Using the NIH All of Us database, a target trial emulation compared people starting metformin with those starting sulfonylureas. With stabilized inverse probability weighting, metformin was linked to a markedly lower risk of death without dementia (HR 0.61, 95% CI 0.49-0.75) and a smaller reduction in dementia incidence (HR 0.90, 95% CI 0.82-1.02). Estimates matched earlier analyses in UK primary care and Mass General Brigham records, and held when social and behavioral confounders were excluded, calendar year and prior drug exposure were adjusted, and people without baseline type 2 diabetes were included.
The dementia result deserves careful reading: the confidence interval crosses 1.0, so the headline benefit is a trend, not a statistically firm effect. The mortality signal is far stronger, which creates a competing-risk wrinkle. If metformin users survive longer, more of them live long enough to develop dementia, and the authors' competing-risk framework exists to handle precisely that tension. Sulfonylureas also cause hypoglycemia, which has itself been tied to cognitive decline, so part of the gap may reflect harm from the comparator rather than protection from metformin. Residual confounding, such as healthier patients being steered toward metformin, can never be fully excluded in observational data. This work is a preprint and has not been peer reviewed, so figures may shift. It is confirmatory, replicating prior signals across three health systems rather than overturning anything. Adults with type 2 diabetes should not choose a drug for dementia prevention on this basis; randomized trials such as the ongoing MAP study will be needed.