GLP-1 Safety Signals in FAERS

A 2026 pharmacovigilance study found hundreds of statistical safety signals in reports involving GLP-1 receptor agonists, but its most useful result was methodological. Requiring several detection methods, checking whether the drug was named as the primary suspect, and testing signals in a second national database removed some dramatic-looking associations while retaining others. The findings can guide investigation. They cannot show how often an event occurs or prove that a medicine caused it.^[1]

The analysis covered 32 quarterly files from the US FDA Adverse Event Reporting System, or FAERS, from 2018 through 2025. The FDA describes FAERS as a repository for adverse-event reports, medication-error reports and product-quality complaints linked to adverse events.^[2] The study compared 288,408 unique reports involving GLP-1 receptor agonists with 11,319,150 reports involving other medicines. Across 22,059 drug-event pairs, 544 met the study's rule for a signal in at least three of four statistical methods.^[1]

How the researchers screened the reports

The team used four disproportionality methods. These methods ask whether a particular event is reported more often with a medicine than expected from the rest of the database. The study counted an event as a signal only when at least three methods agreed. That agreement rule is stricter than relying on one reporting-odds calculation alone, but it still detects unusual reporting patterns rather than clinical risk.^[1]

The researchers then stress-tested the 30 signals with the highest reporting odds ratios. Their primary-suspect analysis retained 21 and demoted nine that appeared to be driven by reports in which the GLP-1 medicine was listed only as a secondary suspect. Capillaritis, an inflammation of small blood vessels in the skin, fell from a reporting odds ratio of 18.6 to 0.27. A heart-sounds term fell from 84.3 to 0.55.^[1]

That change matters because people often take several medicines and may have more than one condition. A drug can appear in a report without being the reporter's main suspected cause. Sensitivity analysis does not solve confounding or uneven exposure time. It also cannot repair duplicate submissions and missing clinical detail. It does show how much a headline result depends on one attribution choice.^[1]

Which signals remained

The stable leading signals were mainly gastrointestinal and hepatobiliary, meaning they concerned digestion, the liver, or the gallbladder. The paper reported a reporting odds ratio of 64.8 for cyclic vomiting syndrome, concentrated in semaglutide reports; 84.2 for gallbladder injury; and 45.1 for impaired gastric emptying. These values describe relative reporting within FAERS. They are not percentages and should not be read as incidence or relative risk.^[1]

The analysis also identified a liraglutide-concentrated medullary thyroid cancer signal with a reporting odds ratio of 41.1 and a semaglutide-concentrated optic ischaemic neuropathy signal of 26.5. Optic ischaemic neuropathy involves reduced blood flow to the optic nerve. A concentrated signal can reflect true risk, indication, prescribing patterns, time on the market, publicity, or a mixture of these factors. The study design cannot separate those explanations by itself.^[1]

Readers looking for clinical context can review our evidence summaries for semaglutide and liraglutide. Those pages keep trial and label evidence separate from spontaneous-report screening. This FAERS analysis does not authenticate products sold outside regulated supply chains, even when those products use the same molecule names.

The psychiatric finding and the publicity check

At class level, only semaglutide met the study's signal criterion for suicidal ideation, with a reporting odds ratio of 2.30. That does not establish that semaglutide causes suicidal thoughts. Psychiatric symptoms can be influenced by underlying illness, the reason a medicine was prescribed, concurrent treatment, and reporting attention. FAERS also lacks the complete exposure denominator needed to calculate an event rate among all users.^[1]

The researchers tested whether the European Medicines Agency's 2023 announcement about the issue changed the reporting pattern. Their interrupted time-series analysis found no statistically detected level change at the announcement, with p=0.39, and no slope change, with p=0.75. The median reported time to onset for psychiatric events was 47 days. These checks argue against a simple publicity spike explaining the pattern, but they do not remove other biases or establish causality.^[1]

Why replication helps but does not prove causation

The study compared its priority findings with JADER, Japan's adverse-event reporting database. Sixteen of 17 priority signals, or 94.1%, replicated there. Seeing a similar reporting imbalance in another system makes a purely local database artifact less likely. Both systems still rely on spontaneous reports, however, so replication does not turn the result into a randomized comparison or an incidence estimate.^[1]

A signal is best treated as a question for follow-up. Depending on the event, that follow-up might use medical-record validation, claims data with exposure denominators, prospective cohorts, randomized trials, or regulator review. The current paper helps prioritize those questions because it shows which associations survived several analytic challenges and which collapsed under a simple primary-suspect restriction.^[1]

What we do not know yet

The analysis cannot determine the frequency of any event among people taking a GLP-1 receptor agonist. It cannot fully adjust for differences in the patients receiving semaglutide, liraglutide, or other medicines. It also cannot verify every diagnosis, exclude all duplicate reports, or prove that one drug is safer than another. The authors explicitly describe the results as hypothesis-generating rather than causal.^[1]

The practical reading is narrow. Gastrointestinal and gallbladder signals remained prominent, several rarer signals merit further study, and some eye-catching associations were highly sensitive to suspect-drug attribution. None of that supports starting, stopping, or switching treatment without clinical advice. Questions about symptoms or an individual medicine belong with a qualified healthcare professional who can consider the approved label, medical history, and alternatives.

Medical disclaimer: This article is for educational and journalistic purposes only and does not constitute medical advice. Peptides discussed may be classified as prescription medicines or research chemicals depending on your jurisdiction. Always consult a qualified healthcare professional before using any peptide product. PeptideMethods.com does not sell, distribute, or facilitate the sale of any peptide product.