Study wrapper · #119
Psychiatric Safety Signals of GLP-1 Receptor Agonists: A FAERS-Based Pharmacovigilance Study with Explainable Machine Learning.
Editor's note
A disproportionality (pharmacovigilance) study of FDA adverse-event reports enhanced with machine learning, and its design dictates cautious reading. Spontaneous reporting databases like FAERS can flag potential signals but cannot establish causation, incidence, or risk, they are shaped by reporting biases, media attention, and prescribing volume. Across 211,195 cases, sixteen psychiatric preferred terms met signal criteria, with suicidal ideation most frequently reported (ROR 2.95), and signal magnitudes were consistently higher for semaglutide than tirzepatide. An explainable-ML model (AUROC 0.816) identified semaglutide use and younger age (19-44) as positively associated with psychiatric-event reporting, and age 65+ as negatively associated. These are associations within reporting patterns, not clinical risk estimates, and the authors themselves call for prospective study. Notably, regulators including EMA and FDA have previously reviewed GLP-1 suicidality signals without establishing a causal link. Weight this as hypothesis-generating surveillance that flags subgroups for scrutiny, not as evidence that semaglutide causes psychiatric harm.
Plain-language abstract
This study looked at psychiatric side effects reported for GLP-1 medications, focusing on semaglutide and tirzepatide, using the FDA's public database of adverse-event reports from 2021 to 2025. Such databases collect reports from doctors, patients, and others, but they cannot prove a drug caused an event, they only show what gets reported, which can be skewed by news coverage and how widely a drug is used. The researchers used statistical methods to spot events reported more often than expected, and machine-learning models to find which case features were linked to psychiatric reports. Out of 211,195 cases, sixteen psychiatric conditions stood out, with suicidal thoughts reported most often. The statistical signals were consistently stronger for semaglutide than for tirzepatide. The analysis found that being 65 or older was linked to fewer psychiatric reports, while semaglutide use and younger age (19-44) were linked to more. The authors stress these are reporting patterns, not proof of risk, and that proper prospective studies are needed. Regulators have previously examined similar signals without confirming a causal link.