A team from the University of Pennsylvania had five years of public discussions analyzed by language models and drew signals on menstrual disorders and thermoregulation.
Researchers at the University of Pennsylvania's engineering school analyzed over 400,000 messages published on Reddit by approximately 70,000 users over five years to investigate the effects reported by patients on semaglutide (Ozempic, Wegovy) and tirezepatide (Mounjaro, Zepbound). Language models have been used to translate informal descriptions of symptoms into MedDRA standardized medical terminology, allowing for comparable occurrences where there was only free text.
The work, signed by Neil Sehgal as first author and Sharath Chandra Guntuku as last author, with Lyle Ungar and Jena Shaw Tronieri, was published in Nature Health. It reports that 44% of users report at least one adverse reaction, that nearly 4% describe reproductive symptoms, including menstrual cycle irregularities, and that fatigue is the second most cited symptom. Two signal families emerge as worthy of a dedicated study: cycle disorders and thermal regulation problems, chills and hot flashes.
Signals, not evidence
The authors are explicit about the limit: there is no evidence in these data that drugs cause these symptoms. Reports are spontaneous, significant reporting biases, and concurrent pathologies or treatments are not controlled. Reddit's population is younger, more male and more American than the one who actually takes these treatments. The interest of the method is elsewhere: to identify, on a very large scale and at a lower cost, leads that clinical trials do not have the statistical power to detect.
Read the full article on the original source
Go to the source →