Neil Sehgal, a doctoral student in CIS, used AI to analyze roughly 150,000 Reddit posts about gray-market versions of retatrutide, a new, unapproved GLP-1. Credit: Sylvia Zhang, Penn Engineering
Artificial intelligence has given the first systematic look at side effects reported by people using gray-market versions of retatrutide, a GLP-1 drug still in clinical trials. The most frequently reported symptom was increased appetite, not the gastrointestinal complaints seen in trials.
The study, published in JAMA Network Open, examined nearly 150,000 Reddit posts and comments from roughly 40,000 users discussing unapproved versions of retatrutide, a next-generation weight-loss drug not yet cleared for use.
Unlike semaglutide, which activates one hormone receptor, or tirzepatide, which activates two, retatrutide activates three, raising the question of whether users experience a different pattern of symptoms.
AI tools identified about 14,000 users who described personally taking an unapproved version, extracted the symptoms they attributed to the drug and matched those descriptions to standardized medical terminology. That narrowed the sample to 7,823 users reporting at least one classifiable symptom beyond expected effects such as weight loss or appetite suppression.
Among those users, increased appetite was the most commonly reported symptom, an unexpected result given retatrutide’s weight-loss effects in Phase 2 trials, where nausea and other gastrointestinal complaints were the most common side effects. Compared with an earlier analysis of semaglutide and tirzepatide, the retatrutide posts contained proportionally more reports of appetite-related symptoms, sleep disturbances, increased heart rate and muscle-related changes, and fewer reports of nausea and other gastrointestinal symptoms.
Another research group was inspired to examine the same symptoms in the FDA’s adverse-event database, and its study, published in Obstetrics & Gynecology, also found disproportionate reports of certain menstrual changes.
“That’s what we hope will happen with this study,” said Neil Sehgal, first author of the study and doctoral student in computer and information science at the University of Pennsylvania. “Other researchers can take what we find on Reddit and look for the same patterns in other datasets, using entirely different methods. Whether those signals hold up or not, the goal is to build a better understanding of what people are actually experiencing.”
The researchers hope AI-assisted social media analysis can flag potential safety concerns earlier, giving clinicians and public health researchers a starting point for further investigation.
Data from University of Pennsylvania School of Engineering and Applied Science