Using Large Language Models to Assess Public Perceptions Around Glucagon-Like Peptide-1 Receptor Agonists on Social Media

In the global context, the prevalence of obesity is on the rise, bringing significant impacts to public health. Obesity is independently associated with the incidence and mortality of cardiovascular diseases, with an estimated economic burden exceeding $200 billion annually for healthcare systems. In recent years, glucagon-like peptide-1 (GLP-1) receptor agonists have become practice-changing therapeutic options due to their weight loss and independent effects on reducing cardiovascular risks apart from diabetes management. Against this backdrop, researchers Sulaiman Somani, Sneha S. Jain, Ashish Sarraju, Alexander T. Sandhu, Tina Hernandez-Boussard, and Fatima Rodriguez from Stanford University conducted a study on public awareness of GLP-1 receptor agonists on social media. Their findings were published in the journal “Communications Medicine” in 2024.

The study analyzed over 390,000 discussions related to GLP-1 RAs on Reddit using a large language model. The results indicate a high level of public interest in this therapeutic option, with discussion topics primarily focusing on experiences with GLP-1 RAs for weight loss, comparisons of different GLP-1 RAs and their side effects with other therapies, issues related to the acquisition and supply of GLP-1 RAs, and the positive psychological benefits associated with their use and resultant weight loss. Notably, these discussions mostly exhibited neutral to positive sentiments.

The research methodology included essential steps such as data curation, topic modeling, and sentiment analysis. During the data collection phase, researchers used an application programming interface called PullPush to gather all discussions related to GLP-1 RA on Reddit. For the topic modeling phase, a pre-trained bidirectional encoder from transformer (BERT)-like architecture model was employed to embed the discussions. These embeddings were then utilized for topic separation using density-based clustering algorithms by reducing dimensional complexity. Sentiment analysis was conducted using a pre-trained BERT model to classify the emotions of social media posts.

The study’s findings revealed that there were up to 168 GLP-1 RA-related topics being discussed by the public on Reddit, which were further clustered into 33 discussion groups. These groups covered themes ranging from drug efficacy, side effect comparisons, acquisition methods, to positive mindset. Sentiment analysis showed that 31.8% of the discussion posts exhibited negative sentiments, 50.1% were neutral, and 17.4% expressed positive emotions.

The study concludes by emphasizing the high level of public interest in GLP-1 RAs and suggests potential public health interventions such as monitoring drug side effects, improving drug accessibility, and recognizing the dual benefits of these drugs for weight management. By using large language models and AI-driven topic modeling processes, the research demonstrates an effective tool for mining public sentiments on social media, which can guide future research and public health efforts.

The study published in “Communications Medicine” highlights the potential application of social media data in healthcare research and underscores the efficacy of large language models in processing and analyzing massive unstructured text data. It further illustrates the importance of understanding patient perspectives to guide clinical decisions, research, and policy efforts.