A notable result from a long follow-up

Metformin, a long-established medicine for type 2 diabetes, has been associated with lower odds of dementia in a new analysis from the US Diabetes Prevention Program Outcomes Study. The finding is attracting attention because it comes from a cohort originally assigned at random to metformin, placebo or an intensive lifestyle programme, rather than from healthcare records alone.

The analysis assessed 1,483 surviving participants who completed sufficient cognitive evaluation between 2022 and 2024. Their median age was 74. The participants had entered the original Diabetes Prevention Program between 1996 and 1999 because they had impaired glucose tolerance and were at high risk of type 2 diabetes; they were not a representative sample of all older adults.

Compared with those initially assigned to placebo, people originally assigned to metformin had 60 per cent lower adjusted odds of dementia. The comparison with the lifestyle-intervention group showed 62 per cent lower odds. These figures underpin the claim that metformin may roughly halve dementia risk.

Yet “odds” are not the same as a guaranteed reduction in an individual’s probability of developing dementia, and the estimates were imprecise. The confidence intervals were wide and only narrowly excluded no difference. The researchers themselves say that a longer follow-up with more dementia cases is needed.

Why the design matters

Much of the earlier evidence on metformin and cognition has been observational. Such studies can struggle to separate a drug’s effects from differences between the people who do and do not receive it. Diabetes severity, access to care, weight, cardiovascular health and the choice of other glucose-lowering medicines can all influence both prescribing and dementia risk.

The original Diabetes Prevention Program reduced that problem by randomly assigning 3,234 adults to one of three strategies: metformin, placebo or an intensive programme of dietary change, physical activity and weight loss. During the masked phase, which lasted about 3.2 years, participants in the metformin group received 850 milligrams twice daily.

After the main trial ended, the study became a long-term follow-up. Those initially allocated to metformin were offered open-label treatment until 2021. On average, their total metformin exposure was 15.5 years, compared with about 4.5 years in the original placebo group and 3.8 years in the lifestyle group. Participants in the comparison groups could receive metformin from their own clinicians after developing diabetes.

That history is both a strength and a complication. Randomisation provides a more credible basis for comparison than a conventional prescription database. But the result reflects an initial assignment followed by years of differing treatment access and use, not a simple test of a fixed metformin course versus permanent non-use.

What cognitive outcomes showed

The investigators used detailed clinical and neuropsychological assessments to classify participants as having no cognitive impairment, milder impairment syndromes, mild cognitive impairment or dementia. The adjudicators were masked to the participants’ original treatment assignment, an important protection against biased diagnosis.

The headline dementia result did not extend to the overall distribution of all cognitive impairment categories, which did not differ significantly across the three groups. Metformin assignment was also linked to better performance over time on a memory measure, but the effect was small.

This distinction matters. A convincing dementia-prevention treatment would ideally produce consistent evidence across diagnosed dementia, broader cognitive impairment, longitudinal test scores and, eventually, independent trials. The new analysis provides an encouraging signal, rather than that complete evidential package.

The study also assessed cognition at a late point in follow-up rather than measuring incident dementia continuously from trial enrolment. Participants who remained alive and able to complete the 2022–2024 assessment may differ from those who did not. This creates a potential for survivor or selection effects, even though the original allocation was random.

A mixed earlier literature

The result is compatible with several large observational analyses and reviews that have associated metformin use with a lower risk of dementia among people with diabetes. Possible explanations include improved glucose regulation and indirect benefits through metabolic and vascular health. Scientists have also proposed that effects on inflammation, cellular energy regulation and insulin signalling could be relevant to brain ageing.

However, plausible biological mechanisms do not demonstrate a clinical benefit. Earlier studies have been inconsistent, including studies finding no significant association when metformin initiation was compared with delayed or no initial glucose-lowering treatment. A previous cognitive analysis from the same Diabetes Prevention Program follow-up, conducted when participants were younger, found no difference in cognitive scores between the metformin, lifestyle and placebo groups.

A 2024 umbrella review of evidence from observational research found associations between several diabetes medicines, including metformin, and reduced dementia risk. It also found that results varied by region and study design. Such variation reinforces the need for trials that are designed specifically to test cognitive outcomes.

What should happen next

The new report is a preprint, meaning it has been posted for scientific discussion but has not yet completed journal peer review. Its main contribution is to provide unusually long follow-up after randomised allocation, with dementia classification undertaken in later life. Its central limitation is the limited number of dementia cases, which makes a large apparent effect vulnerable to revision as more outcomes accrue.

Replication is particularly important in groups beyond the trial population: people without prediabetes, individuals already living with diabetes for many years, and populations with different ages, ethnic backgrounds, medical histories and healthcare settings. Future studies should also distinguish between Alzheimer’s disease, vascular dementia and mixed causes, because diabetes may affect those pathways differently.

For now, metformin remains a medicine used primarily to manage type 2 diabetes and, in selected high-risk adults, to help delay diabetes. Current diabetes guidance does not support prescribing it solely to prevent dementia. Long-term users should continue to discuss the established benefits, side effects, kidney function and vitamin B12 monitoring with their clinicians.

The preliminary finding is nevertheless important. If a safe, inexpensive and widely available medicine can eventually be shown to lower dementia incidence, the public-health implications would be considerable. The appropriate conclusion at this stage is cautious optimism: the signal is credible enough to justify further study, but not strong enough to change dementia-prevention practice.

Sources