This study focused solely on data obtained from publicly accessible short-video sharing platforms, with no involvement of human experimental research, thereby obviating the need for ethical approval. Furthermore, no identifiable information regarding individual users or their IDs was included in this study, ensuring privacy and confidentiality.

In designing our search and data collection strategy, we referred to established methodological guidelines for conducting health research using data from social media platforms, which provide a step-by-step framework for healthcare professionals and researchers to follow (30). A search was performed on TikTok, WeChat, and Xiaohongshu from March 1 to March 6, 2025, using the keyword “骨关节炎” (osteoarthritis). To minimize bias from personalized recommendation algorithms, searches were performed while logged out and using newly created accounts. The search results from each platform were sorted according to the default ranking algorithms, without applying any preset filters (e.g., relevance, most viewed). This approach ensured that the collected videos adhered to the platform's standard sorting criteria. We selected the top 100 videos for each search term to ensure a representative sample of the most relevant content (31). Inclusion criteria were as follows: (1) videos related to OA, and (2) videos in Chinese. Exclusion criteria were: (1) duplicate content, (2) advertisements, and (3) videos containing irrelevant content. For the selected videos, characteristics such as titles, number of likes, comments, collections, shares, video duration, sources, type of information, and content were recorded and analyzed.

The primary outcomes of this study were the reliability and quality of OA-related videos, evaluated using the DISCERN scale, JAMA benchmarks, and the GQS. These outcomes aimed to assess the accuracy and overall quality of the health information presented in the videos. The secondary outcomes included engagement metrics (likes, comments, shares, collections). The understandability of the videos was also assessed using the Patient Education Materials Assessment Tool (PEMAT).

In addition to the primary and secondary outcomes, we conducted exploratory analyses to examine differences in video quality based on the content creator type. Specifically, we compared the DISCERN, JAMA, and GQS scores for videos created by healthcare professionals (e.g., physicians, rehabilitation practitioners) and independent users. Furthermore, we examined how the content type influenced engagement metrics (likes, comments, shares, collections). This study also analyzes the impact of platform algorithms on content visibility and recommendation quality through correlation analysis. Specifically, it explores how algorithm-driven factors, such as user engagement (likes, comments, shares), correlate with video quality scores (e.g., DISCERN, JAMA, GQS).

Descriptive statistics were used to analyze all video features, including video source, content, audio, and message type (i.e., DISCERN, JAMA, and GQS). Categorical variables were presented as absolute frequencies and percentages, while continuous variables were reported as mean ± standard deviation, or alternatively as median, interquartile range (IQR), and range. Normally distributed data were presented as mean ± SD, while non-normally distributed data were reported as median and range. Comparisons between two groups were conducted using the Mann–Whitney U test or the Student's t-test, while comparisons among three or more groups were performed using the Kruskal–Wallis test or one-way ANOVA. Spearman correlation analysis was used to assess the relationship between quantitative variables. P value less than 0.05 (P < 0.05) was considered statistically significant. All statistical analyses were conducted using SPSS (version 22.0; IBM Corp) and GraphPad Prism (version 9.0; Dotmatics).

As shown in Figure 1, TikTok, WeChat, and Xiaohongshu each contained 100 videos after excluding advertisements, duplicates, and irrelevant content. Analysis of the general characteristics of the videos revealed no significant difference in video duration across the three platforms (p = 0.312). Videos on TikTok received more likes, comments, shares, and favorites than those from WeChat and Xiaohongshu (both p < 0.001). DISCERN ratings showed that TikTok videos scored higher than Xiaohongshu, which in turn scored higher than WeChat (mean 32.42, SD 0.37 vs. mean 30.21, SD 0.10 vs. mean 24.57, SD 0.34; P < 0.001); The JAMA score for TikTok videos was higher than that for Xiaohongshu, which was higher than that for WeChat (mean 1.36, SD 0.07 vs. mean 1.21, SD 0.05 vs. mean 0.76, SD 0.06; P < 0.001); The GQS score for TikTok videos was higher than for WeChat, which was higher than for Xiaohongshu (mean 2.46, SD 0.08 vs. mean 2.16, SD 0.06 vs. mean 1.90, SD 0.04; P < 0.001). Statistical differences were observed in the understandability and actionability of videos across different platforms (P < 0.001; Table 1). The detailed characteristics of OA-related videos across platforms are presented in Table 1.

Table 2 presents the sources and content of OA-related videos. Doctors were the primary uploaders of videos (255/300, 85.0%), and the content primarily focused on physical therapy (184/300, 61.3%) and etiology (41/300, 13.6%). The video content predominantly focused on educational science (215/300, 71.7%) and rehabilitation guidance (28/300, 9.3%).

When examining differences in information publishers, we found that videos uploaded by independent users had significantly lower DISCERN, JAMA, and GQS scores compared to those uploaded by physicians and rehabilitation practitioners, with significant differences observed (all P < 0.001; Table 3; Figures 2A–C). Similarly, the understandability and actionability of videos uploaded by private users were significantly lower than those uploaded by rehabilitators and physicians (P < 0.05, P < 0.001; Table 3; Figures 2D,E). When examining differences in the type of information, our study found no significant difference in likes across types (p = 0.053). However, regarding the number of discussions, clinical examination outperformed physical therapy (mean 91.12, SD 156.53 vs. mean 61.55, SD 216.80; P < 0.05), and anatomy outperformed etiopathogenesis (mean 526.65, SD 1438.78 vs. mean 120.40, SD 565.86; P < 0.05). Anatomy outperforms physical therapy in terms of shares (mean 2,720.85, SD 8,060.86 vs. mean 1,081.93, SD 7,717.04; P < 0.01). Patient experience ranked lower than anatomy in terms of number of collections (mean 284.08, SD 899.09 vs. mean 1,689.88, SD 3,875.34; P < 0.05) (Figures 3A–D).

Spearman correlation (ρ) analyses identified the relationships between various video variables. The results showed a positive correlation between video duration and video sharing (P < 0.05), but no correlation between duration and the number of likes, comments, or collections (all P > 0.05). And there is a significant positive correlation between the number of likes and the number of comments, shares and collections (P < 0.001; Table 4). The number of likes and comments were positively correlated with both the DISCERN and GQS scores (P < 0.001). The number of shares was positively correlated with the GQS score (P < 0.001). The number of collections was positively correlated with both the DISCERN and GQS scores (P < 0.01, P < 0.001; Table 5).

Correlation analysis revealed a significant relationship between interaction metrics and video quality scores. Specifically, videos with higher interaction rates are generally associated with higher quality scores (Table 5). On TikTok, videos with high interaction rates and quality scores are significantly more prevalent than on other platforms (P < 0.001; Table 1), possibly due to TikTok's recommendation algorithm. TikTok's algorithm tends to prioritize videos with high interaction rates, which generally receive higher quality scores. After viewing high-quality videos, viewer interaction frequency tends to increase, creating a positive feedback loop. This finding further supports the potential influence of platform recommendation algorithms on video visibility and the quality of recommended content.

The evaluation using the DISCERN, JAMA, and GQS revealed that OA-related videos on TikTok significantly outperformed those on Xiaohongshu and WeChat across multiple indicators. Specifically, TikTok videos were not only more credible but also more comprehensive and practical. This finding aligns with previous studies and confirms TikTok's leading role in health science popularization (33–35). This difference also highlights the potential and significance of TikTok in public health education, offering insight into how communication media directly influence the quality of health information.

The DISCERN and JAMA assessments revealed that videos created by healthcare professionals (e.g., physicians, rehabilitation therapists) received higher ratings compared to those produced by independent users. Videos created by non-professionals typically exhibited two major shortcomings: unclear presentation of specialized knowledge and a lack of practical instructional value, often aimed more at garnering attention than providing viewer benefit. These findings are consistent with previous studies (36, 37). These results emphasize that professional qualifications are crucial for ensuring the accuracy and credibility of health information, and that increasing the involvement of medical professionals in health education is essential for disseminating both professional and practical health knowledge to the public.

This study also found that content, subject matter, and presentation style significantly influenced user engagement (likes, shares, comments). Videos providing detailed explanations of bone and joint anatomy and clinical symptoms received more engagement than those focused on rehabilitation. Viewers preferred scientific content that was information-dense, detailed, and practically applicable, offering immediate utility. These findings are consistent with previous studies (38–40).

The results of this study demonstrate a high degree of consistency with previous research on social media health education. D’Ambrosi et al. found that, although physical therapists garnered significant attention on TikTok, the quality of their content varied (41). Similarly, several studies have emphasized the potential of short-form video platforms in advancing public health education (42, 43); however, the issue of fluctuating content quality persists, failing to persuade all viewers. These findings highlight a central paradox: when health discussions are dominated by laypeople, ensuring the accuracy and reliability of information becomes challenging. Our results are consistent with studies by Ozsoy (39) and Uprak (37) on YouTube health videos, which demonstrated that content created by healthcare professionals consistently outperformed that created by laypeople in terms of reliability and quality (37, 39). This finding achieves consensus across platforms and geographies, as an increasing number of researchers focus on online health issues. The deep involvement of healthcare professionals is indispensable for ensuring that video content is both credible and beneficial.

Although recommendation algorithms significantly influence the dissemination of health information on short-video platforms, there has been limited research on the impact of platform-specific recommendation systems on the visibility and quality of health-related videos, particularly those concerning OA. This study is the first to examine how the recommendation algorithms of TikTok, WeChat, and Xiaohongshu influence the display, visibility, and dissemination effectiveness of video content. Using TikTok as an example, its algorithm primarily relies on user interaction metrics such as likes, comments, and shares to determine video recommendations (44–46). The results of this study suggest that videos with higher interaction metrics are generally associated with higher video quality. This positive feedback loop leads TikTok's algorithm to prioritize higher-quality OA-related videos, which subsequently generate more interaction. This interaction-based recommendation mechanism may cause certain health content to gain higher visibility, while content with fewer interactions may be overlooked, despite potentially possessing significant educational value. Similarly, WeChat and Xiaohongshu have their own recommendation algorithms, although their priorities may not rely solely on interaction data but also include content relevance and user behavior. Due to the differing recommendation mechanisms across these platforms, health videos about OA may display varying levels of visibility and dissemination patterns, thereby affecting the accuracy and breadth of health information accessible to users. This disparity highlights potential biases inherent in recommendation algorithms in health information dissemination. We believe that while platform recommendation algorithms can increase video exposure, they may inadvertently enhance the visibility of certain content, causing misleading videos to attract more audience attention. Consequently, algorithmic bias may have detrimental effects on the health education functions of these platforms (47, 48).

This study relied on manual scoring and conventional evaluation tools. Future research could leverage artificial intelligence (AI) and machine learning (ML) techniques to automate the classification and analysis of health-related videos (49). For instance, AI could be employed for content categorization, misinformation detection, and sentiment analysis, enabling large-scale analysis of videos and their potential impact on public health. Additionally, natural language processing (NLP) could be applied to analyze viewer comments and interactions, offering deeper insights into audience engagement, trust, and behavior (50).

This study focused on a limited set of videos from Chinese platforms. Future research could broaden the dataset to include videos from a wider range of platforms, including YouTube, Instagram, and Facebook, for a comparison of content across different platforms and regions. This would provide a more comprehensive evaluation of social media's role in global health education. Furthermore, including multilingual videos from different cultural contexts would enhance the generalizability of the findings and provide insights into the cross-cultural applicability of health communication strategies.

Further research could explore how platform algorithms influence the dissemination of health information across diverse cultures. This could involve studying the differences in algorithmic impact between Western and Eastern platforms and the role of algorithms in shaping health education across culturally diverse contexts.

In summary, our study emphasizes the need for quality assessment of social media content, particularly on short-form video platforms. We identified significant disparities in the quality of OA-related videos across platforms and creators, underscoring the need for stricter content regulation and greater involvement of healthcare professionals in content creation. From a practical standpoint, enhancing the quality of health science communication on social media is essential. Short-form video platforms can play a pivotal role in chronic disease management by optimizing content presentation and advancing public health education, ultimately empowering individuals to take charge of their health.