81% of Dementia Prevention RCTs Failed. What Actually Worked

The most honest number in dementia prevention is 81%

If you want the cleanest summary of the dementia-prevention literature, start here: a 2022 systematic review counted 112 completed randomized controlled trials and found that only 21 met their prespecified primary outcome. That leaves an 81% failure rate. In a field crowded with supplements, brain-training pitches, and lifestyle hacks with excellent branding, that number does real housekeeping.

The review by Nicola Coley et al. did something refreshingly unglamorous. It counted the whole record, not just the winners. The authors grouped the trials into pharmacological agents (33 trials, 29%), nutritional supplements (27, 24%), physical activity (18, 16%), cognitive stimulation (13, 12%), and multidomain interventions (22, 20%). The point is not that nothing helps. The point is that most things tested at scale did not work as designed on the endpoint that mattered most in each trial.

That matters because dementia prevention is full of slippage between observational promise and randomized proof. A risk factor can correlate with later dementia and still resist intervention. A biomarker can improve while cognition does nothing useful. A secondary endpoint can look nice while the primary one stays stubbornly null. Randomized trials do not solve every problem, but they are where a lot of attractive theories go to get a reality check.

Two large successes, and neither fits in a pill bottle ad

Across those 112 trials, the review identified two large-scale successes: blood pressure lowering in the Syst-Eur trial, where incident dementia was the primary endpoint, and the FINGER trial, a multidomain lifestyle intervention whose primary endpoint was change in a neuropsychological test battery composite score in the 2022 review.

Neither result is especially friendly to the wellness economy. Blood pressure control requires diagnosis, treatment, follow-up, and adherence. The strongest synthesis of that literature is a 2021 Cochrane review of pharmacological treatment of hypertension in people without prior cerebrovascular disease. Cochrane reviews use preregistered methods, duplicate screening, formal risk-of-bias assessment, and GRADE ratings. In this area, that is about as close as you get to the adult table.

The FINGER intervention was not a cute single tweak either. It combined supervised aerobic and strength exercise, computerized group cognitive training, nutritional counseling, and vascular risk monitoring, including blood pressure, cholesterol, and metabolic parameters. According to the evidence summarized in the review literature, FINGER produced a small but statistically significant cognitive benefit in the trial record summarized by Coley et al. The published sources available here do not report the exact primary effect size, confidence interval, or p-value, so the right description is limited but still meaningful: small, significant, and multidomain.

Why the 81% figure holds up

The 81% failure rate is persuasive for a boring reason: the review used completed randomized trials and judged them by prespecified primary outcomes. That reduces the usual forms of optimism inflation. You cannot rescue a failed trial by spotlighting a subgroup, a secondary measure, or a post hoc explanation after the primary endpoint misses.

It also gives readers a denominator. Plenty of prevention stories quietly imply that the field is a string of near-breakthroughs. This review shows a different pattern. Most interventions did not cross the line they set for themselves.

That still does not make 81% a ceiling on failure. Publication bias cuts the other way. If unpublished negative trials exist and were not captured, the true failure rate could be even higher. The review cannot quantify that possibility, but it should make anyone cautious about confident sales copy.

Multidomain looks promising. It does not yet look settled.

FINGER is the best-known positive multidomain trial, but it is not the category average. In the same 112-trial review, other prominent multidomain studies did not meet their primary endpoints: MAPT failed on its 3-year memory-function endpoint, preDIVA failed on incident dementia, and the German AgeWell trial failed on global cognitive performance according to Coley et al.

That is the awkward part often lost in summaries. "Multidomain" is not a magic word. It is a label covering different protocols, populations, durations, and levels of implementation. FINGER may have worked because it hit the right combination at the right intensity in the right risk group. Or because it was delivered with better fidelity. Or because other trials included too much biological heterogeneity for an average effect to show up cleanly.

A 2024 meta-analysis of long-term physical exercise and multidomain interventions supports a narrower conclusion. Single-domain physical activity trials yielded mainly non-significant primary outcomes, while multidomain approaches produced more consistent benefit signals. That is useful, but it is not the same as saying the case is closed. No factorial-design trial has cleanly separated which components matter most, what dose they require, or which combinations actually drive the signal.

A 2025 systematic review by Erika Bereczki et al. makes the same tension explicit. The field has moved toward combined multidomain, pharmacological, and nutraceutical strategies precisely because so many single-domain interventions failed. That review also references the Lancet Commission estimate that addressing 14 modifiable risk factors could prevent or delay up to 45% of dementia cases. Important distinction: that estimate comes from observational epidemiology. It describes prevention opportunity, not randomized proof that one intervention package has already delivered it.

The supplement record is bad, and the trial count makes that hard to ignore

Nutritional supplements accounted for 27 of the 112 completed trials, making them the second-largest category in the review by Coley et al. If isolated nutrient supplementation were a reliable route to cognitive protection, this is where you would expect the scoreboard to show it. It does not.

The broader review literature does not support supplement efficacy as a class for preventing cognitive decline or dementia. Bereczki et al. describe the modern field as shaped by the failure of numerous pharmacological and non-pharmacological single-domain interventions in their 2025 review. That includes many of the supplement-style approaches that dominate consumer marketing.

There is also a known interpretation problem in nutrition trials: funding and reporting bias can add spin even when the result is thin. The available abstracts here do not consistently report the funding status of the supplement trials in the 112-trial review, so it is not possible to quantify that bias directly. The fair conclusion is simpler anyway. Even before you get into motive, the category has not delivered dependable randomized success.

Why results may differ so much from one trial to the next

One likely reason is that "older adults at risk" is not a biologically uniform group. A 2025 randomized trial analysis by Thais Lorenzo et al. found that prediabetes, diabetes, and glucose-related markers substantially modified cognitive and neuroimaging outcomes in a 2-year multidomain lifestyle trial. In plain terms, metabolic status changed how participants responded.

That is not a side note. It may explain why one multidomain protocol works in one cohort and fizzles in another. If glucose dysregulation changes response, then a pooled average across metabolically mixed participants can wash out a real effect in one subgroup and hide a null effect in another. Precision prevention is still more slogan than finished science, but this is the kind of result that gives the idea some actual structure.

Vascular and motor signals may intersect here too. A 2025 secondary analysis of SPRINT MIND found that slow baseline gait speed predicted later cognitive impairment in older adults with hypertension. That does not test whether gait training prevents dementia, and it does not prove the mechanism of blood pressure treatment. It does support the broader view that vascular, metabolic, motor, and cognitive decline often travel together instead of politely staying in separate boxes.

What readers should do with this evidence

The practical verdict is less flashy than the market wants. Blood pressure management has the strongest replicated support in the randomized literature, backed by the best trial infrastructure in the Cochrane synthesis. Multidomain intervention is the most plausible broader framework, but FINGER remains more important than fully replicated. Exercise, diet, and cognitive training may help more when combined than when isolated, yet the exact recipe still looks unsettled.

So the right reading of this literature is neither nihilism nor hype. It is triage. First, control vascular risk factors that already have strong clinical justification. Second, treat multidomain lifestyle work as serious long-haul prevention, not as a 30-day challenge with better typography. Third, be skeptical when a single supplement claims to solve a problem that has chewed through 112 randomized trials and spat out an 81% failure rate.

The field's least marketable finding is also its most credible one: most dementia-prevention interventions have failed in randomized trials, and the best-supported successes are blood pressure treatment and one intensive multidomain program.

That is not a glamorous answer. It is, so far, the honest one.