Patients on GLP-1 drugs describe burning skin, tingling, and oral numbness. The clinical-trial language for this is paraesthesia and dysesthesia (abnormal cutaneous sensation), and the question coming out of social media reports and case-aggregator blogs has been whether these scattered descriptions belong to one class-wide effect or to distinct drug-specific patterns. A pharmacovigilance disproportionality analysis published in the European Journal of Clinical Pharmacology ↗ on May 22 used the WHO VigiBase database (the global spontaneous-report system maintained by Uppsala Monitoring Centre) to ask the question quantitatively. Each agent in the class came out with its own subtype.
The investigators queried VigiBase under ATC class A10BJ, the GLP-1 receptor agonists (exenatide ↗, semaglutide ↗, liraglutide, dulaglutide, lixisenatide), and added tirzepatide ↗ (a GIP/GLP-1 dual agonist filed under a separate ATC stub). The endpoint was the High Level Term "Paraesthesia and dysesthesia," meaning any abnormal cutaneous sensation including burning, tingling, hyperaesthesia (over-sensitive skin), hypoaesthesia (numbness), and oral variants. The summary statistic was the Information Component, a Bayesian disproportionality score that flags when a drug-event pair appears more often than chance against the rest of the database.
Three signal patterns separated cleanly. Exenatide ↗, the longest-marketed agent in the class, was significantly associated with hypoaesthesia and oral paraesthesia, meaning numbness, particularly around the mouth. Semaglutide ↗ and tirzepatide ↗ both produced a hyperaesthesia signal, meaning over-responsive skin. Semaglutide ↗ carried an additional and distinctive disproportionality for dysesthesia per se and for the burning-skin variant; that combination did not turn up on any other agent in the analysis. The authors characterize burning skin as a discrete clinical form rather than a synonym for general paraesthesia.
The dose pattern was direction-consistent across agents. Reports clustered at higher doses and on the higher-potency molecules, the same axis that drives most of the gastrointestinal tolerability picture for this class. The French Pharmacovigilance database, which holds narrative-level detail the global VigiBase set does not, supplied the clinical course. Discontinuation produced spontaneous favourable resolution in most reported cases. Rechallenge attempts (resuming the drug after stopping) surfaced in the narratives, presumably for clinical-need reasons; those are the rows that matter most for whether the signal is causal versus coincident, and the paper reports their existence without a tabulated outcome rate.
This sits alongside a separate PLOS Medicine meta-analysis on optic-nerve ischemia ↗ published the day before, which produced its own GLP-1 safety stack. The two pieces are scoped differently. The NAION analysis was a quantitative pooled hazard ratio (semaglutide ↗ vs all-other glucose-lowering, HR 2.17, absolute floor about 1 additional case per 7,000 patient-years). The current paraesthesia analysis is qualitative-disproportionality. It tells you which subtype each drug owns, but it cannot tell you how often. VigiBase is a denominator-free signal-detection tool. The Information Component flags asymmetries in spontaneous reporting; it does not give you cases per patient-year.
What this study moves is the structure of the safety conversation, not its magnitude. Burning skin on semaglutide ↗ is now in print as a distinctive sub-form, separable from the more general paraesthesia signal that the class has carried for years. Tirzepatide ↗ gets a more focused hyperaesthesia profile. Retatrutide ↗, still pre-approval and visible mainly via its trial dataset rather than its (small) post-marketing footprint, did not separately disproportionate here but is named in the discussion as carrying the same clinical-trial signal the older agents do.
The mechanism is unsettled. None of these molecules cross the blood-brain barrier in bulk; peripheral GLP-1R is expressed on neurons of the dorsal-root and trigeminal ganglia, which is the obvious anatomical substrate for sensory side effects but does not by itself explain the molecule-by-subtype mapping. The next data point worth tracking is whether prospective ascertainment in the TRIUMPH safety pool reproduces the semaglutide ↗-versus-tirzepatide ↗ subtype split, since that comparison is what would let the field decide whether the difference is pharmacological or detection-bias.