Health-Related Rumor Debunking on Sina Weibo in China (2009-2024): 15-Year Retrospective Infodemiology Study.

Background: Understanding how health-related rumor debunking evolves and spreads on social media is critical for public health communication and policy. Existing research, however, has been largely crisis-centered-dominated by studies of specific events, such as the COVID-19 pandemic-and offers limited insight into the longer-term patterns of thematic evolution, demographic targeting, and engagement dynamics of official debunking practices. Objective: This study aimed to provide an integrated understanding of official health rumor debunking in China over a 15-year period by delineating its thematic evolution, demographic disparities, and features associated with engagement and information diffusion. Methods: We collected rumor debunking posts published on Sina Weibo between 2009 and 2024. A 2-stage classification pipeline using Sentence Transformer Fine-Tuning was constructed to identify health-related debunking posts. We applied BERTopic (BERT: Bidirectional Encoder Representations from Transformers) to map the thematic landscape and used a large language model to extract demographic mentions (gender, age, and social roles), and health content referenced in the posts. Finally, we used Extreme Gradient Boosting regression models with Shapley Additive Explanations to quantify the relative contributions of predictors to engagement and information diffusion, incorporating a multidimensional feature space spanning user metadata, content characteristics, and temporal and contextual attributes. Results: Across 377,520 rumor debunking posts from 24,742 blue-verified accounts, 93,799 posts (24.8%) were health-related. Health debunking volume surged during the COVID-19 pandemic and remained elevated above prepandemic levels through 2024, although its relative share declined markedly in 2024, indicating a shift away from pandemic-centered topics, even as overall debunking activity continued to rise. Prevalent themes included vaccines, debunking reports, personal care, and sleep hygiene. Temporal trajectories fell into 3 patterns: event-driven spikes, sustained growth, and recurrent fluctuations. Demographic mentions were uneven: women were referenced more often than men, youth were the most frequently mentioned age group, and older adults were the least mentioned. Engagement was predicted primarily by source-level reach: follower count was the strongest predictor of reposts, comments, and likes alike, showing a nonlinear, threshold-like pattern in each case, whereas the secondary predictors differed across the 3 interaction types. Conclusions: This 15-year longitudinal analysis shows that health-related debunking on Sina Weibo was strongly shaped by major public health crises, surging during the COVID-19 pandemic and remaining elevated, while broadening toward routine, lifestyle-related concerns. Across all interaction types, engagement was shaped more by source reach than by the content of corrections. Although most posts addressed the general public, those with explicit demographic references revealed uneven representation across gender and age groups, with distinct health concerns linked to each. Taken together, these findings reveal a structural asymmetry in the debunking ecosystem: content is diversifying while distribution remains governed by source-level reach, suggesting that platform-level mechanisms that help credible but less-followed sources circulate corrections may be valuable for amplifying reliable health information.