Supplements and dementia

The evidence

The COSMOS trial programme randomised more than 21,000 US adults aged 60 and over. Three separate cognitive sub-studies — COSMOS-Mind (n=2,262, telephone testing), COSMOS-Web (n=3,562, online testing) and COSMOS-Clinic (n=573, in-person neuropsychological testing) — each found a benefit of a standard daily multivitamin on memory or global cognition ^20,21,22. A meta-analysis of all three found a benefit on global cognition equivalent to reducing cognitive ageing by about two years ²².

In practice

The honest limits: the effect is small, the population was predominantly white, well-educated and American, and a prespecified analysis found no reduction in the incidence of mild cognitive impairment or dementia over three years — though that analysis was underpowered with only 110 MCI and 14 dementia cases ²³. So: a modest cognitive benefit is reasonably established; dementia prevention is not. A standard multivitamin is inexpensive and safe, and this is the closest thing to a defensible supplement recommendation in the field.

The evidence

The Oxford VITACOG trial gave high-dose folic acid, B6 and B12 to 271 people over 70 with mild cognitive impairment for two years. Brain atrophy was 30% slower overall, and 53% slower in those with baseline homocysteine above 13 µmol/L ⁹. Cognitive benefits were concentrated in the same high-homocysteine group ¹⁰. Follow-up analyses found the effect only occurred in people who also had good omega-3 status ^11,12. A subsequent Hong Kong trial of 279 people with MCI and raised homocysteine, using lower doses, found no benefit at two years ¹³. A 2025 meta-analysis of 17 trials in 5,275 people found a very small benefit on global cognition once outliers and single-blind studies were removed ¹⁴.

In practice

If your relative is B12 or folate deficient, that should be corrected — it is a reversible cause of cognitive impairment in its own right, and it is standard practice to test for it. Taking high-dose B vitamins without a measured deficiency or raised homocysteine is not supported.

The evidence

A meta-analysis pooling 48 longitudinal studies in 103,651 people found around a 20% lower risk of dementia or cognitive decline with higher dietary omega-3 intake ¹⁷. Randomised trials tell a different story. A Cochrane review found no benefit on cognitive function in cognitively healthy older people ¹⁵. The MAPT trial (n=1,525, three years) found no effect ⁷. A 2025 meta-analysis of 11 placebo-controlled trials in cognitively unimpaired older adults found a standardised mean difference of −0.02 (95% CI −0.07 to 0.04) — as close to zero as results come ¹⁶.

In practice

Subgroup analyses hint that people starting with low omega-3 levels, or with mild cognitive impairment specifically, might benefit — but these are post-hoc findings from trials that were negative overall. A 2026 observational analysis even reported faster decline among supplement users, which is almost certainly confounding by indication rather than harm, but it does not help the case ⁷². Oily fish twice a week remains sensible dietary advice.

The evidence

Observational meta-analyses consistently find vitamin D deficiency associated with around 40–50% higher dementia risk ¹⁹. The Finnish Vitamin D Trial randomised 2,492 generally vitamin D–sufficient adults to placebo, 1,600 IU or 3,200 IU daily for up to five years and found no effect on dementia incidence ¹⁸. The Canadian SYNERGIC trial found no cognitive benefit from vitamin D added to exercise and cognitive training ⁴².

In practice

Test and treat genuine deficiency — it matters for bone health, falls and fractures, which are far more immediate concerns in residential aged care. Do not take high-dose vitamin D expecting a cognitive return.

The evidence

The LipiDiDiet trial randomised 311 people with prodromal Alzheimer’s disease to the multinutrient drink or an identical-tasting control. At 24 months the primary cognitive endpoint was not met (p=0.166) — largely because the control group declined far less than the trial was designed for ²⁴. Over 36 months, significant differences emerged on the cognitive composite (60% less decline), Clinical Dementia Rating Sum of Boxes (45% less decline), memory and brain atrophy, with small-to-medium effect sizes comparable to established Alzheimer’s medicines. Only 81 participants contributed 36-month efficacy data ²⁵.

In practice

The trial is well conducted, industry-funded, and its positive findings are secondary endpoints in a trial that missed its primary. It is safe and well tolerated. It is also expensive and not subsidised in Australia. If a family wants to try it in early disease, that is a defensible choice; it should not be presented as proven, and it has no role in advanced dementia.

The evidence

The US Ginkgo Evaluation of Memory trial randomised 3,069 adults aged 75 and over to 120 mg twice daily and followed them for a median of 6.1 years. There was no reduction in dementia (hazard ratio 1.12) and no slowing of cognitive decline in any domain ^26,28. The French GuidAge trial randomised 2,854 adults over 70 with memory complaints for five years and found no reduction in progression to Alzheimer’s disease ²⁷. Manufacturer-sponsored meta-analyses of shorter treatment trials in established mild dementia report benefit; the independent prevention trials do not ⁷⁰.

In practice

Ginkgo also increases bleeding risk, which matters for anyone on aspirin, clopidogrel or an anticoagulant — a very large proportion of aged care residents ⁸⁸.

The evidence

The PREADViSE trial randomised 7,540 men to vitamin E, selenium, both or placebo and found no reduction in dementia incidence for any arm ²⁹. A Cochrane review found no evidence that vitamin E prevents progression from mild cognitive impairment to dementia or improves cognition, but moderate-quality evidence from a single trial that it may slow functional decline in Alzheimer’s disease ³⁰.

In practice

The Cochrane review found no increased risk of serious adverse events or mortality in the dementia trials it examined ³⁰, so the case against it is absence of benefit rather than demonstrated harm. Either way, the risk–benefit calculation does not favour routine high-dose use.

The evidence

Meta-analyses in healthy adults find modest benefits on memory, with a larger effect in older adults (66–76 years) ^31,33. The first study in Alzheimer’s disease, published in 2025, gave 20 g/day for eight weeks to 20 patients: brain total creatine rose 11%, and cognitive scores improved. But it was single-arm, open-label, uncontrolled and eight weeks long ³² — a design that cannot distinguish a drug effect from practice effect and expectation.

In practice

Creatine is cheap, extensively studied for safety in exercise contexts, and has separate evidence for muscle strength and mass in older adults — which matters for falls and frailty ⁷⁵. If someone is taking it for muscle, the possible cognitive upside is a bonus. Taking it specifically to treat or prevent dementia is not supported by the evidence at this stage.

The evidence

The science here is genuinely interesting, and the reporting of it has run some way ahead of the evidence. A small Brazilian trial randomised 45 people with amnestic mild cognitive impairment to low-dose lithium or placebo for a year and found reduced phosphorylated tau in spinal fluid and slightly better cognitive scores ⁹³. When the same group traced that cohort 13 years later, those originally given lithium had markedly better cognitive scores — but only 36 of the 61 could be found, a third had died, and comparing survivors of two groups a decade on is not a fair comparison ⁹⁵.

In practice

The drinking-water studies, often quoted as though they settle it, do not agree with each other. A Danish study of 73,731 people with dementia found lower dementia above 15 µg/L — but higher dementia at 5–10 µg/L, which is not how a protective effect behaves ⁹⁶. A Scottish study of 37,597 people found no protection at all, and a trend towards more dementia in women ⁹⁷. Then in 2025 a Nature paper showed lithium is depleted in the Alzheimer’s brain and that replacing it prevented damage in mice ³⁴ — and in 2026 the trial designed to test that idea in humans, two years of low-dose lithium in 80 people with mild cognitive impairment, missed all six of its main outcomes ³⁵. Lithium orotate is sold as a supplement and is not regulated as a medicine. Lithium has a narrow safety margin, affects thyroid and kidney function, and interacts with fluid tablets, blood pressure medicines and anti-inflammatories — all common in older people. This is worth following as the research develops, but it is not something to start on your own. If you are considering it, discuss it with the treating GP first.

The evidence

A 22-week randomised double-blind trial in 54 people with mild-to-moderate Alzheimer’s disease found saffron 30 mg/day performed comparably to donepezil 10 mg/day, with less vomiting ⁹⁸. Further trials have reported saffron matching donepezil and memantine, and beating placebo, over 16 to 52 weeks ¹⁰³.

In practice

The catch is replication. These trials are small, and the Alzheimer’s saffron literature comes overwhelmingly from a single research group in Iran. A finding that has not been reproduced by independent investigators elsewhere is a promising lead, not an established treatment. Saffron is also expensive and variable in strength between products.

The evidence

A Cochrane review pooled six trials in 454 patients and found apparent benefits on cognition, global assessment, behaviour and daily function. Its conclusion, though, was that the methodological quality of most trials was low, only one was of adequate quality and size, and there is inadequate evidence to make any recommendation ⁹⁹.

In practice

Huperzine A is a cholinesterase inhibitor — the same mechanism as donepezil — sold as a supplement rather than a medicine. That means the same class of side effects (nausea, slow pulse, vivid dreams) without the dose control, quality assurance or prescriber oversight that comes with the registered drug. If the mechanism is what you want, use the medicine.

The evidence

A Cochrane review of 14 trials found some benefit on memory and behaviour, and a stronger signal on global clinical impression. But most trials ran only 20 to 30 days, they were heterogeneous in dose and population, and they largely enrolled people with vascular cognitive impairment rather than Alzheimer’s disease ¹⁰⁰.

In practice

Well tolerated. The honest summary is that a modest short-term effect in vascular cognitive impairment has never been followed up with the long trials that would show whether anything durable happens.

The evidence

A 2026 critical appraisal of clinical trials across nine prominent dietary bioactives found that none demonstrated consistent cognitive efficacy in Alzheimer’s disease. Phase III trials of DHA, curcumin and tricaprilin missed their primary cognitive endpoints. Resveratrol reduced a cerebrospinal fluid amyloid marker without cognitive benefit ³⁷. An eight-week randomised crossover trial of nicotinamide riboside 1 g/day in 46 older adults with subjective cognitive decline or MCI lowered plasma p-tau217 but produced no cognitive change ³⁶.

In practice

Poor oral bioavailability is a recurring problem for several of these compounds. Biomarker movement without cognitive benefit is a pattern seen across the whole field, including with far more powerful drugs.

The evidence

The ketone-body hypothesis is plausible, and tricaprilin — the most rigorously developed medium-chain triglyceride product — failed to meet its primary cognitive endpoint in phase III trials ³⁷. Coconut oil specifically has no adequate randomised evidence in dementia.

In practice

Coconut oil is high in saturated fat, which cuts against every cardiovascular recommendation relevant to vascular dementia risk.

The evidence

Within the COSMOS programme, cocoa extract providing 500 mg flavanols daily produced no benefit on global cognition in COSMOS-Mind (n=2,262) or COSMOS-Clinic (n=573), across three years and two years respectively ^20,22.

In practice

A useful internal control: the same trial design that detected a multivitamin effect detected nothing for cocoa. That makes the multivitamin result more credible and the cocoa result harder to explain away.