Within-individual study designs resolve GLP-1 psychiatric safety divergence by eliminating confounding by indication that creates spurious risk signals in matched cohort studies

The apparent divergence in GLP-1 psychiatric safety evidence—matched cohort studies showing 195% increased MDD risk versus RCTs and within-individual studies showing protective or neutral effects—is resolved by understanding confounding by indication. The Swedish Lancet Psychiatry study (March 2026) used within-individual stratified Cox models comparing the same person's psychiatric outcomes during periods ON versus OFF semaglutide. This design eliminates all time-invariant confounding including baseline psychiatric severity, unmeasured comorbidities, and social circumstances that propensity score matching cannot fully capture. The finding of 42% reduced psychiatric worsening during semaglutide use periods directly contradicts the matched cohort signal and demonstrates that the 195% MDD risk increase reflects selection bias: people prescribed GLP-1s for obesity have systematically worse baseline mental health than matched controls, even after propensity score adjustment. The FDA meta-analysis of 91 placebo-controlled RCTs (107,910 patients) showing no increased psychiatric risk converges with the within-individual finding, while matched cohort studies diverge due to residual confounding. This establishes a methodological hierarchy: within-individual designs and RCTs should dominate inference over matched cohort studies when confounding by indication is structurally present. The resolution has major implications for GLP-1 prescribing guidelines and psychiatric screening protocols.