Topic and Sentiment Trends in Semaglutide Discussions on Twitter: Subpopulation-Based Longitudinal Analysis (Preprint)

BACKGROUND User experience has a significant impact on pharmaceutical drug effectiveness. Social media platforms, particularly Twitter, have become prominent spaces where individuals share their medication-related experiences, especially with widely marketed drugs such as semaglutide. Despite the volume of conversation, a comprehensive understanding of how various user subpopulations respond to and participate in semaglutide-related discussions remains underdeveloped. OBJECTIVE This study aims to explore how semaglutide is perceived and discussed across different Twitter user groups. Within these user groups, we investigate (1) the evolution of sentiment patterns toward semaglutide and (2) the evolution and prevalence of semaglutide-related discussion topics. METHODS We prepare a dataset consisting of 859,751 tweets pertaining to semaglutide, along with related metadata, that were posted between July 2021 and April 2024. We apply sentiment analysis and topic modeling to the snippets of the tweets and augment our dataset with the results. We group the dataset by specific user attributes and time periods to analyze the evolution of sentiment patterns and discussion topics across user subpopulations. RESULTS Our analysis reveals a mean sentiment score of -0.24 across all tweets, with all user subpopulations experiencing a decline in sentiment during the study period. User discussions focus on semaglutide’s applications (particularly in weight loss) and potential side effects, along with economic factors and celebrity/political influence. We also uncover differences in sentiment and discussion topics across user subpopulations. As a notable example, organizational user accounts consistently express less negative sentiment (mean score -0.014) than individuals (-0.24), particularly in discussions related to drug efficacy and regulatory concerns. Temporal analysis shows a marked decrease in sentiment during the November 2022-January 2023 period, coinciding with regulatory announcements about potential adverse effects. In addition, we observe regional and gender-based variations, such as the increased likelihood of female users engaging in discussions involving celebrities and politicians (19.24% vs. 14.6% for male users) and male users expressing more positive sentiment overall. CONCLUSIONS This study helps advance the understanding of how diverse user groups perceive and discuss one of the world’s most widely marketed drugs. Although we observe a general throughline of negativity, there are nuanced differences among the subpopulations. Our results offer valuable implications for healthcare communication strategies and pharmacovigilance.