Heidi Taipale's national Swedish cohort study in The Lancet Psychiatry follows 95,490 people diagnosed with depression or anxiety between 2009 and 2022 who used antidiabetic medication, of whom 22,480 used GLP-1 receptor agonists. [1] Within-individual analysis — comparing periods of use against periods of non-use in the same person — produces an adjusted hazard ratio of 0.58 for worsening mental illness on semaglutide, 0.62 for anxiety, 0.53 for substance-use disorder, and 0.56 for self-harm. [1] The class-as-a-group signal lowered self-harm risk by 44 percent; the magnitude is concentrated in semaglutide and liraglutide. Exenatide and dulaglutide produced flat hazard ratios of 1.01 in the same data. The paper's May 7 brief on the opposing-study FDA-desk frame named the two-cohort question. Today the prescription-volume side of the question crosses a documentable threshold.

The opposite signal sits in a 2024 Scientific Reports propensity-matched study of 162,253 obese GLP-1 users versus comparators, which found a 98 percent increased risk of psychiatric disorders, a 195 percent higher risk of major depression, and 106 percent elevated risk of suicidal behavior in the GLP-1 cohort. [2] The two designs are not testing the same question. The Swedish cohort enrolled patients with prior psychiatric history; the TriNetX-database design that produced the opposite signal excluded them. The Swedish design uses each patient as their own control across treatment and non-treatment periods; the TriNetX design uses a matched comparator population. Both methods are observational; both produce hazard ratios that point in opposite directions and arrive on the FDA's desk in the same regulatory moment. [3]

The prescription-volume side has crossed two million scripts on the Wegovy oral pill alone since the January 5 launch. Novo Nordisk's CEO call Thursday named a "record-breaking start" — the second consecutive quarter that framing has been used. Foundayo's week-four print landed Friday at 7,335 scripts, behind the comparator. The combined U.S. oral GLP-1 cohort is now well over two million prescriptions, with 80 percent naive share. That base — patients new to GLP-1 therapy — is the population where the Lancet Psychiatry signal is most policy-relevant: previously untreated, in primary care settings, with diagnosed but managed anxiety and depression. The clinical workflow is running ahead of the label.

The label is what FDA action gates. Neither EMA nor FDA has found a causal link between GLP-1 medications and suicidal ideation in regulatory review to date; both agencies have been monitoring the signal since 2023 reports out of Iceland and Europe surfaced spontaneous reports of suicidal thoughts on semaglutide. [4] The Lancet Psychiatry result is the largest population-scale within-person dataset to enter the regulatory record. The Scientific Reports propensity-matched study, with its larger absolute sample but its prior-psychiatric-history exclusion, is the counterweight. The FDA's options at this point are three: revise the labeling toward a benefit signal (politically and methodologically aggressive given the contradiction); add a class warning that names the inconsistency (the conservative path); or take no labeling action and let the cohorts accumulate (the action gap that has held since 2023). The third path is the live one.

What the action gap means in practice is that prescribers and pharmacy benefit managers are operating without label guidance while two opposing observational cohorts of nearly equal sample size sit unreconciled. Primary-care reading of the Lancet Psychiatry paper — Taylor and colleagues' clinical commentary in the same issue — names "selection bias" as the most likely explanation for the contradiction: people who continue to use GLP-1 receptor agonists may be inherently different from people who do not. [5] That framing produces a research recommendation, not a label change: a randomized controlled trial of GLP-1 receptor agonists in people with diabetes, depression, and anxiety. No such trial is currently registered with the FDA. Novo Nordisk's PIONEER and SUSTAIN trials, which produced the cardiovascular and weight-loss labeling, did not power for psychiatric outcomes.

The class-specificity reading is the operational one. Semaglutide and liraglutide produce the signal; exenatide and dulaglutide do not. [1] If the Lancet Psychiatry result holds in randomized data, the implication is that semaglutide's pharmacology — its central nervous system penetration, its receptor-binding profile — produces the benefit, and that the GLP-1 class label is too broad to capture the molecule-specific effect. That is a structural argument the FDA's Division of Endocrinology has heard before in its diabetes-drug labeling history; it is what produced the differential labeling for the SGLT2 class on cardiovascular outcomes. The same architectural fix — molecule-specific labeling within a class — is the path the Lancet Psychiatry result quietly proposes.

What this paper now tracks: whether the FDA acknowledges the comparator-class divergence in any safety-signal communication this month, and whether Novo Nordisk or Eli Lilly fold the Lancet Psychiatry signal into post-marketing language before the Q3 close. The two-million-script threshold means the regulatory absence is now a population question. The cohort, the cohort's methodology, and the cohort's contradiction all sit on the same desk. The label has not moved.

-- NORA WHITFIELD, Chicago