Penn researchers publish an AI study of 400,000 GLP-1 Reddit posts

Researchers at the University of Pennsylvania School of Engineering and Applied Science analyzed more than 400,000 Reddit posts from people using semaglutide (Ozempic, Wegovy, Rybelsus) and tirzepatide (Mounjaro, Zepbound) to see what patients say about side effects outside the setting of a clinical trial ^[1]. The study, published in Nature Health, drew on posts from nearly 70,000 Reddit users spanning more than five years ^[1].

The team used large language models to sort informal, everyday descriptions of symptoms into standardized medical categories, a process that would have taken far more manual effort in the past ^[1]. About 44% of users in the study described at least one side effect, and gastrointestinal problems were the most common complaint, which matches what is already known about these drugs ^[1].

Two other patterns stood out. Nearly 4% of users who reported side effects described reproductive symptoms, including bleeding between periods, heavy bleeding, and irregular cycles ^[1]. Users also described chills, feeling unusually cold, hot flashes, and fever-like symptoms ^[1]. Fatigue was the second most frequently reported complaint overall, even though it has not shown up often enough in many clinical trials to meet standard reporting thresholds ^[1].

The researchers are careful to say this does not prove the drugs caused these symptoms. "We can't say that GLP-1s are actually causing these symptoms," said Neil Sehgal, the study's first author, though he noted the menstrual reports were frequent enough in a mixed-gender sample to be "a signal worth investigating" ^[1]. Co-author Jena Shaw Tronieri pointed to the hypothalamus, a brain region these drugs are thought to engage that also regulates hormones and body temperature, as a possible biological link worth studying further, while stressing that this does not confirm cause and effect ^[1].

Why it matters for patients

Clinical trials are built to catch the most serious and common safety problems, but they are not designed to capture every symptom that matters to the people taking a drug once it reaches a much larger population ^[1]. Lyle Ungar, a co-author on the study, said trials "can fail to find what symptoms patients are most concerned about," even though social media conversations are not necessarily representative of everyone using the medication ^[1].

For patients, this means that some experiences discussed widely online, like menstrual irregularity or feeling unusually cold, may not be prominently listed on drug labels yet, even though they are common enough in patient conversations to draw research interest ^[1]. The study is not a reason to conclude a symptom is caused by a GLP-1 drug, but it may help explain why some patients feel their experiences are not fully reflected in official information.

The researchers also caution that Reddit users skew younger, more male, and more U.S.-based than the general population of people on these drugs, so the findings may not generalize to everyone taking semaglutide or tirzepatide ^[1]. Sehgal noted that a menstrual-symptom rate of roughly 4% among a sample that includes men would likely be higher in a female-only group, but the exact scope remains unclear ^[1].

What happens next

The research team says it wants to expand the analysis beyond Reddit and beyond English-language posts to see whether the same symptom patterns show up among different groups and platforms ^[1]. Ungar said it is not yet known whether the Reddit findings reflect the experience of GLP-1 users worldwide or are specific to the type of person who posts on Reddit in the United States ^[1]. The authors frame this kind of AI-driven "social listening" as a potential early-warning complement to clinical trials, not a replacement, particularly useful for catching signals faster when a drug's use expands quickly ^[1]. One of the four authors, Tronieri, has disclosed a research grant from Novo Nordisk and consulting fees from Currax Pharmaceuticals; the other authors reported no conflicts of interest ^[1].