Professional Creators Provide Higher-Quality Ankylosing Spondylitis Content on TikTok and Bilibili: A Cross-Sectional Study

Abstract

Ankylosing Spondylitis (AS) is a complex autoimmune disease for which early diagnosis and treatment are critical for prognosis. The quality of health information shared on social media significantly impacts patient education. This study aims to evaluate the quality and completeness of AS-related health information on 2 major Chinese video platforms, TikTok and Bilibili. On August 27, 2025, a cross-sectional study was conducted, retrieving and analyzing the top 100 AS-related videos (N = 200) from each platform. Video reliability was assessed using the DISCERN tool and JAMA, and video quality was assessed using GQS and the 6-Dimensional Content Integrity Scale. Bilibili was found to be significantly superior to TikTok only in academic rigor (defined as information reliability and source transparency, measured by DISCERN-Dimension 1 and JAMA; P < .001), while both platforms performed poorly in practical utility (defined as content completeness and instructional value for patients, measured by GQS and DISCERN-Dimension 2; P > .05). Video “quality” was found to be completely uncorrelated with “popularity” (eg, likes, favorites) (P > .05), while video duration was the only significantly correlated external factor (rho = .41, P < .001). Content completeness exhibited a “cancelation effect” (Referring to the complementary nature of content strengths across platforms where one excels in areas the other lacks. Bilibili excelled in “diagnosis,” while TikTok excelled in “treatment”). Notably, orthopedists and other healthcare professionals (HCPs; 59.5%) were the main content creators, not rheumatologists (27%). On Bilibili, the “patient” group scored significantly higher in “treatment information” (Dim2) than the “rheumatologist” group (P = .029). Bilibili provides more academically rigorous AS information, but both platforms severely lack depth in core practical content, and platform algorithms fail to effectively screen for high-quality content (quality-popularity disconnect). The video ecosystem (eg, “orthopedist-dominated” and “professionalism inversion” [a phenomenon where patient-led narratives outperformed specialists in treatment-related metrics due to the inherent structural alignment between comprehensive personal storytelling and traditional assessment criteria].) profoundly reflects real-world diagnostic dilemmas and the limitations of traditional assessment tools in the modern video era.

Introduction

Ankylosing Spondylitis (AS) is a chronic inflammatory autoimmune disease primarily affecting the spine and sacroiliac joints, belonging to the spondyloarthritis (SpA) spectrum. ¹ The prevalence of AS is approximately 0.3%, affecting about 20 million people globally.^2,3 The prevalence of AS in China is 0.20% to 0.42%. ⁴ This figure is roughly comparable to the global figure. The core features of the disease are inflammatory low back pain, restricted spinal mobility, and inflammatory changes in the sacroiliac joints visible on imaging, which can lead to spinal ankylosis, kyphotic deformity, and functional disability in advanced stages. Diagnostic delay and high misdiagnosis rates for AS are major contributors to the severe consequences of the disease. It is estimated that the average diagnostic delay for AS is 6 to 7 years globally; in China, the probability of a patient’s first consultation being with a rheumatologist is only 25%.^5,6 This delay imposes a severe burden on patients and society, making correct AS health education an urgent priority.^7,8

With the development of mobile terminals, people are increasingly turning to online videos for information. In 2018, 9 out of 10 American adults used the World Wide Web, and 75% of them made medical searches with medical content. ⁹ According to the “Digital 2025 Global Statistical Report” by DataReportal (in partnership with We Are Social and Meltwater), the global average daily time spent watching online videos in 2025 is approximately 99 min. ¹⁰ Due to their accessibility and propensity for dissemination, videos have the potential to be a vital component of patient education.^11,12 Given that online videos often contain extensive written information in addition to verbal content, the concept of health literacy is crucial. To ensure broad accessibility, the National Institutes of Health (NIH), the U.S. Department of Health and Human Services (DHHS), and the American Medical Association (AMA) recommend that patient education materials be written at a sixth-grade reading level. If readability exceeds this threshold, it will be difficult for the average reader to understand, potentially hindering effective disease prevention and treatment compliance. ¹³ So, this ease of access is a double-edged sword. The variable quality of videos makes it difficult for patients to guarantee the accuracy and timeliness of the information they receive, a persistent challenge in the dissemination of health science videos. ¹⁴

In this study, we choose TikTok and Bilibili as the platforms for this study. First, both are the most widely used short video platforms in China and have been commonly used in previous health information content quality assessments, making the results comparable with the existing literature. ¹⁵ Secondly, global video platforms such as YouTube, Instagram, and Facebook are limited in assessing the relevance of Chinese health content due to reasons such as language preference. ¹⁶ Third, although other Chinese social media (eg, Kuaishou, Xiaohongshu) also provide health-related content, but their primary formats or user engagement patterns are different (eg, lifestyle posts or mixed media rather than systematic health videos), existing research suggests that studies using TikTok and Bilibili provide a clearer context for medical information dissemination in China. ¹⁷

TikTok and Bilibili are widely used video-sharing platforms in China, each with a distinct style. TikTok is known for its short-video format and algorithm, while Bilibili has a younger user base and no specific restrictions on video length.^18,19 Research suggests that platform-specific factors can influence the reliability and educational value of health information; therefore, a comparison between these 2 platforms is valuable in the analysis of health education videos.

Although the value of health education videos is gradually being recognized, the quality assessment of videos for specific diseases remains underexplored. ²⁰ Such explorations are essential for future health communication strategies and patient education initiatives. This study aims to fill this gap by conducting a comprehensive, comparative assessment of AS-related video quality on these platforms using standardized tools (DISCERN, GQS, JAMA, and a 6-dimension content completeness scale).

With the rise of social media, the dissemination of medical information has changed significantly. In addition to traditional peer-reviewed publications, the academic visibility of research is increasingly influenced by social media interactions. A bibliometric analysis of the top 100 most cited papers on AS revealed a moderate correlation between citation metrics and altmetric attention scores, suggesting that high-impact scientific findings in AS (eg, prevalence and biological therapies) also attract significant attention on social media. ²¹ Therefore, evaluating AS-related content on platforms such as TikTok and Bilibili is not only critical for patient education, but also reflects the general trend of medical literature promotion in the digital age.

Materials and Methods

Ethical Review

This study was based on publicly available data and did not involve any clinical data, human samples, or laboratory animals. Therefore, ethical review was not required. This cross-sectional study was conducted and reported in accordance with the STROBE statement. ²²

Video Retrieval Strategy

We retrieved videos using the native Chinese keyword for “Ankylosing Spondylitis” (Qiangzhi xing jizhu yan) on the Chinese versions of Bilibili and TikTok. Before starting a search, we logged out of all accounts, cleared our search history, and enabled incognito mode to reduce algorithm-driven personalized recommendations. Videos were sorted by the default comprehensive ranking, with no restrictions on upload date or author. A preliminary pool of the top 110 videos was retrieved from each platform (N = 220). These videos were then screened based on the pre-defined exclusion criteria. Specifically, a total of 10 videos from each platform were removed based on the exclusion criteria (including technical retrieval failures for 2 videos on Bilibili): (1) content irrelevant to the topic; (2) lack of substantive health education; (3) silent videos; or (4) technical extraction failures. Ultimately, a total of 100 videos from Bilibili and 100 videos from TikTok met all criteria and were included in the final analytical sample (Total N = 200). Figure 1 illustrates the specific search strategy.

Figure 1.

Video screening process.

Data Collection

We recorded basic video information (duration, likes, textual interactions, favorites, shares, upload date, presence of subtitles, presence of human subjects) and publisher information (account name, professional certification). We categorized publishers into 3 types: (1) Rheumatologists, (2) Orthopedists and other HCPs (Healthcare Professionals), and (3) Patients and other non-professionals.

When collecting user textual interactions, we adopted a platform-specific strategy. For TikTok, we recorded the “comment count” below the video. For Bilibili, we opted to record its unique “Danmaku” (live bullet-comment) count instead of the traditional comment section. Bilibili is renowned for its real-time Danmaku, which are synchronous with the video content and characterized by high immediacy and fragmentation. This format is functionally and behaviorally more analogous to the rapid, brief comment style on TikTok’s short-video platform. In contrast, Bilibili’s traditional comment section often features longer, asynchronous discussions, making it less comparable to TikTok’s comments.

Assessment Tools

We used the DISCERN instrument, GQS, and JAMA benchmarks to assess video quality, and a 6-dimension scale for content completeness.

The DISCERN instrument is an internationally recognized tool for assessing the quality of health information. ²³ It aims to help researchers, clinicians, and the public objectively judge the reliability of information and the quality of treatment information. DISCERN consists of 16 items, each scored on a 5-point scale (1 = “No” - does not meet the standard, 5 = “Yes” - fully meets the standard, 2-4 = partially met). These items are divided into 3 sections: Part 1: Reliability of the publication (Items 1-8), Part 2: Quality of treatment information (Items 9-15), and Part 3: Overall rating (Item 16). The total score is categorized into 5 levels: Very Poor (16-26), Poor (27-38), Fair (39-50), Good (51-62), and Excellent (63-80). DISCERN has been widely used to evaluate health information ²⁴ and to assess the quality of health education videos. ²⁵

The Global Quality Scale (GQS) is another concise tool commonly used for online health information assessment. ²⁶ It is often used alongside DISCERN, especially for evaluating online video quality. The GQS ranges from 1 to 5 and assesses the overall quality, comprehensibility, and utility of health information, particularly for online videos or webpages, covering dimensions such as accuracy, completeness, comprehensibility, usability, and structure.

The JAMA benchmarks were proposed by the Journal of the American Medical Association in 1997 to establish quality control standards for online medical content. ²⁷ They are particularly suitable for assessing the reliability and transparency of information sources in health content on websites, videos, and social media. As a concise, structured score (0-4 points), it evaluates content based on authorship, attribution, currency, and disclosure.

The assessment of video content utilized the JAMA benchmarks, the DISCERN instrument, and the Global Quality Scale (GQS). While these tools were originally developed in Western medical contexts, their core evaluative dimensions—transparency, evidence-based reliability, and educational quality—are universally applicable to health information regardless of geographic region. To ensure cultural and linguistic suitability for Chinese social media, the criteria were interpreted with consideration for the specific ecosystem of TikTok and Bilibili. For instance, in evaluating “Authorship” under the JAMA benchmarks, we accounted for platform-specific professional verifications (eg, “Bilibili-certified expert”) as an equivalent indicator of credentials. Furthermore, the DISCERN and GQS were used to maintain international comparability, allowing our findings on AS to be contextualized within the broader global literature. To ensure conceptual equivalence during the evaluation, all criteria were assessed by 2 bilingual rheumatology specialists, minimizing bias related to cultural or instrumental translation.

The combination of JAMA, DISCERN, and GQS provides a comprehensive assessment covering source transparency, clinical reliability, and educational utility, which are the 3 pillars of online health information quality.

Finally, we used a 6-dimension scale to assess content completeness, evaluating videos on: (1) Disease definition, (2) Symptoms, (3) Risk factors, (4) Evaluation and diagnosis, (5) Treatment and management, and (6) Prognosis. Each dimension was scored from 0 to 2 (0 = no content, 1 = partial content, 2 = comprehensive content).

Video Assessment

On August 27, 2025, 2 authors (LIU and WANG) independently assessed the videos. Both are medical postgraduates with specialized training in rheumatology. Before scoring, they thoroughly reviewed the criteria for the DISCERN, GQS, JAMA, and 6-dimension scales. Disagreements were resolved by a third reviewer (HUANG), a chief physician in the Department of Rheumatology and Immunology.

This study employed a cross-sectional design. All data collection and quality assessment were based on the videos available on the retrieval date (August 27, 2025). We did not conduct longitudinal tracking or reassessment, aiming to provide a cross-sectional snapshot of video quality at the time of data collection.

Statistical Analysis

Statistical analysis and visualization were performed using R statistical software (version 4.5.0) with RStudio. For comparisons between 2 groups (eg, Bilibili vs TikTok platform differences), the Welch 2 Sample t-test was used. For comparisons among 3 or more groups (eg, differences between publisher types), a one-way analysis of variance (ANOVA) was used, followed by a Tukey HSD post-hoc test if significant differences were found.

Spearman’s rank correlation analysis was used to assess the relationship between continuous variables (eg, video duration, like count) and quality scores. All statistical tests were 2-tailed, and a P < .05 was considered statistically significant.

Results

General Characteristics of Videos

We collected the top 110 videos from TikTok and Bilibili via default sorting. After screening, 200 videos (Bilibili N = 100, TikTok N = 100) were included. The screening process is detailed in Figure 1.

As of the data collection date (August 27, 2025), the temporal distribution of videos on the 2 platforms showed significant differences. Bilibili’s videos spanned a wide time frame, with days since upload ranging from 1 to 2879 days (median = 1010.5 days, IQR: 158.5-1424.5), exhibiting an “archive” characteristic. Its median video duration was 123.47 s (IQR: 70.79-238.67), significantly longer than TikTok’s. In terms of publisher composition, Bilibili’s videos were dominated by orthopedists and other HCPs (53%), followed by rheumatologists (26%). The vast majority of content (88%) was medical knowledge dissemination.

In contrast, TikTok’s content was highly concentrated in the near term, with all videos posted within 260 days (median = 63.0 days, IQR: 43.0-75.75), exhibiting a “snapshot” characteristic. The median video duration was 79.91 s (IQR: 51.20-134.59). TikTok also had a higher proportion of videos from orthopedists and other HCPs (66%), followed by rheumatologists (28%). Medical knowledge dissemination was still the largest category (83%).

The production quality of the videos was generally high. Of the 200 included videos, 96% contained subtitles and 92.5% featured human subjects. Videos lacking subtitles or human subjects were mostly concentrated among the patient and non-professional group. Other general characteristics are detailed in Table 1.

Table 1.

Basic Characteristics of Video.

Characteristic	Descriptive statistics
	Bilibili (n = 100)	TikTok (n = 100)
Video duration(n)
<60	20	33
60-300	62	63
>300	18	4
Video upload year(n)
2017-2020	11	N
2021	22	N
2022	20	N
2023	8	N
2024	13	1
2025	26	99
Publisher category(n)
Rheumatologist	26	28
Orthopedist & other HCPs	53	66
Patients & non-professionals	21	6
Production features (n)
Are there any characters in the video?
Yes	89	96
No	11	4
Are there subtitles in the video?
Yes	96	100
No	4	0
Video content type(n)
Medical knowledge dissemination	88	83
Treatment recommendations	6	10
Patient experience sharing	5	6
Rehabilitation & prevention	1	1
User engagement metrics(total)
Number of video likes	69 336	63 571
Textual Interactions	3499	13 286
Number of video collections	28 240	21 686
Video Retweets	15 223	19 409

Note. Due to differing platform architectures, this metric comprises “Comments” for TikTok and “Danmaku” (real-time bullet comments) for Bilibili, representing the primary mode of immediate textual feedback on each respective platform.

Content Completeness Analysis

To assess content completeness, we used the 6-dimension scale. The average scores for Bilibili on the 6 dimensions (Definition, Symptoms, Risk factors, Evaluation & Diagnosis, Treatment & Management, Prognosis) were 0.88, 0.93, 0.66, 0.84, 0.79, and 0.75, respectively. The corresponding scores for TikTok were 0.91, 1.00, 0.63, 0.53, 0.97, and 0.67. Although the overall completeness scores appeared “superficially convergent,” the radar plot (Figure 2) revealed a significant “cancelation effect” in specific dimensions: Bilibili scored higher in “Evaluation and Diagnosis” (0.84 vs 0.53), whereas TikTok performed better in “Treatment and Management” (0.97 vs 0.79). Furthermore, both platforms performed poorly in the “Risk factors” dimension (scores <0.70), reflecting a common content gap.

Figure 2.

Video integrity assessment.

Quality Assessment by Publisher Type

Based on DISCERN total score categorization, the quality distribution of Bilibili’s videos was: 3 Very Poor, 23 Poor, 37 Fair, 31 Good, and 6 Excellent. The platform’s overall mean DISCERN score was 45.56, GQS was 3.46, and JAMA was 2.09. However, subgroup analysis by author type revealed that rheumatologists had a mean DISCERN score of 47.00, orthopedists 42.75, and patients/others 50.86. ANOVA revealed a statistically significant difference between the patient group and the orthopedist group (P < .01), while no significant difference was found between rheumatologists and the other 2 groups.

To investigate the source of this discrepancy, we further analyzed the DISCERN sub-dimensions. Results showed that in the DISCERN Dimension 2 (Treatment Information Quality, Q9-Q15), the patient group achieved an excellent mean score of 20.7, which was statistically significant compared to the poor performance of rheumatologists (15.4) and orthopedists (15.8). Conversely, in Dimension 1 (Reliability, Q1-Q8), rheumatologists demonstrated superior performance with a score of 27.9, significantly higher than orthopedists (23.6, P < .05).

In comparison, TikTok’s video quality was generally lower, with no videos reaching the “Excellent” category (3 Very Poor, 32 Poor, 42 Fair, 23 Good). Its overall mean DISCERN score was 42.56, GQS 3.39, and JAMA 1.92. Notably, on the TikTok platform, no statistically significant differences (P > .05) were found among publisher types for DISCERN total, GQS, or JAMA scores. Furthermore, no significant differences were found among author types for GQS and JAMA scores on either platform.

Heatmap Analysis of Quality Metrics by Publisher Type

To visually assess publisher performance across specific DISCERN, GQS, and JAMA items, a detailed heatmap was generated (Figure 3). This revealed several key quality patterns across platforms and author types.

Figure 3.

Data hot spot map.

Universal Strengths: On both platforms, all publisher types achieved relatively high scores on DISCERN items 1 to 3 (concerning the clarity and relevance of aims). For example, on Bilibili, rheumatologists scored 4.70, 4.62, and 4.46 on these items, respectively, and even the lowest-scoring patient group remained above 3.7. This indicates that the basic informational framework of most videos was clear.

Universal Weaknesses: The most significant finding was the “quality trough” (lightest colored areas) present on both platforms. DISCERN items 4 and 5 (citing sources) and DISCERN 11 (discussing treatment risks) were common shortfalls for all publishers. Even the most specialized rheumatologists scored only 1.38 (Bilibili) and 1.32 (TikTok) on DISCERN 11, reflecting a systematic deficiency in information transparency and risk communication.

Publisher Differentiation: Differences emerged in items related to treatment depth (DISCERN 9-15). On TikTok, although rheumatologists’ scores appeared visually higher (deeper color), this trend did not reach statistical significance (P = .604, as shown in 3.3), indicating that the professional quality gap in treatment depth has been “flattened” in the short-video ecosystem. On Bilibili, the patient group showed surprisingly high scores (deeper color) on DISCERN items 10 to 12 (benefits, risks, and non-treatment consequences), aligning with the anomalous findings in 3.3 and likely related to the platform’s user profile and longer video durations.

Overall, the heatmap showed that rheumatologists led in most technical metrics, while patients excelled in specific “experiential” metrics. However, all creators require improvement in citation and risk disclosure. GQS and JAMA scores were not significantly impacted by author type.

Comparison of Overall Quality Scores Between Platforms

To compare the overall video quality of the 2 platforms, we assessed DISCERN, GQS, and JAMA scores (Figure 4). Welch’s t-test results revealed that Bilibili held a significant advantage in metrics measuring academic rigor and reliability: its mean DISCERN Dimension 1 (Reliability) score (25.34) was significantly higher than TikTok’s (22.64, P < .001), and its mean JAMA Total score (2.09) was also significantly higher than TikTok’s (1.92, P < .001).

Figure 4.

Comparison of total video quality scores between platforms.

However, in metrics measuring practical utility, the platforms performed equally. Bilibili and TikTok showed no statistical difference in DISCERN Dimension 2 (Treatment Information Quality; P = .93) or GQS (Global Quality Scale; P = .57).

Ultimately, driven primarily by its significant advantage in reliability metrics, Bilibili also achieved a marginally significant victory in the DISCERN Total score (45.56 vs 42.56, P = .03). Overall, the results indicate that Bilibili provides more academically rigorous information than TikTok, but both platforms lack sufficient high-quality content regarding core treatment depth and utility.

Correlation Between Video Duration and Quality

To investigate the influence of video duration on quality, we first conducted platform-level Spearman correlation analyses (Figure 5).On Bilibili, video duration showed a significant positive correlation with all quality total scores (DISCERN_Total: rho = .46, P < .001; GQS: rho = .28, P = .004; JAMA_Total: rho = .33, P < .001). However, when analyzing DISCERN sub-dimensions, this relationship disappeared: duration was not significantly correlated with Dim1 (Reliability, P = .20) or Dim2 (Treatment Info, P = .17).On TikTok, a similar platform-level positive correlation was observed (GQS: rho = .44, P < .001; DISCERN_Total: rho = .32, P < .001), but it was notably not correlated with JAMA_Total (P = .24).

Figure 5.

Correlation between video duration and quality metrics.

To clarify this complex non-linear relationship, we performed a deeper subgroup analysis (stratified by Platform + Publisher Type). Results showed that within all 6 author subgroups (eg, Bilibili-Patients, Bilibili-Rheumatologists), video duration was not significantly correlated with Dim1 or Dim2 (all P > .05). Similarly, no significant correlations were found in the GQS and JAMA subgroup analyses.

In summary, although platform-level data suggests that longer videos have higher quality, the in-depth subgroup analysis confirms that no simple linear relationship exists between video duration and quality metrics within any specific creator group.

Correlation Matrix of Video Features and Quality

To explore factors influencing video quality, we conducted a Spearman correlation analysis of video features (duration, engagement) and quality scores (Figure 6).

Figure 6.

Spearman correlation matrix heat map.

The analysis revealed a critical finding: video “quality” is completely disconnected from “popularity.” As shown, all 4 user engagement metrics (like_count, comment_count, collect, forward) showed no statistically significant correlation with any of the core quality metrics (DISCERN_Total, GQS, JAMA_Total; all P > .05). This indicates that, for AS, videos with high likes or comments are not necessarily of higher quality.

Conversely, video duration was the only external feature significantly correlated with quality. Duration showed a moderate positive correlation with DISCERN_Total (rho = .41, P < .001), GQS (rho = .35, P < .001), and JAMA_Total (rho = .28, P < .001).Furthermore, the assessment tools demonstrated good internal consistency (DISCERN_Total vs GQS: rho = .72, P < .001), and user engagement metrics were highly inter-correlated (rho values .69-.95).

Discussion

Summary of Principal Findings

In this study, we assessed AS-related health videos on Bilibili and TikTok using DISCERN, GQS, JAMA, and a 6-dimension completeness scale. The key findings are:

• ① Content Completeness: The platforms showed superficial convergence, masking a “cancelation effect” where Bilibili excelled in “Diagnosis” and TikTok in “Treatment.” Both platforms were deficient in “Risk Factors.”

• ② Platform Quality: Bilibili was superior only in “academic rigor” (JAMA, DISCERN-D1), while platforms were equally mediocre in “practical utility” (GQS, DISCERN-D2) (P > .05).

• ③ Quality versus Popularity: Video quality was completely uncorrelated with user engagement (likes, favorites) (P > .05).

• ④ Publisher Ecosystem: Orthopedists and other HCPs (59.5%), not rheumatologists (27%), are the main creators of AS content.

These findings diverge from other disease-specific video analyses, offering new perspectives on platform-specific features and reflecting the real-world predicament of rheumatology in China.

Discussion 1: “Professionalism Inversion” and the Limits of DISCERN in the Video Era

This study revealed a thought-provoking “professionalism inversion” on Bilibili: non-professional “patients” (mean 20.71) significantly outperformed “rheumatologists” (mean 15.38, P = .029) on the highly complex DISCERN Dimension 2 (Treatment Information Quality). We argue this anomaly is not accidental, nor does it imply patients are more knowledgeable, but rather stems from a systematic misalignment between the DISCERN tool and the modern video education ecosystem.

First, DISCERN was created in 1999 to evaluate paper-based health pamphlets. It therefore heavily favors “comprehensive” content that discusses all treatment options, risks, non-treatment consequences, and quality of life impacts (ie, Dim2). Second, this standard fundamentally conflicts with the modern doctor’s objective. In an average 80 s video, a doctor’s goal is to deliver high-density, reliable (as evidenced by high Dim1 scores) core knowledge, not to replicate a paper pamphlet. This leads to their failure in Dim2 due to “lack of comprehensiveness.”

The key “coincidence” is that the “patient narrative” (avg. 466 s) coincidentally aligns perfectly with the DISCERN Dim2 criteria. A structurally complete “patient story” (eg, “back pain -> medication -> side effects -> improvement”) happens to hit every scoring item in Dim2. Our subgroup correlation analysis (P > .05) confirmed the true role of duration: it is a “threshold variable,” not a linear driver. Doctors (80 s) are below the threshold, while patients (466 s) are above it. Once this threshold is crossed, it is the narrative content, not the duration itself, that determines the high score.

In summary, the “professionalism inversion” is a systematic bias arising from the mismatch between an outdated assessment tool and modern creator objectives. This suggests an urgent need to develop new assessment tools for the short-video era that balance “comprehensiveness” (valued by DISCERN) with “precision” (at which doctors excel).

Discussion 2: Online Ecosystem as a Reflection of Offline Dilemmas

Another key finding is that orthopedists (59.5%) have replaced rheumatologists (27%) in the AS health education ecological niche on both platforms. This phenomenon, not widely observed in other studies,^{28 -30} is significant. As an autoimmune disease, AS treatment should be managed by rheumatologists, especially in the early stages. ³¹ Orthopedic intervention is typically required only in late-stage disease with severe spinal ankylosis. ³²

In China, AS diagnosis faces high misdiagnosis rates (>50%; 2.1, 2.3), long diagnostic delays (7 years; 2.1), and a scarcity of rheumatology resources. While national health policy mandates the establishment of rheumatology departments in secondary hospitals and above, ³³ implementation is lacking in many grassroots hospitals due to uneven resource distribution and insufficient specialist training. This results in a large number of early AS patients with “low back pain” mistakenly seeking treatment from orthopedists (75% in some reports). ³⁴ Consequently, orthopedists have taken on a significant role in online education.

However, blaming this phenomenon solely on China’s specific system may be simplistic. Interestingly, in market-driven systems like the United States, rheumatology faces a similar “accessibility crisis.” ³⁵ This reveals a deeper, transnational mechanism: the intrinsic characteristics of the rheumatology discipline itself. Whether the healthcare system is “policy-driven” (like China) or “market-driven” (like the US), the nature of rheumatological diseases (chronic, long treatment cycles, low emergency rates, limited procedural billing) ³⁶ places the specialty in a weak position within the hospital’s overall profit structure.

Thus, despite differing healthcare systems, the developmental dilemmas of rheumatology show a high degree of similarity: low economic returns, scarcity of specialists, and limited departmental development. This study not only provides a snapshot of China’s diagnostic dilemma but also offers a novel perspective, corroborating the universal challenges faced by the rheumatology discipline due to its intrinsic characteristics.^37,38

Discussion 3: Platform Characteristics and Implications for Patient Education—“Homogeneous Superficiality”

Ankylosing spondylitis (AS), a chronic inflammatory rheumatic disease, exhibits a distinct age distribution, predominantly affecting young men. Given that nearly all young adults are active social media users, the dissemination effectiveness of AS health education videos is potentially higher compared to conditions affecting older populations. ³⁹ TikTok’s user base is notably young, with approximately 70% of users aged between 16 and 35 years. This high proportion of young viewers aligns closely with the peak age of AS onset, positioning TikTok as a critical channel for reaching potential young patients and delivering early-stage health education. ⁴⁰

In contrast, the user demographics of Bilibili demonstrate a stronger inclination toward systematic knowledge acquisition. While Bilibili serves a broad audience—including 30% of users in the 30 to 50 age bracket—its core community remains heavily composed of young individuals. ⁴¹ Compared to TikTok users, who often prioritize entertainment and brevity, Bilibili users are more inclined toward deep learning and the pursuit of comprehensive health information. As AS is a chronic condition necessitating long-term self-management, young patients are more likely to utilize Bilibili’s long-form video format to gain a thorough understanding of disease mechanisms and complex rehabilitation exercise protocols. ²⁸ This precise alignment between audience needs and platform attributes explains why Bilibili videos are typically longer and possess significantly greater depth than those on TikTok. Such a digital ecosystem not only satisfies the demand for high-quality science communication among young AS patients but also provides professional medical institutions with a more effective front for patient education.

Despite our efforts to clear search history, the 2 platform samples showed a fundamental temporal gap: Bilibili (median 1010 days) acts as a “historical archive,” while TikTok (median 63 days) functions as an “instant newsfeed.” This ecological difference largely explains the findings from section 3.5. Bilibili’s “archive” algorithm retains older, “PC-era” style videos that emphasize academic rigor, resulting in significantly higher JAMA and DISCERN-D1 scores (P < .001). In contrast, TikTok’s “newsfeed” algorithm rewards novelty and does not reward “academic rigor” (Duration vs JAMA, P = .24), favoring instead “practical utility” (Duration vs GQS, rho = .44). This leads to its failure in rigor-based metrics.

However, this difference in rigor masks a deeper similarity. As shown in 3.5, the platforms showed no statistical difference in GQS or DISCERN-Dim2 (utility/treatment). Although the DISCERN total scores were significantly different, the overall quality on both platforms was “Fair,” a finding consistent with other studies.^28,29 We observed that virtually all creators avoided complex “pathological mechanisms” and “medication regimens,” leading to a “homogeneous superficiality” in medical depth. Consequently, both platforms lack sufficient depth in core medical content.

Discussion 4: Implications of the Study

A critical finding of this study is the total disconnection between user popularity metrics (likes, comments, favorites) and medical information quality (DISCERN, GQS) (P > .05). This suggests that patients cannot rely on platform algorithms or “like counts” to judge reliability and must maintain critical thinking and cross-verify information to avoid being misled by popular but low-quality content.

For creators (especially doctors), our findings suggest a need to reconsider content goals. The “professionalism inversion” (patients > doctors on Dim2) highlights the mismatch between doctors’ objectives (high-density, precise knowledge) and traditional tools (which demand comprehensiveness). In the future, doctors might benefit from adopting “patient narrative” techniques, balancing their Dim1 rigor with more discussion of Dim2 (risks, quality of life).

Given the disconnect between quality and popularity, platform companies must assume greater social responsibility. Algorithms should not only reward engagement but also identify and promote content with high “academic rigor” (high Dim1 or JAMA scores). Platforms should collaborate with medical institutions to establish scientific content assessment systems, helping high-quality medical information gain visibility and truly “do good.”

Discussion 5: Comparison With Emerging AI-Supported Platforms

While our study focuses on short-video platforms, it is essential to acknowledge the rapid emergence of AI-supported tools in health information dissemination. Unlike TikTok and Bilibili, where professional certification of creators helps ensure that content is produced by qualified healthcare personnel, AI chatbots like ChatGPT, Gemini, and Perplexity offer different advantages and challenges. A recent study evaluating AI responses to Ankylosing Spondylitis (AS) queries found that while tools like Perplexity demonstrate high information quality and reliability, the readability of AI-generated content often exceeds the recommended sixth-grade level, potentially creating barriers for patients with limited health literacy.

In contrast to the interactive and human-centric nature of videos, which provide a bridge for patients to approach professional doctors, AI models possess vast self-learning capabilities that can surpass individual human memory and address the lag in medical knowledge updates. However, AI applications are still prone to factual errors and imprecise information during data entry. For AS patients—who often face significant diagnostic delays—relying on unverified AI advice may cause unnecessary anxiety or treatment non-compliance. While professional human creators in China face challenges in effective communication due to high patient volumes—reflecting the “Even Hercules had his troubles” dilemma—the lack of transparency and regulatory oversight in AI algorithms remains a substantial risk for public health. Therefore, short-videos currently remain an irreplaceable medium for reliable, human-verified patient education. ⁴² Our findings thus underscore that short videos serve as a crucial, human-centered bridge that supplements the analytical yet often unvetted nature of AI-generated advice.

Discussion 6: the Unique Landscape of the Chinese Health Care System

Within China’s multi-tiered health care system, “grassroots hospitals” refer to primary care institutions—such as township health centers, community health service centers, and county hospitals—that serve as the first point of contact for routine health needs.^{43 -45} Despite long-term reforms aimed at strengthening grassroots service capacity, the Chinese system remains largely hospital-centric.^46,47 Persistent inequities in resource distribution between large tertiary hospitals and primary care institutions often lead patients to bypass grassroots services in favor of higher-level hospitals, even for minor conditions. These systemic characteristics help explain why “grassroots hospitals” retain a distinct functional and perceptual identity and why patients increasingly turn to social media platforms like Bilibili and TikTok to seek high-quality information from experts typically based in top-tier institutions.

This structural imbalance is further reflected in the professional distribution of online health educators. The observed dominance of orthopedists (59.5%) over rheumatologists (27.0%) as online educators is a stark reflection of China’s real-world rheumatology resource scarcity and referral patterns. Currently, while China has over 200 million patients with rheumatic diseases, there are only approximately 12 189 specialists. At the primary care level, the gap is acute: 93.3% of primary hospitals lack independent rheumatology departments, and 36.8% do not have any rheumatology-certified physicians. Since Ankylosing Spondylitis (AS) typically presents as chronic back pain, the vast majority of patients instinctively seek initial care in orthopedic or spine surgery departments rather than rheumatology, contributing to a misdiagnosis rate exceeding 50% and an average diagnostic delay of over 7 years.

Furthermore, economic and institutional factors exacerbate this disparity. Orthopedic departments are often better resourced and more ubiquitous in “grassroots hospitals” due to their higher surgical throughput and established diagnostic infrastructure. In contrast, many primary institutions lack specific diagnostic tools for AS, such as HLA-B27 testing (66.5% lack this) or sacroiliac joint MRI (54.3% lack this). Consequently, orthopedists become the “de facto” first point of contact both in physical clinics and on social media platforms, as they capture the large volume of patients searching for “spinal pain” or “disk herniation”—which are often the initial (mis)interpretations of AS symptoms. ⁴⁸ This phenomenon highlights the urgent need to balance digital health resources by amplifying the voices of rheumatologists in online spaces.

Strengths of the Study

Despite its limitations, this study has several notable strengths. First, to our knowledge, this is the pioneer study to systematically evaluate the quality, reliability, and completeness of Ankylosing spondylitis (AS) content on major Chinese short-video platforms (TikTok and Bilibili). While previous research has focused on AS-related websites, our analysis addresses the critical gap in the rapidly evolving domain of mobile-based video education. Second, the comparative design between Bilibili and TikTok provides unique insights into how different platform algorithms and user cultures (eg, “historical archive” vs “instant snapshot”) influence health information delivery. Third, the use of multiple validated instruments (DISCERN, GQS, and JAMA benchmarks) ensures a multi-dimensional and rigorous assessment of content. Finally, the study offers practical clinical implications by identifying the dominance of specific creator roles, which can guide healthcare providers in optimizing online patient education strategies.

Limitations of the Study

This study has several limitations. First, temporal disparity: as discussed in 4.4, we compared Bilibili’s “historical archive” with TikTok’s “instant snapshot,” limiting the generalizability of the comparison. Second, metric comparability: We compared Bilibili’s “Danmaku” with TikTok’s “Comments.” While we argue for their functional similarity, inherent differences in information density and user behavior may introduce bias. Third, study design: This cross-sectional design provides only a “snapshot” and cannot assess long-term content evolution or real-world patient impact. Fourth, search term limitations: We used only “ Ankylosing Spondylitis, ” omitting “AS,” “ankylosing spondylitis,” or common aliases, potentially missing relevant videos.

The included videos cover a significant timespan (1 -2879 days). During this period, global and Chinese clinical knowledge of Ankylosing Spondylitis (AS) has undergone evolutionary refinement rather than radical shifts. While the 2022 ASAS-EULAR recommendations introduced newer therapeutic options such as IL-17A and JAK inhibitors compared to the 2016/2017 updates,^49,50 the core principles of AS management—including the emphasis on early diagnosis using the 2009 ASAS criteria, ⁵¹ the fundamental role of physical exercise, and the use of NSAIDs as first-line therapy—have remained remarkably consistent. In China, national guidelines have similarly maintained these core pillars while expanding biological agent accessibility. ⁵²

While JAMA and DISCERN are internationally recognized for health communication, their application in the context of Chinese short videos requires nuanced interpretation. These instruments were primarily developed for Western, text-based environments where citation practices and shared decision-making discussions are culturally normative. In contrast, Chinese medical communication—particularly on platforms like TikTok and Bilibili—is deeply influenced by a hospital-centric care landscape. On these platforms, content often emphasizes professional authority and practical guidance over explicit bibliographic attribution or detailed comparative treatment analysis.

Consequently, certain metrics—such as “Attribution” and “Comparative Discussion”—may systematically yield lower scores due to culturally-specific communication norms rather than inherent information inaccuracy. Furthermore, for Chinese patients, “Authorship” is often perceived through the lens of institutional prestige (eg, the tier of the doctor’s hospital) rather than individual credentials alone, a dimension that current Western metrics do not fully capture. We acknowledge that while structural indicators of disclosure may be cross-culturally transferable, items assessing the depth of citation or shared decision-making require caution. Future research should prioritize the development of culturally adapted assessment tools tailored to the high-context and mobile-first nature of Chinese medical dissemination.

The Role of Disease Knowledge in Treatment Adherence and Prognosis

Patient knowledge regarding disease etiology, pathophysiology, treatment options, and prevention strategies plays a critical role in treatment compliance and clinical outcomes. Patients who understand the mechanisms and long-term course of Ankylosing spondylitis (AS) are more likely to actively participate in shared decision-making, maintain medication compliance, and implement recommended lifestyle changes.

Emerging evidence further supports this association. For instance, the study Digital Guidance: Quality and Readability Analysis of Artificial Intelligence-Generated Spondyloarthropathy Texts emphasizes that accessible, high-quality medical information enhances patient comprehension, which may in turn promote disease awareness and engagement in care. This aligns with the broader health literacy framework, wherein improved informational clarity and reliability can indirectly contribute to better disease self-management and therapeutic compliance.

Therefore, the quality of short video health content is not just a matter of information accuracy, but may also have downstream effects on treatment compliance and patient empowerment in chronic inflammatory diseases such as AS. ⁵³

Conclusion

This study systematically compared AS-related health videos on Bilibili and TikTok. We found that Bilibili provides more academically rigorous information (JAMA, DISCERN-D1), but both platforms are equally mediocre in core practical utility (GQS, DISCERN-D2), lacking sufficient medical depth.

Critically, video quality is completely uncorrelated with “popularity” (likes, favorites) (P > .05), revealing a systematic failure of platform algorithms to screen for high-quality content, which poses a risk of misinformation.

This study also revealed 2 profound ecological phenomena. “Professionalism Inversion”: On Bilibili, a systematic mismatch between traditional assessment tools (DISCERN) and modern video objectives (short, high-density) caused “patient narratives” (long vlogs) to coincidentally align with scoring criteria, resulting in significantly higher treatment information scores than doctors. “Offline Dilemma Mapping”: The dominance of orthopedists (59.5%) over rheumatologists (27%) as online educators is a stark reflection of China’s real-world rheumatology resource scarcity and patient referral issues.

In conclusion, future research should focus on developing novel assessment tools suitable for the short-video era, and platforms must optimize algorithms to balance content “rigor” with “utility” to help high-quality medical information reach patients.