Do GLP-1 Drugs Reduce Dementia Risk? Three Analyses, Three Answers

Three respectable analyses published in 2025 looked at whether these drugs reduce dementia risk, and they do not agree. A meta-analysis of 26 randomized trials found a 45% reduction.^[1] A 396,963-patient cohort study found a 33% reduction.^[2] A continuously updated network meta-analysis in the BMJ found no significant effect at all, and graded its own certainty as low.^[3] Anyone who has read one headline about this has read a third of the evidence.

Why anyone thinks there might be an effect

Diabetes is an established risk factor for dementia, and the mechanisms proposed to link them — vascular damage, insulin signaling in the brain, chronic inflammation — are all things a glucose-lowering drug might plausibly touch. GLP-1 receptors also exist in the brain, which is why these drugs affect appetite at all, so a direct neurological effect is not far-fetched.

Plausible mechanism plus observational signal is exactly the combination that has repeatedly failed in dementia research. Hormone replacement therapy, vitamin E and NSAIDs all looked protective in cohort studies and did not survive randomized testing. That history is the reason to read what follows carefully rather than hopefully.

What each analysis found

The first two are the ones that made the news. The third is the one that keeps being updated, and its authors describe the question as uncertain in as many words.^[3]

The disagreement inside the positive results

Here is a detail that gets lost, and it is more informative than any single number. The meta-analysis of trials found GLP-1 drugs significantly protective and SGLT2 inhibitors not (OR 1.20, 0.67–2.17), with a formal test confirming the two classes behaved differently.^[1] The cohort study found both classes protective, and no difference between them at all (HR 0.97).^[2]

If this were a straightforward property of the drug, those two results would agree.

That they do not points at something else — possibly that what is being measured is better diabetes care rather than a specific molecule, possibly that dementia diagnoses in health records depend on who is being seen and how often. Both studies are good. The tension between them is a finding in itself.

What would settle it

A trial designed for the question, powered for it, and long enough to matter. Those are being run, and they will report years from now.

Until then the honest position is the BMJ’s: uncertain. That is not a satisfying answer and it is the accurate one, and it is worth holding against the marketing that has already begun. If someone is selling you a GLP-1 on a dementia claim, they are ahead of three analyses that cannot agree with each other.

For what these drugs demonstrably do across other systems, our beyond-the-scale article sorts the established effects from the associations, and the substance use piece works through a similar pattern of large cohort signals meeting disappointing trials.