Adults with a psychiatric diagnosis who started semaglutide had fewer cognitive problems recorded over the next year than similar patients who started most other diabetes drugs. The exception is the finding worth holding onto: against one comparator, a pill that works on the same hormone system, the advantage disappeared.
The study, published August 20 in BMJ Mental Health ↗, pulled de-identified US electronic health records from a psychiatric-care database spanning 1999 to 2024. The researchers found 13,007 adults, average age 62, who carried a psychiatric diagnosis and had a clinician rate their cognition both before and after they began a diabetes drug. Of those, 1,261 started semaglutide, the GLP-1 receptor agonist sold as Ozempic and Wegovy. The rest started something else, and that comparison is the whole study.
Cognition here was not a formal test. It was a 0-to-100 score built from routine clinician notes on memory, attention, orientation and other functions, where 0 means no problems recorded and 100 means problems noted across every domain. Over 12 months the semaglutide group scored 11.1. The group on no diabetes drug scored 14.9, which works out to roughly a quarter fewer recorded cognitive signs on semaglutide (mean ratio 0.75). Semaglutide also came out ahead of glipizide, an old sulfonylurea that squeezes the pancreas to release insulin (13.5, mean ratio 0.83), and ahead of empagliflozin, an SGLT2 inhibitor that flushes glucose through the urine (13.7, mean ratio 0.81). The gap was widest for memory. It was consistent across depression, psychosis and bipolar disorder in size, but only reached statistical significance in major depression.
Then the comparator that behaved differently. Sitagliptin, the fourth drug, scored 12.5, and against it semaglutide's edge shrank to a mean ratio of 0.89 and stopped being statistically significant (the confidence interval crossed the no-difference line, and the p value was 0.50, effectively a coin toss). Sitagliptin is not a random control. It is a DPP-4 inhibitor, a pill that blocks the enzyme that breaks down the body's own incretin hormones, so it raises natural GLP-1 rather than injecting a synthetic version. The one drug semaglutide could not clearly beat is the one that nudges the same pathway.
That single null does a lot of work. Read one way, it suggests any cognitive benefit here is a property of raising incretin signaling in general, not of semaglutide in particular, which would fit a mechanism story rather than a marketing one. Read the other way, it is a warning about confounding. Sitagliptin patients look more like semaglutide patients than the untreated or glipizide groups do (similar enough for a clinician to reach for either), so matching them equalizes the hidden differences (health-seeking behavior, baseline severity, who gets offered a newer drug) that could manufacture an apparent benefit against sicker or less-treated comparisons. Both readings point the same direction: this is not evidence that semaglutide specifically protects the brain.
The rest of the caveats are the standard ones for this kind of work, and the authors flag them. Clinician-recorded cognitive signs are softer and noisier than a validated test battery, and they depend on what a busy provider bothered to write down. Retrospective records cannot randomize who gets which drug. Confounding by indication runs through every arm. The benefit reaching significance only in major depression, where semaglutide's mood and metabolic effects could move a symptom score on their own, muddies whether cognition is the thing being measured at all. The authors call for randomized trials to test causality, which is the honest place to land.
This is the third time peptidemodel has covered a GLP-1 cognition signal that leans on a database rather than a hard endpoint, after a piece where semaglutide lowered a dementia risk score while the disease itself stayed unproven ↗. The pattern keeps repeating: the association shows up, and then a closer look at the comparators dissolves the part that would have made it a drug story. Semaglutide holds a card ↗ on the platform under the GLP-1 receptor ↗ and neuroprotective ↗ targets. Sitagliptin, empagliflozin and glipizide are small molecules, not peptides, and have no cards here.