Penn mines Reddit for drug side effects

University of Pennsylvania researchers are testing whether Reddit can function as an early-warning system for drug side effects. In a study published April 10 in Nature Health, the team used AI to analyze 410,198 Reddit posts from 67,008 users who mentioned semaglutide or tirzepatide, the GLP-1 drugs sold under brands including Ozempic, Wegovy, Mounjaro and Zepbound. (seas.upenn.edu) The researchers said the method surfaced both expected complaints and signals that are less visible in clinical trials and drug labels. (nature.com) Penn identified two symptom clusters that it said warrant more study: reproductive symptoms, including irregular menstrual cycles, and temperature-related complaints such as chills and hot flashes. (seas.upenn.edu) ### Why are Penn researchers looking at Reddit in the first place? Sharath Chandra Guntuku, a research associate professor in computer and information science at Penn Engineering and the study’s senior author, said the point was to capture what patients report in their own words outside formal reporting systems. He said well-known side effects such as nausea helped validate that the method was detecting a real signal. (nature.com) Lyle Ungar, a Penn computer and information science professor and co-author, said clinical trials tend to identify the most dangerous side effects but can miss the symptoms patients talk about most. He said people taking these drugs are “swapping notes” in real time online in ways that may not reach doctors’ offices or official reports. (seas.upenn.edu) ### What did the AI actually find in those posts? Nature Health said the analysis covered Reddit posts from 2019 to 2025 and found a wider range of self-reported effects than standard labeling captures. The paper highlighted reproductive symptoms and temperature-related complaints as potential underrecognized effects alongside the expected gastrointestinal complaints. (seas.upenn.edu) Neil Sehgal, the paper’s first author and a Penn doctoral student, said nearly 4% of the Reddit users in the sample reported menstrual irregularities. He added that the share would likely be higher in a female-only sample and called it a signal worth investigating. (seas.upenn.edu) ### Does this mean Reddit posts can prove a drug caused a symptom? Neil Sehgal said no. He said the findings are not causal and do not show that GLP-1 drugs actually caused the symptoms described in posts. Nature Health described the work as a complement to traditional pharmacovigilance rather than a replacement for it. (nature.com) The paper said large-scale social media analysis can help detect emerging safety signals and broaden understanding of real-world drug experiences. ### Why does the language on Reddit matter? Penn’s release said social media can capture symptoms in the vocabulary patients actually use rather than only in clinical terminology. (seas.upenn.edu) That matters because product teams, researchers and safety monitors often need to translate everyday descriptions into standardized medical concepts before they can compare reports across sources. The Jefferson City News-Tribune article that highlighted the study pointed to everyday phrases such as “crashed,” “flared” and “wired but tired” as examples of the kind of language that can be useful when building symptom taxonomies from community posts. (nature.com) ### What happens next? The April 10 Nature Health paper leaves the next step with follow-up research and clinical review. Penn researchers said the Reddit signals should be treated as leads for further investigation, not as confirmed adverse events. The study remains one of the clearest recent examples of AI-based social listening being used as a pharmacovigilance input. (seas.upenn.edu) For now, the published record is the Nature Health paper by Sehgal, Guntuku and Ungar, along with Penn Engineering’s April 10 summary of the findings. (nature.com) (newstribune.com)