Source Preprint paper:
Twenty-Four Years On, the Cheapest Pill in the Pharmacy Shows a Hint of Brain Protection
Researchers followed participants from the Diabetes Prevention Program, a large US randomized trial that began in 1996, for more than two decades and asked whether the people originally randomized to metformin ended up with less dementia. In 2022 to 2024, when the surviving participants had a median age of 74, a masked expert panel classified 1,483 of them into cognitive categories. The overall spread of cognitive diagnoses was essentially identical across the three original trial arms. Within that null overall result, one specific category moved: dementia was diagnosed in 1.9 percent of the metformin group versus 3.5 percent of the placebo group and 4.2 percent of the intensive lifestyle group, giving adjusted odds ratios of roughly 0.40 and 0.38. Metformin participants also scored marginally higher on a word-list memory test across five testing waves, a difference the authors themselves label as modest (Cohen’s d of 0.1). The result rests on only 48 dementia cases in total and 9 in the metformin arm, and is not statistically significant against placebo before covariate adjustment.
Metformin has been the most talked-about candidate for a general anti-aging drug for over a decade, largely on the strength of observational data that could never rule out the possibility that healthier people simply get prescribed it more often. The obvious fix is a randomized trial, but nobody has run one long enough to count dementia cases. This preprint is the closest anyone has come, and it comes by accident.
The Diabetes Prevention Program randomized 3,234 American adults at high risk of type 2 diabetes to metformin, placebo, or an intensive lifestyle program in the late 1990s. The masked phase ran only about three years, but participants in the metformin arm were then kept on open-label metformin until 2021. By the end, the metformin group had accumulated an average of 15.5 years of drug exposure against roughly 4 years in the comparison arms, who mostly picked up metformin later from their own doctors after developing diabetes. That exposure gap is the natural experiment.
In 2022 to 2024 the surviving cohort, by then a median 74 years old, sat through a full dementia workup: the standard national Alzheimer’s centre neuropsychological battery, informant questionnaires, a video neurological exam, blood-based Alzheimer’s markers, and adjudication by a five-person panel blinded to who had taken what.
The headline number is that dementia was about 60 percent less likely in the metformin arm. The number underneath it is that this represents 9 cases versus 18 and 21. Everything else about the cognitive picture was flat. Mild cognitive impairment, the stage that normally precedes dementia, showed no difference at all. Memory and executive function declined at identical rates in all three groups over 14 years of repeated testing; the metformin group simply sat a fraction of a point higher on the memory test throughout, which is a different phenomenon from slower aging.
The most provocative detail is buried in an exploratory analysis. Splitting dementia cases by blood p-tau217, the difference between groups came almost entirely from amyloid-positive cases, meaning Alzheimer’s-type dementia specifically, rather than from vascular dementia, which is what a glucose-lowering drug might have been expected to prevent. If real, that points at a mechanism beyond blood sugar control. If not real, it is what a handful of cases scattered across cells looks like. Both readings remain live.
Actionable Insights
The honest take-home is that this study lowers the probability that long-term metformin harms cognition and slightly raises the probability that it helps, without settling anything.
On magnitude. Over roughly 24 years, dementia occurred in 1.9 percent of the metformin group versus 3.5 percent on placebo. That is an absolute difference of 1.6 percentage points, meaning about 62 people would need to take metformin for two decades to prevent one case of dementia. Against the lifestyle arm the gap was 2.4 points, or about 43 people. The widely quoted “60 percent reduction” is the relative version of that same small absolute number, and relative numbers always sound larger.
On the memory benefit. The metformin group scored 0.58 points higher on a word-recall test scored out of 45. Cohen’s d was 0.1, which is a rule-of-thumb way of saying the two groups overlap almost completely. Roughly speaking, if you picked one person from each group at random, the metformin person would have the better score about 53 percent of the time instead of 50 percent. It is not a difference you would notice in yourself.
Practical framing: this is a signal worth tracking, not a reason to start a drug. Everyone here was overweight and pre-diabetic at entry, so nothing generalizes to metabolically healthy people. Anyone on long-term metformin should have B12 monitored, since depletion is common and itself causes cognitive symptoms.
Context and Source
- Full title: Randomized metformin and cognitive outcomes in the Diabetes Prevention Program Outcomes Study
- Institutions: A 27-centre US consortium. Coordinating and biostatistics centre at George Washington University. Lead authorship spans VA Puget Sound and University of Washington, Columbia University Irving Medical Center, Pennington Biomedical Research Center, Georgetown, Michigan, Joslin/Harvard, Northwestern, New Mexico, Pittsburgh, UC San Diego, Colorado Anschutz, and Massachusetts General Hospital.
- Country: United States
- Funded by: National Institute on Aging and NIDDK, National Institutes of Health. Merck KGaA donated study medication.
- Peer review status: Not peer reviewed. The preprint carries the standard medRxiv notice that it should not be used to guide clinical practice.
Impact evaluation. medRxiv is a preprint server, not a journal, and by design has no Journal Impact Factor
Related Reading:
