The Lancet Psychiatry's 95,000-patient Swedish cohort study now sits on a Food and Drug Administration desk that already changed its position. In January 2026, the FDA's Division of Metabolism and Endocrinology Products requested that semaglutide and tirzepatide labels remove the suicidal-behavior warning the agency had added in 2023. The new cohort — 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 — confirms the disposition the FDA already made. The question on the desk is no longer whether the warning should come off. It is whether the new data tightens the removal, or whether the comparator-class counter-study sitting on ScienceDirect breaks it. [1]

The Lancet study is the larger of the two papers, and the better-powered. Across 95,000 Swedish adults with type-2 diabetes initiated on semaglutide between 2018 and 2024, the cohort tracked against propensity-matched controls who initiated DPP-4 inhibitors or SGLT-2 inhibitors. The mental-illness signals all favored semaglutide. The signal sizes — 42-percent lower hazard of worsening mental illness, 44-percent lower for depression, 38-percent lower for anxiety, 47-percent lower for substance-use, 44-percent lower for self-harm — are larger than any prior GLP-1 mental-health observational study has produced. Five separate endpoints in one direction is not the pattern that confounding alone tends to produce. [2] The paper's Sunday note framed the FDA disposition as Day Five; today is Day Six, and no FDA statement has emerged.

The comparator-class counter-study is the part of the regulatory question MSM has not picked up. A ScienceDirect publication in early May reported, across a U.S. claims-database cohort of similar size, that GLP-1 receptor agonists were associated with higher rates of psychiatric adverse events compared to SGLT-2 inhibitors and DPP-4 inhibitors when the comparator restriction was tightened to patients without prior psychiatric diagnoses. The signal direction was opposite to the Lancet's. The interpretation issue is the same one observational pharmacoepidemiology has been arguing for two decades: which comparator class produces the unconfounded comparison, and which one produces the regression-to-the-mean artifact?

The Science Media Centre's expert-reaction round-up captured the methodological problem in plain language. Patients prescribed GLP-1 receptor agonists are, on average, less depressed and less anxious at baseline than patients prescribed SGLT-2 inhibitors — because GLP-1 prescriptions require sustained adherence to injection or pill schedules that depressed and anxious patients adhere to less reliably. The Lancet cohort attempted to address this with propensity matching across forty-plus baseline covariates. The ScienceDirect comparator-class study attempted to address it by restricting to patients without prior psychiatric diagnoses. Neither approach eliminates the residual confounding that comes with self-selection into a therapy that requires sustained behavioral engagement. [3]

What the FDA has to decide is whether the larger, methodologically cleaner Lancet cohort supports the January warning-removal disposition more strongly than the ScienceDirect comparator-class study undermines it. The agency's January 2026 letter to manufacturers cited the 2023-2025 pharmacovigilance signal as not meeting the threshold for a class-wide warning. The Lancet cohort, on its face, provides the affirmative evidence the January letter did not cite. The ScienceDirect paper, on its face, provides the next signal the agency's own pharmacovigilance system will have to absorb. Both papers will sit inside the FDA Adverse Event Reporting System review the agency has run continuously since 2023.

The psychiatric-medication reframing is the part of the story Psychiatric Times has been printing. The Lancet paper has been read in some quarters as evidence that GLP-1 receptor agonists are not just weight-loss agents but next-generation psychiatric medications. The reframing is premature. The Lancet cohort tracked patients prescribed semaglutide for type-2 diabetes, not for psychiatric indication. The mental-health endpoints were secondary, not primary. The hazard ratios apply to a population taking the drug for its approved indication, not to a population starting the drug for depression. Whether the same signal would survive in a depression-indication trial is the prospective question the field does not yet have an answer to. [3]

What patient-advocates have asked for, in response to the cohort, is the FDA convening an advisory committee. The agency has not announced one. The disposition the January letter made was administrative, not adjudicative. The next administrative step — whether the warning removal is finalized, paused for the new data, or expanded with a positive psychiatric-benefit claim — would, under standard procedure, require an FDA committee hearing or a published guidance document. Neither has been scheduled. The regulator-inaction-compression frame the paper has tracked through this story compresses again at Day Six.

The Swedish cohort is the cleanest data the field has on this question. The comparator-class counter-study is the next data the field has produced. Both will sit inside the same regulator's silent disposition. The next move belongs to the FDA. [1]

-- NORA WHITFIELD, Chicago