This study analyzed publicly available video content from TikTok and Bilibili without collecting personal data or interacting with users. According to institutional and international guidelines, the study did not constitute human subjects research. Nonetheless, the protocol was reviewed and approved by the Institutional Review Board of Beijing Hospital of Traditional Chinese Medicine. The authors confirm compliance with TikTok's Terms of Use and Bilibili's platform policies.

To ensure the breadth and representativeness of the video sample, TikTok and Bilibili were selected as the primary data sources based on their large user bases and popularity in disseminating health-related content in China.

A new user account was registered on each platform to minimize the influence of personalized recommendation algorithms. The keyword “Osteoarthritis” (in Chinese: “骨关节炎”) was used to search for relevant videos. Searches were conducted in February 2025, without geographic restrictions, and included publicly available videos only. No filters were applied to sort the search results, thereby preserving the platforms' default ranking algorithms.

Although the study focused on the top-ranked search results to reflect real user exposure, the potential effect of algorithmic recommendation bias is acknowledged as a limitation.

Videos were eligible if they met the following criteria: (1) explicitly related to osteoarthritis health education; (2) publicly accessible without privacy restrictions; and (3) duration between 0 and 60 min. Exclusion criteria included: (1) videos primarily addressing other diseases or pediatric populations; (2) advertisements unrelated to OA; and (3) duplicate videos or content lacking substantive OA relevance. An initial pool of 200 videos (100 from TikTok and 100 from Bilibili) was retrieved. After screening based on the predefined criteria, 189 videos were included in the final analysis, comprising 96 videos from TikTok and 93 videos from Bilibili. All inclusion and exclusion decisions were independently cross-checked by two reviewers to ensure consistency and transparency.

For each included video, a standardized set of characteristics was recorded, including basic information (video duration, main topic, and uploader identity), uploader type (categorized as healthcare professionals, such as osteopaths and rehabilitation therapists, or non-professionals, such as patients and health bloggers), content type (classified into disease knowledge, treatment methods, or lifestyle interventions, allowing for multiple categorizations per video), and user engagement metrics (number of likes, comments, favorites, and shares).

To systematically evaluate video quality, four validated instruments were employed: the Journal of the American Medical Association (JAMA) benchmarks, the modified DISCERN (mDISCERN) scale, the Global Quality Score (GQS), and the Health On the Net Foundation's Code of Conduct (HONcode).

The Journal of the American Medical Association (JAMA) benchmarks assessed the reliability of information sources and disclosure of conflicts of interest, with scores ranging from 0 to 4, where higher scores indicated greater credibility (Table 1) (20). The modified DISCERN (mDISCERN) scale evaluated clarity, reliability, balance, citation of external sources, and acknowledgment of uncertainties, using a binary scoring system (0 or 1 per item, total score 0–5) (Table 2) (29). The Global Quality Score (GQS) provided an overall assessment of educational value, rated from 1 (poor) to 5 (excellent) (Table 3) (25). The Health On the Net Foundation Code of Conduct (HONcode) assessed adherence to eight ethical principles for online health information, with a cumulative score ranging from 0 to 16 (Table 4) (30).

Video evaluations were performed independently by four trained experts following a randomized allocation of videos. A computer-generated randomization sequence assigned 189 videos into two groups of approximately equal size. Each group was evaluated by two independent reviewers (Group I: Shuming Li and Ren Guo; Group II: Ping Chen and Chenyu Zhang), who were blinded to the video source and uploader identity. In cases of substantial discrepancy between raters, a third independent clinical expert adjudicated the scores, with the final rating determined by the median value of the three evaluations. Prior to assessment, all reviewers underwent calibration training based on standardized scoring guidelines to ensure consistency and minimize subjective bias.

Continuous variables were presented as mean ± standard deviation (SD) or median and interquartile range (IQR), depending on the distribution. Categorical variables were expressed as frequencies and percentages. The Shapiro–Wilk test was used to assess the normality of continuous variables. Independent sample t-tests or Mann–Whitney U tests were applied to compare video quality scores (JAMA, mDISCERN, GQS, and HONcode) between platforms and uploader types. Spearman's rank correlation analysis was conducted to evaluate the associations between video quality scores and user engagement metrics. A two-sided P-value < 0.05 was considered statistically significant. All analyses were performed using SPSS version 27.0 (IBM Corp, Armonk, NY, USA).All analyses were independently verified by two statisticians to ensure accuracy and reproducibility.

A total of 200 osteoarthritis-related videos were initially retrieved, with 100 from TikTok and 100 from Bilibili. After applying predefined inclusion and exclusion criteria, 189 videos were included in the final analysis, comprising 96 TikTok and 93 Bilibili videos (Figure 1). Among these, 6 videos (6.3%) from TikTok and 28 videos (30.1%) from Bilibili were uploaded by non-professionals.

Figure 2 summarizes the basic characteristics of the included videos. TikTok videos had a median duration of 75.5 s (IQR 49–108), compared to 152 s (IQR 72.5–332.5) on Bilibili (P < .001). The number of comments was significantly higher on Bilibili (median 128; IQR 15.5–424) than on TikTok (median 27; IQR 4–108.5; P = .002), as were favorites (median 697 vs. 179.5; P = .001). No statistically significant differences were observed in the number of likes (median 898 vs. 662.5; P = .351) or shares (median 297 vs. 148; P = .085) between the two platforms. The number of followers of each uploader was also recorded to provide additional context for engagement differences, although this variable was not included in inferential analyses.

Video quality was assessed using four established scoring tools (Figure 3). TikTok videos had significantly higher JAMA scores (median 2.4; IQR 2.2–2.6) than Bilibili videos (median 2.1; IQR 2.0–2.6; P = .001). GQS scores were also higher on TikTok (median 3; IQR 2–3) compared to Bilibili (median 3; IQR 2–3; P = .006). HONcode scores were higher for TikTok (median 11; IQR 9–11.75) than for Bilibili (mean 9.33 ± 2.59; P = .005). No statistically significant difference was found in mDISCERN scores (median 3; IQR 2–3 for both platforms; P = .196).

Spearman correlation analysis was conducted to examine the associations between user engagement metrics and video quality scores on TikTok and Bilibili. Results are presented in Figure 4.

On TikTok, all engagement metrics were significantly positively correlated with JAMA scores, with the highest correlation observed for likes (r = 0.732, P < .001), followed by comments (r = 0.704), favorites (r = 0.684), and shares (r = 0.617). mDISCERN scores were also positively correlated with likes (r = 0.611), comments (r = 0.592), favorites (r = 0.561), and shares (r = 0.590; all P < .001). For GQS, correlation coefficients ranged from r = 0.467 (shares) to r = 0.578 (likes), and for HONcode from r = 0.483 (shares) to r = 0.618 (likes). Video duration was not significantly correlated with any score (P > .05).

On Bilibili, likes were moderately correlated with JAMA (r = 0.470, P < .001), mDISCERN (r = 0.399), GQS (r = 0.390), and HONcode (r = 0.449). Similar correlations were found for favorites (e.g., r = 0.480 for GQS; r = 0.495 for HONcode) and shares (r = 0.414 for HONcode). Comments also showed positive but weaker correlations with most scores (r = 0.309–0.430). Unlike TikTok, video duration was significantly correlated with JAMA (r = 0.344, P = .001), mDISCERN (r = 0.382, P < .001), and HONcode (r = 0.266, P = .010), but not with GQS (P = .071).

Comparison by uploader type revealed that videos from healthcare professionals had significantly higher GQS scores (median 3; IQR 2–3) compared to those from non-professionals (median 2; IQR 2–3; P = .004; Figure 5). HONcode scores were also significantly higher for professional uploaders (median 10; IQR 8–11) than for non-professionals (mean 8.76 ± 2.82; P = .010). No significant differences were found in JAMA scores (median 2.3; IQR 2.0–2.6 vs. mean 2.16 ± 0.76; P = .067) or mDISCERN scores (median 3; IQR 2–3 for both groups; P = .473).Subgroup analysis by content theme (disease knowledge, treatment, lifestyle) suggested a similar trend, though differences were not statistically significant.

This study systematically evaluated the quality and reliability of OA-related health information on TikTok and Bilibili, revealing significant differences between the platforms. TikTok videos demonstrated higher quality across JAMA, GQS, and HONcode assessments, while mDISCERN scores showed no statistically significant difference. These findings suggest that, overall, TikTok may provide more credible and educational OA-related content than Bilibili. These results indicate an association rather than a causal relationship, given the cross-sectional naturse of the data.

Several factors may explain this disparity. First, platform algorithms likely influence the visibility of high-quality content (31). TikTok's algorithm is hypothesized to favor professionally produced, high-engagement videos, aligning with prior evidence showing that platforms such as YouTube prioritize trustworthy health information (25). In contrast, Bilibili's more decentralized and community-driven recommendation system may offer less curation, allowing content of variable quality to proliferate.

Second, differences in user demographics and video formats may shape content quality. TikTok's short-form, attention-driven design incentivizes concise, information-dense videos, while Bilibili's long-form culture may foster verbose or diluted messaging. Additionally, TikTok's recent moderation efforts targeting health misinformation may have elevated the baseline quality of content on the platform, while Bilibili continues to host a larger proportion of videos created by non-professionals, as reflected in our sample.

Importantly, while TikTok outperformed Bilibili in overall quality metrics, it remains susceptible to health misinformation (32). Prior research has demonstrated the presence of inaccurate or misleading videos even on well-regulated platforms. For example, in a study of sinusitis-related TikTok content, the majority of non-medical influencer videos contained inaccurate claims (33).This underscores the persistent need for quality assurance, regardless of platform. This highlights that algorithmic advantages do not necessarily eliminate misinformation risks, underscoring the need for ongoing monitoring and quality regulation.

Our results are consistent with previous studies demonstrating that healthcare professionals produce higher-quality content than non-professionals (34). Liang et al. similarly found that TikTok outperformed Bilibili in content reliability and educational value for videos about gastroesophageal reflux disease (20). This cross-condition consistency suggests structural platform advantages. However, such consistency should be interpreted as an observed pattern rather than a causal conclusion about platform mechanisms.

The observed superiority of professional content also echoes findings from YouTube-based studies, where professional authors achieved significantly higher GQS and mDISCERN scores than lay creators (25, 35).These results reinforce the importance of content source credibility in health communication.

Nonetheless, platform performance can vary by disease topic and cultural context. One study found that Chinese-language videos on Bilibili and TikTok outperformed English-language content on YouTube for gastric cancer, suggesting that public familiarity and topic sensitivity may influence content quality (24).

Regarding the relationship between content quality and user interaction, prior research has been mixed. Some studies have reported no significant correlation between engagement metrics and quality, while our findings demonstrate a strong positive association, particularly on TikTok. This may reflect algorithmic designs that amplify credible content, or user behavior that favors well-produced and informative videos (36, 37). Nevertheless, these correlations cannot confirm directional causality, as engagement may also drive algorithmic exposure or creator adaptation over time. In addition, We did not perform comment coding in this study, Further work may classify interaction types (e.g., medically relevant vs. casual comments) to assess “effective engagement” and its link to video quality.

This study underscores the critical role that short video platforms can play in patient education and chronic disease management. For individuals with osteoarthritis, access to accurate, high-quality online content can support informed decision-making, improve adherence to self-care practices, and potentially enhance long-term outcomes. The higher prevalence of reliable, professionally produced videos on TikTok suggests that certain platforms may offer more trustworthy health information environments than others (38). Nevertheless, these findings reflect associations observed within the current dataset and should not be generalized as an inherent superiority of one platform over another, since audience demographics and algorithmic mechanisms differ.

Given the chronic and self-managed nature of osteoarthritis, continuous access to concise and evidence-based content—particularly outside clinical settings—is essential. Videos produced by healthcare professionals may be especially effective due to their clarity, scientific accuracy, and ability to foster trust. Such content can serve as an extension of clinical care, reinforcing key messages and promoting sustained patient engagement (39, 40).

From a public health perspective, these findings suggest that social media platforms should be integrated into broader health communication strategies (41). To maximize public health benefits, platform operators should consider prioritizing evidence-based content in recommendation systems, introducing visible quality indicators (e.g., verification badges), and implementing mechanisms to flag or demote misleading material. Enhancing the discoverability of credible content—particularly from verified medical professionals—can improve the informational environment and reduce exposure to misinformation (42).

Furthermore, the observed association between video quality and user engagement supports the potential for high-quality content to achieve broader reach and impact. Public health agencies and healthcare institutions may benefit from partnering with digital creators or investing in content development to amplify accurate messaging at scale.

Finally, this study highlights the dual responsibility in digital health education: clinicians and public health professionals must actively engage in content dissemination while also equipping patients with critical appraisal skills to navigate an increasingly complex information landscape. Sustained improvement in online health information quality will thus require coordinated efforts among platforms, professionals, and users to balance accessibility with accuracy.