertain clinical importance, with no evidence the effect on global cognition persists at twelve months — only one small trial looked that far.[50]

The most-quoted exception is narrower than advertised. ACTIVE randomised 2,832 cognitively healthy adults aged 65 and over to memory, reasoning or speed-of-processing training, or no contact.[51] At ten years the reasoning and speed groups still did better on the skill each had trained, while memory training's gain had faded.[52] Twenty years on, speed trainees who also did booster sessions had about 25% lower dementia incidence — but boosters went only to trainees who had stuck with the first sessions, so that comparison is not randomised; without boosters there was no benefit, and none of the whole-arm comparisons against control reached statistical significance.[53] The trial also carries the pillar's only harm signal for a cognitive intervention: among participants with 12 years of education or less, memory and speed training were associated with higher 20-year mortality, on imprecise, unreplicated estimates.[54]

Two popular claims deserve less confidence than they get. Crosswords beat a computerised trainer in one trial, but its 107 participants already had mild cognitive impairment, there was no do-nothing group, and the advantage on the main memory-and-thinking test was small, about 1.4 points on a 70-point scale.[55] And the familiar claim that bilingualism delays dementia by four to five years overstates the pooled estimate of about three and a half years from studies of age at onset;[56] in prospective studies bilinguals had essentially the same dementia risk, a 4% difference well inside the range expected from chance,[57] and a 2026 snapshot of older adults found no advantage in any cognitive domain.[58] Learn a language for its own sake, not as insurance.

Cognitive engagement covers occupational complexity, retirement, music, the multidomain trials and the dose-and-novelty question.

Vaccines, drugs and purpose

The shingles vaccine probably lowers dementia diagnoses, by a disputed amount. In Wales, a birthdate cut-off for eligibility (2 September 1933) created a natural experiment: new dementia diagnoses over seven years fell by 3.5 percentage points, about a fifth in relative terms.[59] The design mimics randomisation, and similar analyses in Australia and Canada found the same direction.[60][61] But so few eligible people were vaccinated that critics calculate some of these results imply implausibly large protection in those who were.[62] The natural experiments used the older live vaccine; for the now-standard recombinant one the evidence is observational so far.[63] There is no randomised trial, but the vaccine is recommended against shingles anyway.

Anti-amyloid antibodies (lecanemab, donanemab) modestly slow early symptomatic Alzheimer's and are not prevention for healthy adults. Over 18 months lecanemab slowed decline by 0.45 points on an 18-point rating scale, about 27%, and brain swelling showed on MRI in about 13% of lecanemab and 24% of donanemab recipients, mostly without symptoms, with small brain bleeds counted separately on top.[64] A slower donanemab dose build-up lowered swelling to about 16%.[65]

Purpose is an observational association. In a 28-year study of 4,632 adults, purpose measured at 63 and 70 predicted modestly lower odds of a dementia-range score on a telephone screen at about 80, while purpose at 52 did not clearly do so; the link could run backwards, since purpose drifts down before memory problems show.[66] Late midlife — retirement, empty nest, role loss, early widowhood — is a sensible time to pay attention, though no special window has been proven. See Purpose.

Practical brain-health checklist

  1. Treat hearing loss. A baseline audiogram in your forties — our suggestion; the AAO-HNSF guideline starts at 50 — and sooner with any concern. Struggling in noise despite a normal test: ask for extended high-frequency audiometry. Wear the aids you're prescribed.
  2. Get a comprehensive dilated eye exam by 40 and treat any correctable loss — glasses, cataract surgery, glaucoma treatment, low-vision rehabilitation. Eat leafy greens daily: the pattern, not the pill. Take an AREDS2-type formula, the version without beta-carotene, only if an eye doctor has diagnosed intermediate macular degeneration or advanced disease in one eye.
  3. Treat blood pressure and cholesterol in midlife. The Commission's target is systolic below 130 mmHg from age 40 if tolerated — stricter than routine practice, looser than SPRINT-MIND's below 120.
  4. Stay metabolically healthy. Track HbA1c (the three-month blood-sugar average), a lipid panel and waist circumference in midlife rather than waiting.
  5. Train both ways, and don't sit through the day; work towards about 7,000 steps.
  6. Sleep 7–8 hours on a regular schedule, and get screened for sleep apnea.
  7. Eat MIND-pattern: leafy greens daily, berries twice a week or more, fish, nuts, olive oil, whole grains, beans, poultry; limit red and processed meat, butter, fried foods, sweets and ultra-processed food. See Dietary patterns.
  8. Stay socially connected — real relationships, regular contact, groups you actually turn up to.
  9. Keep learning things that are genuinely hard, ideally with other people, and rotate as a skill becomes automatic.
  10. Treat depression, at any age. It is both the highest individual risk of the 14 and an early symptom — a reason to treat it, not dismiss it.
  11. Plan your next role before the current one ends.
  12. Don't smoke; drink minimally; reduce air-pollution exposure — run a true-HEPA air cleaner in the bedroom every night. It more than halves the fine particles you breathe there, though no trial has yet shown a health benefit; it matters most if you live near a major road. See Air purifiers and Environmental toxins.
  13. Wear seatbelts and helmets; avoid repeated head impacts.
  14. Take the shingles vaccine at the age it's offered.

What's overrated

  • Brain-training apps in place of demanding real-world learning: small immediate gains, no evidence they last, no far transfer.
  • Cognitive supplements and "anti-Alzheimer's" stacks: null or inconsistent, with no dementia-endpoint evidence.
  • Clear blue-light glasses for eye strain or sleep.
  • Crosswords and bilingualism as prevention. Fine things to do; neither has the evidence its reputation implies.
  • Anti-amyloid antibodies as prevention: a treatment for early symptomatic disease with a real harm profile.
  • Treating cognitive decline as inevitable past a certain age. Late-life engagement tracks dementia risk about as strongly as early-life education does, though that signal is the most open to reverse causation.

For the full ordered action list drawn from every pillar, see the healthspan long list.

— § —