People with bipolar disorder were about 21 percent less likely to land in a psychiatric hospital during the stretches when they were taking semaglutide, the drug sold as Ozempic and Wegovy, than during stretches when the same people took no GLP-1 drug at all. Two close chemical relatives, liraglutide and dulaglutide, showed no such effect.
That split is the story. The finding comes from a Swedish national-registry study published June 21 in Acta Psychiatrica Scandinavica ↗, covering everyone in the country diagnosed with bipolar disorder who also used a diabetes drug between 2009 and 2024. That worked out to 14,694 people, of whom 5,200 used a GLP-1 receptor agonist at some point.
GLP-1 receptor agonists mimic a gut hormone the body releases after eating. They are prescribed for type 2 diabetes and obesity, both of which run heavily alongside bipolar disorder. What they do to mood disorders has been an open question, and the authors framed it against a backdrop of stalled bipolar drug development, where genuinely new treatments have been scarce for years.
The design is the part worth trusting. Rather than comparing GLP-1 users against different people who never took the drugs, the researchers used a within-individual model: each patient serves as their own control, with hospital admissions during treated months weighed against admissions during untreated months for that same person. That cancels out the stable differences between people (genetics, baseline severity, lifestyle) that wreck most observational drug comparisons. What it cannot cancel is anything that changes inside a person over time and happens to track with starting the drug.
Within that design, semaglutide ↗ carried an adjusted hazard ratio of 0.79 (95 percent confidence interval 0.69 to 0.91) for psychiatric hospitalization, the 21 percent reduction. It was also tied to a 17 percent lower risk of a hospitalization specifically following a bipolar relapse (aHR 0.83, 0.69 to 0.99), a number whose upper bound nearly touches the no-effect line. Sick leave taken for psychiatric reasons did not move for any of the drugs.
Then the class fell apart. Liraglutide ↗, the daily injection sold as Victoza and Saxenda, and dulaglutide ↗, the weekly Trulicity, were not individually linked to lower hospitalization. All three drugs hit the same receptor ↗, the GLP-1 receptor, which is what makes the divergence interesting rather than tidy. If the benefit were a straightforward consequence of activating that receptor, or of the weight and glucose changes the whole class delivers, the other two should have moved the needle too. The authors say plainly that this implies the signal "is not a class effect."
That leaves two readings, and the paper does not settle between them. Either semaglutide does something to the brain that liraglutide and dulaglutide do not, perhaps through differences in how well each molecule reaches central tissue, or semaglutide users in this cohort differ from the others in some time-varying way the model did not catch. A within-individual design narrows that second possibility but does not erase it.
It is not the first time semaglutide specifically has thrown a central-nervous-system signal that the rest of its mechanism does not obviously explain. A separate analysis this month tied the drug to roughly half the seizure risk ↗, an effect weight loss alone did not account for. A registry association is not proof, and the authors are explicit that the semaglutide result now needs a randomized controlled trial to confirm. But a drug-specific finding inside a class is exactly the kind of result that points a trial at the right molecule instead of the whole shelf.