Popular weight loss and diabetes drugs such as semaglutide and tirzepatide have revolutionized the treatment of obesity and blood sugar control. Now, researchers at the University of Pennsylvania say artificial intelligence could also help spot side effects that patients discuss online but aren’t always fully reflected in clinical trials or official drug documentation.
In a new study published in natural healthResearchers analyzed more than 400,000 Reddit posts written by approximately 70,000 users over five years. Their findings highlighted several commonly discussed symptoms, including some that deserve more scientific attention, such as menstrual irregularities and temperature-related complaints such as chills and hot flashes.
“Some of the side effects we found, such as nausea, are well known, indicating that our method is picking up real signals,” said Sharath Chandra Gunthuk, research associate professor of computer and information science (CIS) at Penn Engineering and lead author of the study. “Underreported symptoms may be derived from the patient themselves and clinicians may want to pay attention to them.”
Lyle Unger, a professor at CIS and a co-author of the study, said social media could provide insight into concerns that patients don’t necessarily bring up during a consultation.
“Clinical trials typically identify a drug’s most dangerous side effects,” Unger says. “But sometimes we don’t find out what symptoms patients are most concerned about. Even though social media isn’t necessarily representative, a large collection of posts can reflect additional concerns.”
AI and Reddit reveal new GLP-1 concerns
The researchers stress that the study does not prove that the drug is the cause of the symptoms discussed online. Rather, the findings point to a pattern that may warrant further investigation.
“We can’t say that GLP-1 is actually causing these symptoms,” says Neil Sehgal, lead author of the study and a CIS doctoral student advised by Guntuku and Ungar. “However, nearly 4% of Reddit users in our sample reported menstrual irregularities, and the proportion would be even higher in a female-only sample. We think this is a signal worth investigating.”
The study builds on years of research examining online conversations for clues about drug side effects. In 2011, Ungar joined one of the earliest projects to mine user-generated internet content for reporting drug side effects.
“Online patient communities act like neighborhood grapevines,” Unger says. “People living with these drugs exchange notes with each other in real time and share experiences that are rarely documented in medical examinations or official reports.”
Researchers say these discussions have become a valuable source of health-related information, even though data collection and analysis has become more difficult over time as social media platforms expand.
“Clinical trials are the gold standard, but by design, clinical trials are time-consuming,” says Guntuku. “This is not a replacement for clinical trials, but it can go much faster, and when drugs go from niche to mainstream almost overnight, that speed is important.”
Large language models speed up side effect detection
One of the major challenges when studying online health discussions is scale. People describe symptoms in different ways, making it difficult to systematically compare social media posts and the standardized medical terminology from the Medical Dictionary for Regulatory Activities (MedDRA) that clinicians use to classify symptoms.
Things have changed with the rise of large language models such as GPT and Gemini. According to the researchers, these AI systems are now able to process vast amounts of online discussions faster and more consistently.
“Large language models allow us to perform this type of analysis much faster and at a level of standardization that was previously difficult to achieve,” Sehgal says.
Although not entirely representative of the general population, as Reddit users tend to be younger, more likely to be male, and disproportionately based in the United States, many of the symptoms reported were consistent with already known side effects of semaglutide and tirzepatide. Approximately 44% of users in the study mentioned at least one side effect, most commonly gastrointestinal problems.
Unexpected symptoms reported by GLP-1 users
What stood out to the researchers were symptoms that may not be fully represented by current drug labels or standard adverse event reporting systems.
Approximately 4% of users who reported side effects also mentioned genital symptoms such as irregular menstrual cycles, intermenstrual bleeding, and heavy bleeding.
Other users reported temperature-related symptoms, including chills, chills, hot flashes, and a fever-like feeling.
Fatigue also emerged as one of the most frequently discussed symptoms. In fact, it ranks as the second most common symptom reported by Reddit users, despite being less noticeable in many clinical trials.
“These drugs are thought to work by affecting a part of the brain called the hypothalamus, which helps regulate various hormones,” says Jenna Shaw Toronieri, senior research scientist at the Penn State Weight and Eating Disorders Center and co-author of the study. “That doesn’t mean the drug is necessarily causing these symptoms, but it may suggest that reports of menstrual changes or body temperature fluctuations are worth studying more systematically.”
Researchers hope to expand beyond Reddit
The researchers hope that their findings will encourage scientists and health care providers to pay closer attention to the types of side effects patients are discussing online.
“They’re clearly in the patient’s mind, and that’s worth paying attention to,” Sehgal says.
The researchers also plan to expand their analysis beyond Reddit and beyond English-speaking communities to determine whether similar patterns emerge on other social media platforms and populations around the world.
“We still don’t know for sure whether what we’re seeing on Reddit reflects the experiences of GLP-1 users around the world, or whether it’s unique to the type of people posting on Reddit in the United States,” Ungar says.
Ultimately, researchers believe, AI-powered analysis of social media conversations could become an important tool for identifying emerging concerns about medicines and health trends much faster than traditional systems.
For rapidly popular health products, especially substances sold in less regulated or unregulated markets, such as injectable peptides, online conversations on platforms like Reddit and TikTok can provide the earliest clues about what a user’s situation is.
“The key to this kind of approach is that you can move quickly, and that’s exactly when it’s most valuable,” says Guntuku.
The research was conducted at the University of Pennsylvania’s School of Engineering and Applied Sciences. The authors report no external funding. Mr. Tronieri reports receiving investigator-initiated grants from Novo Nordisk on behalf of the University of Pennsylvania and consulting fees from Currax Pharmaceuticals, LLC. The other authors report no conflicts of interest.
