The Lancet Psychiatry's April issue carries Heidi Taipale's national Swedish cohort study of 95,490 individuals diagnosed with depression or anxiety between 2009 and 2022. Of those, 22,480 used GLP-1 receptor agonists during follow-up. Within-person analysis shows semaglutide associated with adjusted hazard ratios of 0.58 for worsening mental illness, 0.56 for worsening depression, 0.62 for anxiety, 0.53 for substance-use disorder, and 0.56 for self-harm. [1] Yesterday's paper carried the cohort's class-specificity finding — semaglutide and liraglutide producing the signal, exenatide and dulaglutide producing nothing. Today's frame is what sits next to it.

A 2024 Scientific Reports propensity-matched study of 162,253 obese GLP-1 users versus comparators found the opposite signal: a 98 percent increased risk of psychiatric disorders, 195 percent higher risk of major depression, and 106 percent elevated risk of suicidal behavior. [2] The TriNetX-database design excluded patients with prior psychiatric history; the Swedish design enrolled them. The two studies are not testing the same question. They are testing the safety of the same molecule in two different populations with two different methods, and they produce signals that point in opposite directions.

The FDA has both papers. Neither EMA nor FDA has found a causal link between GLP-1 medications and suicidal ideation in regulatory review. [3] But the agency's labeling decisions for semaglutide depend on which cohort it weighs more. The Wegovy oral pill scaled to 1.3 million weekly prescriptions in Q1 with 80 percent of new patients GLP-1-naive; expert reaction at the Science Media Centre was direct: this is observational, RCTs are needed, the signal is reassuring on safety but not yet evidence of therapeutic effect on depression. [4] An FDA action gap is not a non-decision. It is the decision to let prescribers and pharmacy benefit managers operate without label guidance while two opposing cohorts of nearly equal sample size sit unreconciled. The clinical workflow already runs ahead of the regulatory text.

-- NORA WHITFIELD, Chicago