A Guide from Dr. Starsiak · Prevention
For most of my career, the honest answer to "can Alzheimer's be prevented?" was "probably, partly, but nobody has proven it." That answer changed in the last few years — and it changed in a way most people haven't heard about. This is the whole protocol, ranked by leverage, with the evidence named.
Why I believe prevention is real
In 2025, a randomized trial of nearly 34,000 people in rural China — the CRHCP trial, published in Nature Medicine — showed for the first time that an intervention reduced actual dementia diagnoses: village health workers treating blood pressure to target cut all-cause dementia by about 15% over four years. Not a surrogate marker. Not a memory-test score. Dementia itself, in a randomized trial — done with ordinary generic blood pressure medicines and consistent follow-up. Nothing exotic.
Around the same time, the 2024 Lancet Commission concluded that roughly 45% of dementia risk worldwide is attributable to fourteen modifiable factors — hearing loss, blood pressure, LDL cholesterol, physical inactivity, social isolation, untreated vision loss, and others. Two large randomized trials — FINGER in Finland and U.S. POINTER (2025) — showed that structured multi-part lifestyle programs measurably bend the cognitive trajectory of at-risk older adults. The individual effects are modest over two years; the point is that they are real, they compound, and they were achieved with things any household can do.
So the frame of this protocol is not "here are some healthy habits, they can't hurt." It is: the trajectory toward dementia is modifiable, several of the levers are proven or close to it, and most of them are cheap. What no one can promise is immunity. Genetics — especially APOE4 — load the dice. What the evidence supports is shifting your odds meaningfully and pushing back the age of onset. Delaying dementia by even a few years changes whether many people ever experience it.
One more thing this page is honest about: lifestyle interventions mostly seem to buy resilience rather than a cleaner brain. In the POINTER imaging substudy, two years of intensive lifestyle change improved cognition without visibly changing amyloid, tau, or brain volume — and the people who benefited most were those whose brains had the least reserve to spare. A separate Harvard study using step counters found physical activity slowed the accumulation of tau itself in people with early amyloid. Either way — resilience or pathology — the clinical translation is identical: start as early as you can, and never stop.
You may be the person this page matters for most. In the Cache County study, people whose husband or wife developed dementia went on to develop dementia themselves at roughly six times the rate of other married people their age — an effect in the neighborhood of carrying an APOE4 gene. The confidence interval on that number is wide, and follow-up research suggests the driver is not caregiving itself but unbuffered caregiver strain: chronic stress, abandoned exercise, broken sleep, and isolation. Caregivers who get support do not show the excess risk — in some studies, supported caregivers actually outlive their peers, possibly because caring for someone is a powerful form of purpose. The practical conclusion: the caregiver is a second patient. Every item in this protocol applies doubly to you, and the stress-and-breath section further down was written with you specifically in mind.
Two principles govern how I weigh these. First, absence of trials usually reflects funding, not merit — a dementia-prevention outcome trial costs hundreds of millions and takes a decade, and that money follows patents. An unpatentable mineral, mushroom, or breathing practice isn't beaten in a trial; it's simply never entered into one. So "unproven" and "worthless" are not the same word. Second, what legitimately breaks a tie is a specific safety signal — a bleeding interaction, a liver signal, a contaminated supply chain — and where one exists, I name it in the exact place it matters. Everything below is ranked by leverage: I refuse to bury a proven 15% risk reduction under a pile of interesting maybes.
The short list
Do daily. Walk 7,000–10,000 steps including one 20–30 minute brisk stretch (a pace where talking takes effort) — and if that's out of reach, know that measurable protection starts under 4,000 steps. Sleep 7–8 hours; get outside into daylight within an hour or two of waking; keep the bedroom dark and cool. Eat from the food pattern below: greens every day, extra-virgin olive oil at least a half-tablespoon, berries and mushrooms through the week, fatty fish twice weekly, live fermented food, real fiber; minimal ultra-processed food and little to no alcohol. Work your mind at something real and hard — learning, building, teaching, performing — not just consuming. Brush twice (electric), clean between teeth once. Practice 10–20 minutes of slow breathing or conscious relaxation — non-negotiable for stressed caregivers. And the basic supplement layer, if it fits your labs and medications: a senior multivitamin, omega-3 (about 1–2 g EPA+DHA, DHA-weighted), vitamin D if you were low, creatine 5 g, and — after reading the lithium section carefully — low-dose lithium.
Do weekly. Strength train twice. One coordinative-social session — dancing is the best-studied; a dual-task tai chi class is a close second. Sauna 3–4 times if you have access and your heart is up to it. At least three real social contacts outside the house, and one role where people count on you.
Do once — or once a year. Home blood pressure protocol; treat to target with your physician — the single best-proven item in this entire guide. Hearing test; correct any loss (over-the-counter aids count). Eye exam; don't postpone cataract surgery. Shingles vaccine (Shingrix) if 50+, and the annual flu shot. Screen for sleep apnea (30 seconds, below); test if positive. The lab panel near the end of this page — including homocysteine, which changes what supplements are worth taking (request an appointment and we'll run it together). Audit every medication and OTC product against the anticholinergic list below — especially "PM" sleep aids and bladder medicines. Pick this year's new skill.
If ten items is too many to start, start with three: blood pressure, the daily walk, and the hearing test. Those three carry more evidence than everything else combined. Add one new item a month.
The Twelve Foundations, ranked
Evidence: Established. This is the only intervention in medicine that has reduced dementia in a randomized trial. In CRHCP (Nature Medicine, 2025; ~34,000 participants), intensive treatment to below 130/80 — run by trained village health workers with free generic medications — reduced all-cause dementia by 15% over four years, with fewer serious adverse events in the treated group. It confirmed what SPRINT MIND (JAMA, 2019) had strongly suggested: targeting a systolic pressure near 120 rather than 140 cut mild cognitive impairment by 19%, with the benefit still visible seven years later.
What I'd have you do. Buy a validated upper-arm cuff (Amazon). Run a 7-day protocol twice a year: two readings each morning and evening, discard day one, average the rest. Bring me the average — request an appointment and we'll set your target together. We treat to an office target below 130/80 — and in robust patients I'll discuss the SPRINT-style target near 120 systolic, with standing pressures checked, because the trade-off at that target is real (lightheadedness, kidney and electrolyte monitoring). Consistency matters more than the specific drug; blood-pressure variability appears to be its own risk factor, which argues for steady daily dosing over heroic intermittent efforts. The quiet lesson of CRHCP: the active ingredient wasn't a novel drug. It was follow-through — somebody checking, adjusting, and not letting it slide. That is a role a spouse or family member can own. My blood pressure guide covers the non-drug levers in detail.
Evidence: Probable for cognition; Established for everything that drives it. No trial has yet proven exercise alone prevents dementia — but the observational and biomarker evidence is deep and consistent, and exercise is a core component of both FINGER and POINTER, the two positive multidomain trials. The numbers worth knowing: in ~78,000 UK adults wearing accelerometers (JAMA Neurology, 2022), dementia risk bottomed out around 9,800 steps/day — about half the risk of the least active — but half of the maximum benefit arrived by ~3,800 steps, and the strongest single predictor was having ~30 minutes of brisk cadence somewhere in the day. A 2025 Harvard study in Nature Medicine found that in older adults with early amyloid, those walking more accumulated tau more slowly — benefit beginning around 3,000 steps/day. In already at-risk brains, the effective dose is startlingly low. Sitting is its own exposure: above ~10 sedentary hours/day, risk climbs regardless of whether you also exercised (JAMA, 2023).
Resistance training earned its own seat: in the SMART trial, six months of twice-weekly progressive strength training in people with mild cognitive impairment improved cognition, with gains persisting a full year after the program ended — and the cognitive benefit tracked the strength gains themselves. And dancing deserves special mention: in the Bronx Aging study (NEJM, 2003), frequent dancing was the only physical activity associated with lower dementia risk, and small trials since show dance grows the hippocampus more than repetitive gym training. The likely reason: it is aerobic exercise, balance work, learning, music, and social contact simultaneously — a multidomain program disguised as a Friday evening.
What I'd have you do. Walk daily with a brisk 20–30 minute segment; lift twice a week — real, progressive resistance; my strength training guide shows you how, and we'll track your grip strength, a legitimate risk marker. Add one weekly session that makes you think while you move — dance, dual-task tai chi, pickleball. Cap sitting below 10 hours. If MCI is already present: prioritize regularity over intensity — four sessions a week of anything tolerable, ideally with other people, is what the trial evidence actually supports.
Evidence: Probable. Midlife short sleep (six hours or less) carried about 30% higher dementia risk over 25 years in the Whitehall II cohort. Disrupting deep slow-wave sleep for even one night measurably raises amyloid in spinal fluid. I'll be honest that the popular "the brain washes itself during sleep" mechanism is genuinely contested right now — a 2024 Nature Neuroscience mouse study found the opposite of what the original glymphatic work predicted — and I'm comfortable with that, because the human risk and biomarker data don't depend on which mechanism wins.
The item hiding inside this section is obstructive sleep apnea. Roughly 80% of American adults who have it are undiagnosed. It raises cognitive-impairment risk by roughly a third to a half in pooled cohorts, and in one cohort untreated apnea was associated with cognitive impairment arriving about ten years earlier — while treated patients tracked closer to people without apnea. Screening takes 30 seconds — the STOP-BANG questions (Snoring, Tiredness, Observed apneas, Pressure, BMI, Age over 50, Neck size, male sex). Three or more: get a home sleep study. If you sleep next to this person, you are the diagnostic instrument — witnessed pauses in breathing are a "yes."
What I'd have you do. Protect a 7–8 hour window. Morning outdoor light within 1–2 hours of waking (indoor light is 1–5% as bright as outside — this is the dose most people miss), dim warm light in the evening, bedroom dark and around 65–67°F. If insomnia is chronic, the first-line treatment is CBT-I — not a pill, and specifically not the anticholinergic "PM" products in the medication section below. Alcohol fragments exactly the deep sleep you're trying to protect. Melatonin, if used, works better low (0.3–1 mg in the early evening, for timing) than high. Side-sleeping is reasonable; mouth taping I'd skip — the trials are unimpressive and it's unsafe in exactly the people who tend to try it. My sleep guide goes deeper on all of this.
Evidence: Probable, with the largest population-level upside on the whole list. The Lancet Commission ranks midlife hearing loss as the single largest modifiable contributor to dementia worldwide (~7% of all cases). The ACHIEVE randomized trial (Lancet, 2023) gave hearing aids to older adults with untreated loss: across everyone, no significant effect over three years — but in the pre-specified higher-risk half, hearing intervention slowed cognitive decline by 48%. My honest framing: hearing aids aren't proven universal prevention; they most help the people most likely to decline — and they cost little to try in that spirit, since 2022 brought FDA-approved over-the-counter hearing aids at a tenth the old price. (A widely shared study claiming 42% dementia reduction from hearing aids was retracted for a coding error; I'm deliberately not citing it. ACHIEVE is the real evidence, and it's good enough.)
Vision: in a cohort of ~3,000 older adults, those who had cataract surgery had about 29% lower subsequent dementia risk — and the elegant control is that glaucoma surgery, which doesn't restore vision, showed no such benefit. The 2024 Lancet Commission added untreated vision loss to its official list. What I'd have you do: audiogram now, then periodically; treat any loss without stigma or delay — and protect the hearing you have (earplugs at concerts and power tools). Annual dilated eye exam. If cataract surgery is on the table, take it off the "someday" shelf.
Evidence: Probable — with one surprisingly durable trial. The brain-training industry earned its FTC fine; most apps make you better at the app. But two findings stand apart. First, the ACTIVE trial: about ten hours of a specific adaptive speed-of-processing training was associated with roughly 29% lower dementia incidence ten years later — and a 2026 twenty-year follow-up still showed a difference among those who completed the boosters. The p-values are fragile, so I hold this as promising-plus rather than proven — but ten hours is a trivially cheap dose, and this exact style of training was the cognitive component inside FINGER and POINTER. Second, the Synapse Project (Psychological Science, 2014): older adults randomized to 15 hours a week of productive learning — digital photography or quilting, real skills with progression — improved memory, with brain-efficiency changes still visible a year later; socializing and easy at-home activities did not. That matches the occupational data (cognitively demanding work → later, less dementia) and the retirement data (each year of later retirement, ~3% lower risk).
The principle I take from all of it: novelty + difficulty + meaning + output. Consuming — TV, scrolling, even light reading — does little. Producing — learning an instrument, a language, a craft that gets harder as you improve; teaching; performing; volunteering in a demanding role — is where the signal lives. What I'd have you do: one genuinely new skill per year, worked most days, with a deliverable — a recital, an exhibit, a conversation in the new language, a quilt. If you like structure, add adaptive speed training ~2 hours/week. Retire from your job if you wish; never retire from difficulty. And for caregivers: pick something you can do at the kitchen table.
Evidence: Probable. In the largest meta-analysis (~608,000 people, Nature Mental Health, 2024), loneliness carried a 31% higher dementia risk; poor social networks in other pooled data, ~59% higher. Two findings elevate this beyond "have friends." Purpose: in the Rush Memory and Aging Project, people with a strong sense of purpose were about 2.4 times more likely to remain Alzheimer's-free — and in the autopsy follow-up, purpose didn't reduce plaques and tangles; it let people function better at the same pathology. Purpose buys resilience. Being needed: the Experience Corps trial put older volunteers into elementary schools ~15 hours/week; over two years, control participants' hippocampi shrank on schedule — volunteers' grew. Obligation, structure, and mattering to someone are biologically active.
What I'd have you do. Three or more real interactions outside the household weekly. One standing group membership that carries obligation — choir, congregation, league, volunteering ≥2 hours/week, ideally intergenerational. Write down what you are for — your purpose in a sentence — and revisit it quarterly; it sounds soft and it is one of the better-replicated risk factors on this page. Widowhood and divorce are high-risk transitions — the year after a loss is precisely when this protocol should intensify, not lapse. Caregivers: your caregiving is purpose — the risk isn't the role, it's doing it isolated and unsupported. Accept the respite. Join the support group. That is not self-indulgence; it is dementia prevention for you.
Evidence: Probable for the pattern; Established for some components' cardiovascular effects. Full honesty first: the 2023 randomized MIND-diet trial in the NEJM was null — but it compared the MIND diet against an active healthy-diet control and both groups improved, so it tells us less than the headlines suggested. What we do have: the Rush cohorts (even moderate MIND adherence, ~35% lower Alzheimer's incidence; top adherence, ~53%), the PREDIMED randomized trial (Mediterranean diet with olive oil or nuts improved cognition while the low-fat control declined), and DIRECT-PLUS (a "green Mediterranean" diet slowed measurable brain atrophy on MRI over 18 months). Diet trials rarely achieve structural brain endpoints; that one did.
The specific foods with their own evidence: extra-virgin olive oil — ≥7 g/day (half a tablespoon) was associated with 28% lower dementia-related mortality in a 92,000-person Harvard cohort; make it your default fat (high-phenolic EVOO — any good early-harvest EVOO from the grocery works too). Leafy greens — one serving daily tracked with cognition ~11 years younger. Berries — ≥2 servings/week associated with ~2.5 years' delay in memory decline. Mushrooms — the sleeper: ≥2 servings/week associated with ~57% lower odds of MCI in a Singapore cohort, plausibly via ergothioneine, a dietary antioxidant with its own dedicated transporter into human tissue; oyster and shiitake are richest. Fatty fish 1–2×/week; nuts ~1 oz/day (walnuts best studied); eggs are fine and their choline is probably an asset. Fiber and fermented foods — a Stanford randomized study found six daily servings of fermented foods lowered 19 inflammatory markers in ten weeks, and a 2024 British twin trial found a cheap prebiotic fiber supplement (inulin/FOS, 7.5 g/day) improved memory-test performance in twelve weeks; target 25–30 g fiber/day plus daily live-culture foods — my gut health guide covers this ground in detail. Coffee and tea — 2–3 cups of caffeinated coffee/day tracked with ~18% lower dementia risk in a 2026 Harvard analysis (decaf didn't); green tea has consistent cohort support. Drink the beverage; skip high-dose green-tea-extract pills (real liver signal at >800 mg EGCG).
And the subtractions, which matter as much: ultra-processed food above ~20% of calories tracked with ~28% faster cognitive decline — keep it under a fifth of your plate. Alcohol — I won't launder this one: alcohol-use disorder is the strongest modifiable risk factor for early-onset dementia, and Mendelian-randomization data suggest no truly safe level for brain volume. The old J-curve was probably confounded. If you don't drink, don't start; if you do, less is better and none is fine. Cocoa pills — enjoy dark chocolate as food, but the big trial (COSMOS) found cocoa-flavanol pills did nothing for memory. The surprise winner in that same trial was the humble multivitamin — see the supplement shelf below.
Evidence: Probable. Alzheimer's has been called "type 3 diabetes" — an overstatement as a diagnosis, a useful heuristic as a warning. The Alzheimer's brain shows insulin resistance and glucose hypometabolism years before symptoms; midlife diabetes, midlife obesity, and even high-normal glucose track with later dementia; severe hypoglycemic episodes roughly double it — so in older adults, glycemic stability beats aggressive lowering. The 2024 Lancet Commission also added high LDL cholesterol as a midlife factor at 7% population-attributable risk — the same weight as hearing loss — and good recent data show statins do not harm cognition. Treating midlife LDL is brain care. A 2025 pause on GLP-1 drugs: the observational data looked spectacular, but the randomized EVOKE trials in early symptomatic Alzheimer's failed outright — these drugs are not a treatment for established disease; whether treating obesity and diabetes with them earlier prevents dementia remains open and plausible. If you're on metformin, stay on it and check B12 annually.
What I'd have you do. Know your fasting insulin and A1c, not just glucose; a triglyceride/HDL ratio above ~3 is a cheap early flag. Treat LDL in midlife on cardiovascular merits and count the brain as a second beneficiary. Midlife excess weight matters; unintended late-life weight loss is a different animal — often an early disease sign, not a virtue. My guides on blood sugar and cholesterol cover the how.
Evidence: Probable for the connection; the hygiene prescription is common sense with cohort support. This one still surprises people. Porphyromonas gingivalis, the keystone gum-disease bacterium, has been found in Alzheimer's brains, its toxic enzymes tracking with tau pathology (Science Advances, 2019); in mice, oral infection alone produced brain colonization and amyloid. Across 47 longitudinal studies, poor periodontal health tracks with cognitive decline; each lost tooth adds measurable risk — and, tellingly, denture wearers lose the association, suggesting restored chewing itself matters. Large Korean and Taiwanese insurance cohorts link regular professional cleanings and improving hygiene habits with lower subsequent dementia.
What I'd have you do. An electric oscillating-rotating brush (Amazon) twice daily; interdental cleaning once daily; professional cleaning every 6 months — every 3–4 if you have periodontitis, which should be treated as seriously as any other chronic inflammatory disease. Replace missing teeth. Xylitol gum after meals is a reasonable adjunct. Oil pulling, honestly placed: the small modern trials show it genuinely reduces plaque, gingivitis, and S. mutans counts, approaching chlorhexidine with slower onset — every endpoint ever measured is dental, no study has tested a brain outcome, and I won't imply otherwise. But the mechanistic bridge is straightforward, and its advantage over chlorhexidine is the harm profile. Swish gently 10–20 minutes, never swallow it, spit it in the trash, then brush as usual; skip it in young children or anyone with swallowing problems. It supplements brushing and flossing; it never replaces them. My dinacharya guide covers the technique.
Anticholinergic drugs. In a 3,400-person prospective cohort with pharmacy records (JAMA Internal Medicine, 2015), the highest cumulative exposure band — roughly three years of daily use at minimum doses of drugs like oxybutynin (bladder), chlorpheniramine, or doxepin — carried 54% higher dementia risk, with a clean dose-response. A nightly 50 mg diphenhydramine (Benadryl — the "PM" in most OTC sleep aids) reaches that band in well under three years. These drugs block acetylcholine — the exact neurotransmitter Alzheimer's depletes and Alzheimer's drugs try to boost. Nobody thinks of Tylenol PM as a brain drug. It is. Swaps exist for nearly all of them (oxybutynin → mirabegron; amitriptyline → almost anything; Benadryl-for-sleep → the sleep section above). Run every med and OTC product through the free ACB calculator (acbcalc.com); I want your total score under 3.
Benzodiazepines and Z-drugs (lorazepam, alprazolam, zolpidem…): the dementia-causation evidence is genuinely debated — but the falls, fractures, delirium, and crash data aren't debated at all. In older adults these should be tapered thoughtfully, not stopped abruptly, and rarely started. Chronic PPIs: evidence mixed; most long-term use has no active indication, and stepping down has other benefits (B12, magnesium, bone). Glucosamine: I used to consider this a harmless joint supplement. A June 2026 Nature Metabolism analysis associated it with roughly 25% higher progression from MCI to Alzheimer's, with supporting imaging and animal work. It's a retrospective association, not a verdict — earlier population data in healthy users pointed the other way — but with modest joint benefit at best (the NIH GAIT trial was null overall), the trade no longer makes sense. I've stopped recommending it; Theracurmin, below, is my replacement. Gabapentin, chronic opioids, chronic sedating antihistamines (switch to loratadine or fexofenadine): minimize; sedation load is cumulative.
Bring every bottle — prescription, OTC, supplement — to one appointment a year. One change at a time.
Evidence: Probable — and the shingles story is one of the most striking findings in modern preventive neurology. Wales rolled out the shingles vaccine with a strict birthdate cutoff: born on or after September 2, 1933, eligible; a week older, never eligible. That accident created a near-randomized experiment in 282,000 people. Result (Nature, 2025): vaccination reduced new dementia diagnoses over seven years by about a 20% relative reduction. Australia's identical natural experiment replicated it (JAMA, 2025), as have analyses in Canada and New Zealand. A separate US study found the newer recombinant vaccine (Shingrix) outperformed the old live one — 17% more dementia-free time over six years. Mechanistically it's plausible twice over: the varicella-zoster virus damages blood vessels and reactivates along nerves for decades, and the vaccine's adjuvant may broadly train innate immunity. These are quasi-experiments, not RCTs — but quasi-experiments of unusual quality, and the vaccine is already indicated for everyone 50+ anyway. Two doses, done. The annual flu shot rides the same logic with weaker evidence: 65+ adults with consistent flu vaccination showed ~40% lower Alzheimer's incidence over four years in a propensity-matched study of nearly a million pairs.
Head injury is an established Lancet factor; risk persists decades after even one significant TBI. Practically this is falls-proofing (lighting, rugs, grab bars, footwear, vision correction, vitamin D if low), helmets for bikes and ladders and ice, and — elegantly — tai chi, which cuts falls by 20–43% in meta-analyses, making it simultaneously a balance program and TBI prophylaxis. Air pollution (PM2.5) is on the Lancet list; the practical, evidence-bearing response arrived in 2026 — a crossover trial of real vs sham HEPA filtration near highways improved an executive-function test within a month. A bedroom HEPA filter is cheap; run it. Prefer an N95 when the sky is orange.
Aluminum. People either roll their eyes or throw out their cookware; the data support neither. Occupational aluminum-dust studies: null. Antiperspirants: dermal absorption ~0.01% — no credible link. Cookware: no demonstrated risk. Vaccine adjuvants: no dementia signal. The one stream with a real, unresolved signal is drinking water — several (not all) studies associate levels above ~0.1 mg/L with roughly doubled risk, confounded by co-occurring minerals. The defensible action is a water filter if your municipal report shows elevated aluminum. Keep the pans; check the water report. Anesthesia, briefly: post-operative confusion is real and worth preventing (see the hospital kit below), but good trials show no long-term dementia difference between spinal and general anesthesia — don't defer needed surgery, especially cataracts, out of dementia fear.
FINGER and U.S. POINTER both improved cognition with the same recipe this page uses — exercise, diet, cognitive training, social activity, vascular monitoring. POINTER's underappreciated finding: both of its arms improved; the structured arm — facilitated peer meetings, prescribed targets, someone checking — beat self-guided advice by a small but real margin. Structure itself was an active ingredient. Translate that: this protocol works to the degree it is scheduled, written down, logged, and reviewed with another person. A calendar, a simple log, a quarterly review (with me, or at your kitchen table), and a partner in it. In the personality research, conscientiousness predicts an ~89% difference in Alzheimer's risk between its highest and lowest deciles — without changing the plaques at autopsy. Resilience, again. Structure is not the packaging of the intervention. It is the intervention.
Stress, breath & the caregiver's nervous system
He or she who keeps another person's brain alive all day needs a way to down-regulate their own. Why stress is on a dementia page at all: higher midlife cortisol tracks with worse cognition and measurably smaller brain volume — an effect concentrated, in the Framingham data, in middle-aged women, the demographic doing most spousal caregiving. "Vital exhaustion" in midlife shows a dose-response with later dementia. No trial proves stress reduction prevents dementia — I won't claim it — but stress physiology runs straight through four proven factors: blood pressure, sleep, depression, and whether you keep doing everything else on this page.
Slow breathing — the practice I'd actually prescribe. Breathing slowly at about 5–6 breaths per minute (inhale ~4 seconds, exhale ~6) sits at the resonance point of the baroreflex; practiced 15–20 minutes daily it lowers systolic blood pressure by roughly 4–6 mmHg in meta-analyses — a real antihypertensive effect from a free intervention, which is how it earns its slot regardless of anything else. The "anything else" is genuinely interesting: a 2023 USC randomized trial had adults practice paced breathing at this frequency 20 minutes twice daily for a month; their plasma amyloid-beta fell while the comparison group's rose. Nobody knows yet whether moving plasma amyloid this way means anything clinically; I file it under mechanistically fascinating, unproven. For acute moments — the 2 a.m. wandering episode, the question asked the ninth time — the fastest validated tool is the physiological sigh: two stacked inhales through the nose, one long full exhale through the mouth, one to three repetitions. Five minutes daily of this "cyclic sighing" beat mindfulness meditation for daily mood in a 2023 Stanford randomized study. Prescription: 15–20 minutes once or twice daily at 5–6 breaths/minute, plus the physiological sigh as needed. Light-headedness means you're forcing it; ease off. Skip prolonged breath-holds in unstable heart disease or pregnancy.
Meditation, honestly. The two best trials of meditation for older-adult cognition (MEDEX; the 18-month French Age-Well trial) were null on cognition and brain structure. Meditation's real, replicated wins are anxiety (non-inferior to escitalopram in a 2023 trial), depression, and sleep — which is exactly the pathway a caregiver needs. Kirtan Kriya, a 12-minute chanting-plus-fingertip meditation, has small trials specifically in dementia caregivers showing improved depression: twelve minutes, free, caregiver-tested — a fine default. For the person who "can't meditate": a 20-minute guided body-scan audio (yoga nidra) or progressive muscle relaxation — plausible, small studies, zero cost, and the use case (the exhausted caregiver at 3 p.m.) is exactly right. Tai chi carries meta-analytic cognitive benefit in MCI; the standout 2023 trial delivered cognitively enhanced tai chi by videoconference and improved cognition by ~1.5 points beyond standard tai chi, sustained a year — and it cuts falls dramatically. Patient and caregiver can do it together in the living room, which I consider a feature, not a footnote. Two hours a week outdoors in green space stacks with the walk you were taking anyway.
The supplement shelf, ranked
Ground rules first. Food pattern, movement, sleep, and the medical housekeeping above outrank everything in this section. Quality is a real problem in this industry: choose third-party-tested products (USP, NSF, or ConsumerLab-verified) and match the form and dose that was studied — a proprietary extract's results do not transfer to a generic powder. And bring the full list to your physician visit, because the interactions flagged below are the practical difference between a sensible stack and a reckless one.
Where to get these: most of what follows is collected, with dosing attached, in my Alzheimer's Prevention protocol in the dispensary (free account, practitioner-grade brands), and each item also carries an Amazon link. Whatever you choose, buy it anywhere you like — just check for third-party testing and match the dose to what was actually studied. Disclosure: as an Amazon Associate, Starsiak Osteopathic Clinic earns from qualifying purchases through the Amazon links on this page, and Dr. Starsiak earns from purchases through the Fullscript dispensary — at no extra cost to you.
Here is something you probably haven't heard: lithium — the mood-stabilizer drug — appears to be, at trace doses, something closer to an essential micronutrient for the aging brain, and the evidence took a dramatic turn in 2025. The epidemiology has hinted at this for years: in the landmark Danish study (JAMA Psychiatry, 2017 — 73,731 dementia cases), people whose tap water carried more lithium had 17% lower dementia incidence than those at low exposure (honesty requires the middle of that curve: the relationship is J-shaped, not linear, and a US analysis found no association once demographics were modeled — ecological data propose; they don't dispose). At the pharmacologic end, bipolar patients who stay on lithium long-term show dementia rates at or below the general population's.
The 2025 Nature paper (from Bruce Yankner's Harvard lab — the group that discovered amyloid-beta's toxicity) is what moved this from curiosity toward centerpiece. Measuring 27 metals in postmortem brains, they found lithium was the only one depleted in the memory cortex — already at the mild-cognitive-impairment stage. The reason is remarkable: amyloid plaques bind and sequester lithium, trapping the brain's own trace supply. Mice fed lithium-depleted diets developed accelerated amyloid, tau, inflammation, and memory failure. And the treatment arm is why "5 mg" entered the conversation: lithium orotate — an organic salt that largely evades capture by plaques (carbonate, the pharmaceutical salt, binds them avidly) — given at trace doses restored brain lithium, prevented pathology when started early, and when started late almost completely reversed memory loss in the mice. Carbonate at the same trace dose did nothing.
The human ledger, with no varnish. Small trials of low-dose lithium carbonate in MCI are genuinely encouraging (less CSF tau phosphorylation, attenuated decline); a Brazilian microdose trial reported stabilized cognition in Alzheimer's while controls declined. Against this: the 2026 LATTICE pilot missed its primary endpoints, and a 2026 meta-analysis of the six conventional-salt trials found no significant overall benefit — with its authors explicitly pointing at orotate as the formulation the field should now test. It is being tested: the first lithium-orotate trial in early Alzheimer's began enrolling at Johns Hopkins in late 2026. No orotate result exists yet in humans. Anyone who tells you 5 mg of lithium orotate is proven to prevent Alzheimer's is ahead of the data; anyone who tells you the idea is silly hasn't read the 2025 paper.
My clinical read. The mechanism is unusually strong, the epidemiology is consistent in direction, the microdose safety profile is excellent, and the cost is a few dollars a month. That combination — high plausibility, low risk, low cost, unproven efficacy — is exactly the profile where a monitored trial-of-one is defensible for a motivated adult. Dose: 1–5 mg elemental lithium daily as lithium orotate (dispensary · Amazon) — check that the Supplement Facts line reads "Lithium (as lithium orotate) 5 mg," since labeling isn't uniform. For perspective, 5 mg elemental is roughly 1/25th to 1/45th of a psychiatric dose. Monitoring: I check TSH and creatinine at baseline and 6–12 months. Who shouldn't: pregnancy or trying to conceive; significant kidney disease (stage 3+); unstable thyroid disease. The hard line: never escalate an over-the-counter product toward "real" lithium doses without prescription-level monitoring — the difference between 5 mg and 150 mg elemental is the difference between a trace nutrient and a drug with a narrow therapeutic window. Tier: Promising (mechanism and epidemiology) / Unproven (human efficacy at this dose). Re-evaluate when the orotate trials read out, likely 2027–28.
The core layer — five unglamorous things with real data
The botanicals — graded plant by plant, introduced one at a time, 8–12 weeks each
A note on names first, because it answers a question I'm often asked: in most of India Brahmi means Bacopa monnieri, but in some regional traditions the same word means Centella asiatica (gotu kola). They are unrelated plants that share a reputation as medhya rasayana — intellect-nourishing tonics. Both are covered here as separate entries. If an Ayurvedic practitioner recommended "Brahmi," find out which plant they meant.
Two traditional entries the dispensary deliberately doesn't cover
Ginkgo biloba — two of the largest, longest prevention trials ever run on any supplement (GEM: 3,069 people, six years; GuidAge: five years) were both null for preventing dementia — as close to a settled negative as this field produces, with bleeding concerns as a tiebreaker it didn't need. Vitamin E for prevention: failed (PREADVISE), with bleeding risk at high dose. Resveratrol: the good trial found no cognitive benefit and more brain-volume loss. Huperzine A: not a gentle herb — it's an unregulated, potent acetylcholinesterase-inhibiting drug with microgram dosing and label-accuracy problems; never in a prevention stack, and never alongside donepezil or bacopa. High-dose green-tea-extract (EGCG) pills: documented liver injury; drink the tea. Alpha-GPC: stroke signal in long-term cohort use; I prefer citicoline or food choline. Glucosamine/chondroitin: dropped, as detailed in the medicine-cabinet section — no purchase link on this site. NAD boosters, taurine, DHEA: nothing cognitive in humans worth your money yet.
Three clusters to respect. Cholinergic cluster (bacopa, sage, huperzine, prescription donepezil/rivastigmine/galantamine): pick at most one botanical, never stack with the prescriptions. Serotonergic cluster (kanna, saffron, SSRIs/SNRIs, tramadol): additive load — flag any combination to your physician. Bleeding cluster (fish oil, curcumin, ginkgo, high-dose vitamin E, aspirin, warfarin/DOACs): the sum matters even when each item is individually "safe" — count them, and stop the botanicals two weeks before surgery. Thyroid watchers: bacopa and ashwagandha push up; shankhpushpi pushes down; all three matter on levothyroxine. And the liver: ashwagandha, green-tea extract, and unpurified anything — one botanical with hepatic potential at a time, and symptoms (itching, dark urine, jaundice) mean stop and test, not push through.
Little experiments worth their cost
Overnight scent training (olfactory enrichment). Smell is the sense with a direct anatomical line into the memory cortex, its loss is one of the earliest Alzheimer's signs, and a 2023 UC-Irvine randomized trial did something delightfully simple: a bedside diffuser rotating seven essential oils, one per night, two hours nightly for six months, in cognitively normal older adults. The enriched group improved dramatically on a standard memory test with measurable white-matter changes on imaging. Forty-three people, single site, awaiting replication — the effect will likely shrink. But the cost is a diffuser with a timer and a rotating oil set, the risk is essentially nil (mind asthma, and some oils are toxic to cats), and the underlying science — use the pathway or lose it — is sound. I'd also flip it around: unexplained loss of smell is worth reporting, not dismissing.
Sauna — the Finnish prescription. Middle-aged Finnish men using a traditional sauna 4–7 times weekly had 66% lower dementia incidence over two decades than once-weekly users, with a second cohort replicating direction at more moderate magnitude — and adding two cautions: no extra benefit beyond ~13 sessions/month, and bathing above 100°C associated with higher risk. No randomized trial exists; the mechanism is credible (a sauna session is cardio-mimetic). Prescription mirroring the data: 15–20 minutes at traditional temperature (80–90°C), building toward 4 sessions/week, hydrate with electrolytes, never with alcohol, and clear it with me first if you have unstable cardiac disease. No sauna? Japanese cohorts tie near-daily tub bathing to better cardiovascular outcomes — a 40–41°C evening soak is a plausible partial substitute, not a proven one. Infrared cabins are gentler and not what the Finnish data measured.
40 Hz light-and-sound stimulation (gamma entrainment). MIT's flicker work spawned a home device whose small blinded trial in mild-moderate Alzheimer's reported strikingly less brain atrophy; the make-or-break 670-person pivotal trial completed enrollment in 2025 with results pending. Experimental — I'm watching, not prescribing; photosensitive epilepsy excludes. Acoustic slow-wave enhancement (phase-locked pink noise during deep sleep): proof-of-concept tier — for the gadget-inclined, harmless. Oil pulling — already placed honestly in the mouth section: dental evidence real, brain evidence nonexistent, cost trivial.
Testing & monitoring
What gets measured gets managed. The once-a-year panel I'd run on anyone serious about this protocol — most of it is standard bloodwork, and it's exactly what a prevention visit is for: home BP average (7-day protocol) — the number that outranks all others; metabolic — A1c, fasting glucose and insulin, lipid panel plus ApoB, Lp(a) once in your life; hs-CRP — not to treat directly, but to hunt its source; homocysteine with B12 and folate — this result decides the B-vitamin question; omega-3 index (target >8%) — decides the fish-oil question; 25-OH vitamin D, TSH, CMP/CBC. Non-lab: audiogram; dilated eye exam; periodontal exam; STOP-BANG; grip strength and gait speed; a baseline cognitive screen; medication ACB score; a stress screen — for the caregiver too. Request an appointment and we'll run it together.
On APOE testing. Knowing you carry APOE4 changes nothing about whether these interventions are worth doing — and changes a lot about how aggressively I'd do them. The REVEAL trial showed disclosure with counseling doesn't cause lasting distress. Two practical warnings: U.S. genetic-nondiscrimination law (GINA) does not cover life, disability, or long-term-care insurance — secure any long-term-care policy before testing — and direct-to-consumer kits will hand you this result with no counseling attached. Offered, never pushed — this is a decision best made in a real conversation, and I'm glad to have it.
On the new blood tests. Plasma p-tau217 testing now rivals spinal fluid for detecting Alzheimer's pathology and earned its first FDA clearance in 2025 — for people with symptoms. In cognitively normal adults I don't screen with it yet: a positive result in a low-risk person is more likely to mislead than help, and there is no approved preventive drug it would unlock. If genuine symptoms arrive, it moves early in the workup — and sorting out which situation you're in is a visit, not a mail-order kit.
If you or your person is admitted: bring glasses, hearing aids, and dentures the same day; post the med list with its ACB scores; family at the bedside; ask for the hospital's delirium-prevention (HELP) protocol; and challenge every newly offered sleeping pill, benzodiazepine, or Benadryl.
Putting it together
Days 1–30: the big three. Home BP cuff and 7-day average → appointment if elevated. Daily walk with one brisk segment. Hearing test booked. (And the shingles vaccine, if due — it's one errand.) Days 31–60: the rhythm. Add strength training twice weekly; set the sleep window and morning light; run the medication audit; start the food pattern's easiest wins — olive oil as default fat, greens daily, fish twice weekly, ultra-processed food down. Days 61–90: the layers. Labs → decide the supplement layer (multivitamin and vitamin D correction are the easy yeses; lithium after reading its section; botanicals chosen against your medication list, introduced one at a time, 8–12 weeks each, so we can tell what's doing what). Choose the year's new skill. Add the breath practice and, if accessible, the sauna habit.
Quarterly: review the log — BP average, steps, sleep, strength numbers, social calendar, supplement tolerability — and remove what isn't earning its place. A protocol that only ever adds is a pill organizer, not a plan. Annually: the full panel, the brown-bag medication review, hearing and vision, and an honest conversation about what stuck.
Two closing calibrations. First: nothing here is a guarantee, and someone who does everything right can still develop Alzheimer's — biology is not a courtroom, and blame has no place in this protocol. What the evidence supports is a meaningfully shifted probability curve and a later, gentler course. Second: the person most likely to abandon this protocol is the caregiver who is keeping someone else alive with it. If that's you — your walk, your sleep, your breath practice, your friendships are not what you do after the real work. They are the real work, applied to the second patient in the house.
The supplement doses, targets, and monitoring above assume review against your personal medication list, kidney and liver function, and history. The most efficient way to do that is a prevention visit: we run the panel, set your blood pressure target, score your medication list, and pick the layers that fit you — then review quarterly what's earning its place.
Everything above with a purchase link — the lithium orotate, algae omega-3, vegan D3, creatine, the iron-free multi substitute, the B-complex and magnesium threonate, and the botanicals — is collected in my dispensary under one protocol, with the dosing instructions attached to each item. Or buy any of it anywhere — check third-party testing and match the studied dose. The advice is the same wherever you buy.
This guide is general education, not medical advice, and it does not create a physician-patient relationship. Nothing on this page is a claim that any intervention prevents, reverses, or cures Alzheimer's disease; the evidence tier is stated for each item, and several are honestly labeled unproven. Do not stop or change a prescribed medication based on this page — including blood pressure medicines, benzodiazepines, and sleep aids, some of which require a supervised taper. Supplements are not FDA-approved to prevent or treat dementia; the botanicals above carry real interactions (bleeding, serotonergic, thyroid, and liver clusters are named in the text) and several are inappropriate in pregnancy, kidney disease, or liver disease — review anything new with your physician or pharmacist first, and never escalate low-dose lithium beyond the trace doses described. These statements have not been evaluated by the FDA and are not intended to diagnose, treat, cure, or prevent any disease. Disclosure: As an Amazon Associate, Starsiak Osteopathic Clinic earns from qualifying purchases through the Amazon links on this page, and Dr. Starsiak earns from purchases through the Fullscript dispensary — at no extra cost to you. The price you pay is unchanged, and the advice is the same wherever you buy. Evidence current through August 2026.