Sellers describe what they stock as a class. Two molecules, a few formats, one story about how the class works — and a price gap between the two that the story never quite explains. This study is the clearest published reason to doubt that the class is a single thing, because it took four drugs in the class and analyzed them one at a time.
The cohort
Swedish national health registers identified 95,490 people carrying a diagnosis of depression or an anxiety disorder who used any antidiabetic medication between 2009 and 2022 — 59.7% women, mean age 50.6 years. Of those, 22,480 used a GLP-1 at some point. [1] The primary outcome was a composite of worsening mental illness: psychiatric hospitalization, more than fourteen days of psychiatric sick leave, hospitalization for self-harm, or death by suicide. Those are hard registry events rather than questionnaire scores, which is the study’s main strength.
The design compares periods when a person was taking a drug against periods when the same person was not. That removes everything fixed about an individual — genetics, history, personality, baseline severity — without removing what changes over time.
The molecules diverge
Against non-use, semaglutide carried an adjusted hazard ratio of 0.58, 95% CI 0.51 to 0.65. Liraglutide came in at 0.82, 95% CI 0.76 to 0.89. Exenatide was 1.01, 95% CI 0.69 to 1.46, and dulaglutide 1.01, 95% CI 0.85 to 1.20. Two drugs in the same class, measured the same way in the same people, showed no association at all.
On the secondary outcomes the pattern held. Semaglutide was associated with lower risk of worsening depression (0.56, 0.44 to 0.71), worsening anxiety (0.62, 0.52 to 0.73) and worsening substance use disorder (0.53, 0.35 to 0.80). Liraglutide moved only depression, at 0.74, 0.64 to 0.87. Taken as a group, the drugs were associated with reduced self-harm, 0.56, 0.34 to 0.92.
What an observational design can and cannot do
The within-individual comparison is a good one and it is not randomization. People start a drug at a moment when something has changed — a new clinician, a new diagnosis, a renewed intention to get healthier. Any of those could improve mental health on its own, and they arrive at the same time as the prescription.
The authors say plainly that randomized trials are warranted. That is the right conclusion for a signal this size, and it is worth noting what the signal is not: nobody prescribes these drugs for depression, and nothing here says they should be. Reasons people actually stop are a different matter, and the survey of quitters puts cost first.
One more thing worth saying because it is rare on this site. The funding came from two private foundations and a government ministry, not from a drug manufacturer. Several authors declare pharmaceutical relationships, but with psychiatric drug companies rather than with the makers of the drugs under study.
What it changes about buying
Nothing directly, and be suspicious of anybody who says otherwise. This cohort had diabetes or was on antidiabetic medication and already carried a psychiatric diagnosis. A person buying semaglutide to lose thirty pounds is not in it. What the study does change is how much weight a class-level claim should carry, and class-level claims are most of what this market publishes.
The practical version is short. If a benefit was demonstrated for one molecule, it was demonstrated for one molecule. Check which drug the study used before letting it justify a purchase, and check what each molecule costs here separately, because the prices diverge too.