Quality and reliability of sarcopenia-related short videos on Douyin

Abstract

Sarcopenia is a geriatric syndrome marked by the progressive decline in skeletal muscle mass, strength, and physical function. It typically develops insidiously and advances gradually, significantly increasing the risk of adverse outcomes such as falls, frailty, and mortality among older adults. With the rise of social media platforms like Douyin, the Chinese domestic counterpart of TikTok, for public health messaging, it is crucial to systematically evaluate the quality of content on this platform related to sarcopenia. In this cross-sectional content analysis, we collected Douyin videos related to sarcopenia using the platform’s default ranking algorithm and, after applying inclusion and exclusion criteria, obtained 156 videos. For each video, we recorded user interaction metrics (likes, comments, shares, collections) and extracted associated metadata. We evaluated content quality with the Global Quality Scale (GQS), the modified DISCERN (mDISCERN), and the JAMA benchmark criteria. We conducted Spearman’s correlation analyses to examine correlations among video variables, GQS, mDISCERN, and the JAMA score. The majority of the content originated from medical professionals (43.6%), with a primary focus on disease overviews (34.6%). Video quality assessments indicated low overall standards: GQS scores frequently reached 3 (44.9%), mDISCERN scores predominantly reached 2 (42.9%), and JAMA scores were concentrated at 2 (55.1%). Videos created by general users scored significantly lower than those from other sources (all P < .05). Furthermore, video duration was positively correlated with GQS (r = 0.470, P < .001) and mDISCERN scores (r = 0.363, P < .001), but not with JAMA scores (r = 0.166, P = .691). In contrast, no significant correlations emerged between quality and interaction metrics. While Douyin has the potential to raise awareness about sarcopenia, the prevalence of subpar content highlights the need for stronger oversight, credential verification, and better algorithmic prioritization to improve the quality of health information on the platform.

1. Introduction

Sarcopenia is a geriatric syndrome characterized by the progressive deterioration of skeletal muscle mass, strength, and physical function.^[1] As the global population ages,^[2] health challenges affecting older individuals are garnering greater societal attention. Despite differences in diagnostic standards leading to varying prevalence rates, research indicates that the worldwide occurrence of sarcopenia among those aged 60 years and older lies between approximately 10% and 27%.^[3] This condition develops insidiously and is associated with several adverse health outcomes, including an elevated risk of falls, functional impairment, diminished quality of life, and increased mortality among patients,^[4,5] thereby exacerbating the disease burden. Sarcopenia poses a significant barrier to daily activities among older adults,^[6] placing substantial economic pressure on healthcare systems.^[7] Modifying the behaviors of at-risk individuals through health literacy initiatives is essential for the prevention and management of sarcopenia.^[8,9]

In today’s era of advanced social networks, a wide range of health-related content is widely disseminated across social media platforms, with a growing number of individuals turning to the Internet for medical information.^[10] Emerging internet technologies offer innovative opportunities to enhance the effectiveness of health communication and improve patient education standards. Douyin, the Chinese domestic counterpart of TikTok, is one of the largest short-video platforms in China. It enables users to create and share short videos on health topics^[11] and is increasingly becoming a primary source of routine health information for the public.^[12,13] However, research on the dissemination of health information on this emerging short-form video platform remains relatively scarce.^[14] Furthermore, existing studies examining TikTok content related to hypertension, breast cancer, diabetes, and other conditions have revealed substantial variability in the educational quality and accuracy of health-related videos.^[15-17] Some videos even propagate inaccurate or unverified health information,^[18,19] potentially misleading viewers’ understanding and adversely affecting their health-related decision-making.

Despite the increasing popularity of Douyin as a source of health information, there is a gap in targeted research on content related to sarcopenia. This study aims to systematically analyze the characteristics and informational validity of Douyin videos concerning sarcopenia. The anticipated outcomes will enhance public understanding of this condition and provide actionable recommendations to improve the quality and communicative effectiveness of health-related video content on these platforms.

2. Materials and methods

2.1. Search strategy

This cross-sectional content analysis was conducted on February 17, 2026. A systematic retrieval of Douyin content related to sarcopenia was conducted. Three keywords were used: “肌少症” (sarcopenia), “肌肉萎缩” (muscular atrophy), and “少肌症” (an alternative term for sarcopenia). Douyin’s search results provide 3 sorting options: “Comprehensive,” “Most Recent,” and “Most Liked.” Given that “Comprehensive” is the platform’s default setting and represents the typical sequence in which users encounter content while browsing, this study employed this sorting method for data retrieval without imposing any restrictions on upload dates. To minimize personalization bias, user accounts were logged out and search histories cleared before each search.

The initial search yielded 765 videos (286, 351, and 128, respectively). For each keyword, the top 100 videos from the ranking results were selected to constitute the analysis sample. The decision to limit the sample size to 100 was based on the following considerations: First, platform ranking algorithms typically prioritize highly relevant content at the top of the results.^[16,20] Beyond this threshold, video content exhibits extremely low thematic relevance,^[21] and most users tend to browse only the first few pages of search results,^[22,23] resulting in significantly reduced exposure for subsequent videos. Second, existing research indicates that videos ranked below this threshold contribute minimally to overall content analysis and have a negligible impact on the validity of research conclusions.^[24-27]

2.2. Video collection and data acquisition

Videos were included if they provided substantive information about sarcopenia, such as its definition, clinical symptoms, diagnostic criteria, management strategies, or preventive measures, and were publicly accessible. Of the 300 candidate videos, we identified 12 duplicates across the 3 keyword-specific searches and removed them before applying the remaining exclusion criteria. We then screened the 288 unique videos to exclude non-Chinese videos, advertisements, videos unrelated to sarcopenia, and those lacking educational information. Detailed video screening criteria are provided in Table S1, Supplemental Digital Content 1.

For each selected video, the following attributes were recorded: title, like count, comment count, share count, collection count, days since upload, duration, video content, and uploader type. All gathered data were organized and managed in an Excel workbook. This study did not involve human participants, animal subjects, or field research requiring ethical approval. Therefore, ethics committee approval was not required.

2.3. Video categorization

The videos were categorized into 5 groups by source and into 6 groups by content. Source classification included medical personnel, science communicators (people who are dedicated to disseminating disease-related knowledge), general users, news agencies (e.g., online media outlets, television stations, and newspapers), and organizations (such as hospitals, public health agencies, research groups, and higher education establishments). Video content was categorized as follows: disease overview, disease symptoms, disease diagnosis, disease hazards, disease prevention, and lifestyle. Each video was assigned to 1 mutually exclusive primary content category based on its predominant purpose and content. Detailed operational definitions are provided in Table S2, Supplemental Digital Content 2.

2.4. Quality assessment

Video quality was independently evaluated and coded by 2 researchers (TZ and JW) using 3 validated assessment tools: the Global Quality Scale (GQS), the modified DISCERN instrument (mDISCERN), and the Journal of the American Medical Association (JAMA) benchmark criteria. Both evaluators are actively engaged in sarcopenia-related research and have completed clinical rotations in geriatric medicine, ensuring their subject-specific expertise. Inter-rater reliability was assessed using linearly weighted Cohen’s kappa (κ_w), calculated from the total GQS, mDISCERN, and JAMA scores independently assigned by the 2 initial raters. Interpretations adhered to established thresholds defined as follows: poor (κ_w ≤ 0.40), moderate (κ_w = 0.41–0.60), substantial (κ_w = 0.61–0.80), or excellent (κ_w > 0.80).^[28]

The inter-rater reliability analysis revealed excellent agreement for both the GQS (κ_w = 0.83, 95% CI [0.77–0.90]) and the JAMA benchmark (κ_w = 0.82, 95% CI [0.75–0.89]). Substantial agreement was observed for the mDISCERN instrument (κ_w = 0.79, 95% CI [0.73–0.86]). Before formal assessment, standardized training was conducted to establish consensus on the interpretation of scoring criteria across all instruments. Discrepancies were resolved through consensus discussion, with adjudication by a third senior researcher (JG) when agreement could not be reached.

The GQS uses a 5-point scale to evaluate overall video quality, with scores ranging from 1 (poor) to 5 (excellent).^[29] Additionally, the modified DISCERN instrument – an adaptation of the original DISCERN tool – evaluates 5 domains: transparency, significance, traceability, rigor, and equity. Each item is scored dichotomously (1 for “yes,” 0 for “no”), with a higher cumulative score indicating greater reliability.^[24] The JAMA benchmark criteria comprise 4 items that assess the reliability of video sources, awarding 1 point for each criterion met.^[30] Consequently, total scores can vary from 0 (poor quality) to 4 (high quality). Detailed scoring criteria for these evaluation systems are provided in Supplementary Tables S3–S5, Supplemental Digital Content 3.

2.5. Statistical analysis

Statistical analyses were conducted using IBM SPSS Statistics for Windows (version 27.0), and GraphPad Prism (version 9.0.0; GraphPad Software, LLC, Boston) was employed for figure generation. The Kruskal–Wallis test was used to evaluate differences among groups, and the Dunn multiple-comparison test was applied for pairwise comparisons of quantitative variables that did not meet the assumptions of normality. We used Spearman’s rank correlation analysis to assess relationships among variables, adjusting P-values with the Bonferroni method to account for multiple comparisons. Continuous variables with non-normal distributions are presented as medians with interquartile ranges (IQRs), whereas categorical variables are reported as frequencies and percentages. A P-value of <.05 was deemed statistically significant.

3. Results

3.1. General characteristics of videos

The initial searches retrieved 765 videos across the 3 predefined keywords. From these results, the top 100 videos ranked under the platform’s default “Comprehensive” sorting option for each keyword were selected, yielding 300 top-ranked candidate videos for screening. After excluding 130 videos that were unrelated to sarcopenia or lacked educational information, 12 duplicate videos, and 2 advertisements, 156 videos were included in the final analysis (Fig. 1). The general characteristics of the included Douyin videos are presented in Table 1.

Figure 1.

Flowchart of video retrieval and selection.

Table 1.

Detailed characteristics of the sarcopenia videos on Douyin.

Parameters	Douyin (N = 156)
Days since upload (d)	466.5 (198–1251)
Duration (s)	83.5 (54.5–157)
Likes	54.5 (22–268)
Comments	4.00 (1.5–11.5)
Collections	13.5 (3–90.5)
Shares	12.5 (3–88.5)
GQS score, n (%)
1	7 (4.5)
2	27 (17.3)
3	70 (44.9)
4	47 (30.1)
5	5 (3.2)
mDISCERN score, n (%)
1	27 (17.3)
2	67 (42.9)
3	40 (25.6)
4	16 (10.3)
5	6 (3.8)
JAMA score, n (%)
1	28 (17.9)
2	86 (55.1)
3	39 (25.0)
4	3 (1.9)

Values are presented as median (interquartile range) unless otherwise indicated.

GQS = Global Quality Scale, JAMA = Journal of the American Medical Association, mDISCERN = modified DISCERN.

These videos collectively accumulated 78,848 likes, 3069 comments, 28,873 shares, and 33,151 collections. Analysis of publication timing revealed that the earliest video was posted on August 26, 2019, 2367 days before the data collection date. In contrast, the most recent video was published on September 26, 2025, only 144 days before the data collection date. The median time since upload was 466.5 days. In terms of video duration, the shortest video lasted 8 seconds, the longest 1192 seconds, and the median 83.5 seconds.

Depending on the uploader’s origin, the included videos were categorized into 5 groups. Videos from medical personnel made up the largest share (n = 68, 43.6%), followed by those from general users (n = 34, 21.8%), organizations (n = 24, 15.4%), and news agencies (n = 16, 10.3%). Science communicators accounted for the smallest portion of the uploaded content (n = 14, 9.0%). The videos were classified into 6 categories based on thematic content analysis. Disease overview was the most common category (n = 54, 34.6%), followed by lifestyle (n = 40, 25.6%), disease diagnosis (n = 26, 16.7%), disease prevention (n = 25, 16.0%), disease symptoms (n = 6, 3.8%), and disease hazards (n = 5, 3.2%; Fig. 2).

Figure 2.

Distribution of videos by source and content category.

3.2. Video quality assessment

Our analysis using the GQS, mDISCERN, and JAMA benchmarks indicates that the quality of sarcopenia-related content on Douyin is generally inadequate. The median (IQR) scores were 3 (IQR: 3–4) for GQS, 2 (IQR: 2–3) for mDISCERN, and 2 (IQR: 2–3) for JAMA. The majority of videos scored at a fair level on the GQS (70/156, 44.9%), while mDISCERN and JAMA scores primarily clustered around 2 points (67/156, 42.9%; 86/156, 55.1%).

The Kruskal–Wallis test indicated significant differences in video quality across sources for the GQS (P < .001), mDISCERN (P = .002), and JAMA scores (P < .001; Table 2). To identify specific differences among source groups, we performed Dunn’s post hoc tests with Bonferroni-corrected P-values, revealing distinct intergroup patterns (Fig. 3). Videos created by general users received significantly lower GQS scores compared to those produced by medical personnel (P = .001), science communicators (P = .013), organizations (P < .001), and news agencies (P = .001). No statistically significant differences were found among these latter 4 groups. Similarly, mDISCERN scores were significantly lower for general users than for medical personnel (P = .008) and organizations (P = .003). JAMA analysis demonstrated that general users scored lower than all other groups, with no significant differences among the remaining categories. In summary, post hoc analyses indicate that content uploaded by general users consistently exhibits lower information quality across all 3 assessment tools compared with content from medical personnel, science communicators, organizations, and news agencies. These groups displayed relatively stable quality levels, with no significant differences among them.

Table 2.

Comparison of video sources based on features on Douyin.

Video features	Medical personnel	Science communicators	General users	News agencies	Organizations	P-value*
Days since upload	298.5 (177.5–1130.5)	664 (308–1073)	242 (189–1300)	806 (561–1429)	710 (361.5–1142)	.006
Duration	78.5 (48.5–107)	137 (76–233)	77 (39–165)	94 (57–196)	94 (69.5–150.5)	.111
Likes	63 (29.5–333)	31.5 (16–279)	45.5 (6–365)	94.5 (11.5–250.5)	37.5 (25.5–124.5)	.570
Collections	10.5 (3–53)	15.5 (10–133)	12 (1–298)	23.5 (2.5–117)	16 (9.5–36)	.910
Comments	5 (2.5–16)	4 (3–13)	2.5 (0–12)	2 (0–7)	2 (1–4)	.004
Shares	8.5 (1.5–57)	18.5 (6–178)	9 (3–145)	12 (3.5–125.5)	21 (11–49.5)	.486
GQS score	3 (3–4)	3.50 (3–4)	2.50 (2–3)	3.50 (3–4)	3.50 (3–4)	<.001
mDISCERN score	2 (2–3)	2 (1–3)	2 (1–2)	2 (2–3)	3 (2–3)	.002
JAMA score	2 (2–3)	2.50 (2–3)	1 (1–2)	2 (2–2.50)	2 (2–3)	<.001

Values are presented as median (interquartile range) unless otherwise indicated.

GQS = Global Quality Scale, IQR = interquartile range, JAMA = Journal of the American Medical Association, mDISCERN = modified DISCERN.

Figure 3.

Global Quality Scale (GQS) scores (A), modified DISCERN (mDISCERN) scores (B), and Journal of the American Medical Association (JAMA) benchmark scores (C), according to video source. * P < .05; ** P < .01; *** P ≤ .001. GQS = Global Quality Scale, JAMA = Journal of the American Medical Association, mDISCERN = modified DISCERN.

Further analysis of individual items in the mDISCERN questionnaire (Fig. 4) revealed that low scores primarily stemmed from a failure to cite evidence-based sources to support introductory statements (item 2) and a lack of additional information resources for patients to consult (item 4). Additionally, most videos presented information and issued recommendations in an overly definitive tone, failing to acknowledge individual variations, potential exercise risks, or the need to consult a physician, and lacking explanations for areas of uncertainty (Question 5). Further analysis of the JAMA criteria indicated that low scores were mainly due to deficiencies in attribution and transparency. Most videos did not clearly label their information sources, and there was a widespread lack of disclosure of key elements, including conflicts of interest, funding sources, and commercial sponsorships (Fig. 4).

Figure 4.

Item-wise score distributions for the modified DISCERN (mDISCERN) instrument (A) and Journal of the American Medical Association (JAMA) benchmark criteria (B). JAMA = Journal of the American Medical Association, mDISCERN = modified DISCERN.

3.3. Correlation analysis

Spearman correlation analysis identified significant positive correlations (all P < .001) among various interaction metrics, specifically: likes and collections (r = 0.861), likes and comments (r = 0.812), likes and shares (r = 0.832), collections and comments (r = 0.667), collections and shares (r = 0.867), as well as comments and shares (r = 0.652; Table 3).

Table 3.

Intercorrelation among video variables.

Variable and analysis†	Likes	Collections	Comments	Shares	Duration	Days since upload
Likes
r-value	1	–‡	–	–	–	–
P-value	–	–	–	–	–	–
Collections
r-value	0.861**	1	–	–	–	–
P-value	<0.001	–	–	–	–	–
Comments
r-value	0.812**	0.667**	1	–	–	–
P-value	<0.001	<0.001	–	–	–	–
Shares
r-value	0.832**	0.867**	0.652**	1	–	–
P-value	<0.001	<0.001	<0.001	–	–	–
Duration
r-value	−0.124	−0.012	−0.129	−0.063	1	–
P-value	1.000	1.000	1.000	1.000	–	–
Days since upload
r-value	−0.042	−0.027	−0.104	0.229	0.074	1
P-value	1.000	1.000	1.000	0.060	1.000	–

The results represent Spearman’s rank correlation coefficients, with reported P-values adjusted using the Bonferroni method.

^*P < .05.

^**P < .001.

^†Spearman correlation analysis was used.

^‡Not applicable.

Video duration was positively correlated with the GQS (r = 0.470, P < .001) and the mDISCERN (r = 0.363, P < .001), but was not significantly correlated with the JAMA benchmark score (r = 0.166, P = .691; Table 4). In contrast, we found no significant associations between any quality metric and user engagement indicators (all P > .05). Specifically, the GQS, mDISCERN, and JAMA scores were not significantly correlated with likes, collections, comments, or shares. Similarly, days since upload were not significantly correlated with any of the 3 quality assessment scores (all P > .05; Table 4).

Table 4.

Spearman correlation analysis for GQS, mDISCERN, and JAMA scores.

Variable and analysis	GQS	mDISCERN	JAMA
Likes
r-value	0.041	0.072	−0.030
P-value	1.000	1.000	1.000
Collections
r-value	0.084	0.108	−0.008
P-value	1.000	1.000	1.000
Comments
r-value	−0.102	0.010	−0.101
P-value	1.000	1.000	1.000
Shares
r-value	0.051	0.085	0.018
P-value	1.000	1.000	1.000
Duration
r-value	0.470**	0.363**	0.166
P-value	<0.001	<0.001	0.691
Days since upload
r-value	0.007	0.010	0.095
P-value	1.000	1.000	1.000

The results represent Spearman’s rank correlation coefficients, with reported P-values adjusted using the Bonferroni method.

GQS = Global Quality Scale, JAMA = Journal of the American Medical Association, mDISCERN = modified DISCERN.

^*P < .05.

^**P < .001.

4. Discussion

4.1. Principal findings

For the general public, social media platforms have become increasingly important channels for accessing medical knowledge and seeking social support.^[31] With its diverse content formats and rapid dissemination capabilities, Douyin has emerged as a significant platform for health information, demonstrating broad public influence.^[13,32] The 156 videos analyzed in this study generated substantial numbers of comments and shares, further underscoring Douyin’s considerable potential as a platform for disseminating health information. However, this entertainment-focused, highly open platform ecosystem also lowers barriers to spreading misinformation and unsubstantiated medical advice,^[28] leaving users susceptible to misleading or exaggerated health-related content.

This study systematically evaluated Douyin videos related to sarcopenia across 3 dimensions: information quality, source composition, and user engagement. The results indicate that the content primarily focused on basic disease knowledge, such as disease overviews, with relatively limited coverage of risk factors, preventive strategies, and management approaches. Assessment using the GQS, mDISCERN, and JAMA instruments further indicated that the overall quality, reliability, and source transparency of the videos were suboptimal. Although Douyin effectively engages audiences and presents health knowledge in accessible formats, users should not rely on its content as the sole basis for health-related decision-making.

4.2. Factors influencing video popularity

User engagement metrics, including likes, comments, collections, and shares, indicate a video’s popularity.^[33,34] Our study identified strong, statistically significant positive correlations among these Douyin engagement metrics, suggesting that videos with higher like counts tend to accumulate more collections, comments, and shares. This consistency may be because high like counts increase a video’s visibility and appeal, motivating further engagement such as commenting and sharing. Alternatively, the platform’s recommendation algorithm may preferentially promote content with a higher interaction rate.^[35] Overall, user engagement with specific content on social media platforms not only reflects individual interests and levels of involvement but also provides a foundation for disseminating that content within social networks,^[36] thereby facilitating its transmission to a broader audience.

4.3. Overall video Quality

While medical personnel serve as the primary source of content related to sarcopenia on Douyin and essentially facilitate the dissemination of information regarding the disease, the overall quality and reliability of these videos remain insufficient. This finding aligns with conclusions from studies on other health topics.^{[21,23,26,37,38]} Our analysis indicates that content creators may oversimplify complex subjects, such as sarcopenia, to conform to platform requirements for duration, pacing, and engagement, thereby omitting essential details or obscuring the sources of information. In this study, videos uploaded by general users consistently received lower scores across all 3 quality assessment tools. This finding aligns with the evaluation conducted by Song et al on chronic obstructive pulmonary disease-related content,^[20] suggesting that uploader background may be associated with information quality. It is noteworthy that some uploaders identified themselves as healthcare professionals but did not display platform-verified medical credentials. Viewers could not confirm their professional status through publicly available profile information, and this uncertainty may make it difficult for the public to assess source credibility. Furthermore, certain videos contained exaggerated or inaccurate claims, likely intended to enhance engagement, further compromising informational integrity. Consequently, we advise the public to exercise critical judgment when consulting these platforms for health information and, whenever possible, to triangulate such information with authoritative medical sources.

Consistent with prior research linking longer video duration to higher information quality,^[15,27] video duration was positively correlated with GQS and mDISCERN scores in our study. One possible explanation is that longer videos allow content creators to convey information more comprehensively, enabling systematic elaboration of topics and the presentation of multidimensional evidence. However, video duration was not significantly correlated with JAMA scores. This distinction may reflect the different constructs these instruments assess: GQS and mDISCERN primarily assess information quality, comprehensiveness, and reliability, whereas the JAMA benchmarks evaluate source transparency, including authorship, attribution, disclosure, and currency.^[30] These elements depend primarily on explicit reporting rather than video length alone. The brevity of short-form videos may limit the depth of health information they can present,^[39] while their limited duration and on-screen space may also constrain the explicit presentation of creator credentials, source citations, conflict-of-interest disclosures, and content-update information. These format-related constraints may partly explain why only 3 of 156 videos (1.9%) achieved the maximum JAMA score. Therefore, longer videos may facilitate a more comprehensive presentation of health information but should not be interpreted as demonstrating better source transparency or JAMA compliance.

4.4. Correlation between video quality and video characteristics

Our analysis revealed no significant positive correlation between video quality scores and user interaction metrics. This finding suggests that a video’s popularity may not be directly linked to its professional quality assessment. Specifically, high-quality videos did not necessarily receive more likes or shares, while visually compelling videos often exhibited lower informational quality. This phenomenon indicates that Douyin users frequently struggle to effectively distinguish between high- and low-quality video content – a conclusion consistent with previous studies on health-related videos on Douyin.^[26,40,41] However, the default “Comprehensive” ranking used in this study may be influenced by both content relevance and user engagement,^[42] potentially favoring more visible or highly engaged content while underrepresenting less popular videos with different quality characteristics. Accordingly, the observed lack of association should be interpreted in light of this sampling strategy, and caution should be exercised when extrapolating the finding to the broader population of sarcopenia-related videos on Douyin.

Beyond these sampling considerations, the observed discrepancy may also reflect the communication environment of short-video platforms. High-quality content typically prioritizes logical rigor and objective presentation, which can appear dull and unappealing in the short-form video medium. Furthermore, on entertainment-oriented social platforms,^[43] users naturally gravitate toward lighthearted, visually engaging formats. Unpackaged professional medical instructions often lack entertainment value and struggle to capture broad interest. Crucially, Douyin’s algorithmic recommendation system heavily relies on engagement metrics, making videos with high like counts more likely to be identified and widely promoted. This mechanism is further amplified by the echo chamber effect,^[14,44] in which social media users are typically exposed to curated content that aligns with their existing beliefs for extended periods. This environment reinforces their cognitive structures and increases their susceptibility to misinformation.^[45]

4.5. Public health recommendations

Given the complexity of sarcopenia as a disease, effective scientific communication must balance professional depth with public comprehension. Content creators should consider extending video duration to enhance information capacity and deliver high-quality content within a reasonable timeframe. Medical personnel must proactively adapt to new media dissemination patterns, ensuring that content remains professional while improving engagement and expressiveness. Individual uploaders must strengthen accountability by verifying health information against mainstream medical journals and national health authorities before publication, thereby improving content quality. Furthermore, Douyin should reform its recommendation algorithm, which currently prioritizes engagement metrics,^[46] by incorporating criteria for assessing information quality. Strict content review procedures must be implemented,^[41,47] with traffic restrictions or warning labels applied to unverified individuals who post content related to disease treatment protocols or risk information. Additionally, the platform could establish a dedicated “Health Science” section that aggregates high-quality content from leading hospitals, medical societies, and renowned experts, providing users with a reliable source of credible health information.^[48]

4.6. Limitations and future directions

The study systematically evaluated the quality of Douyin videos related to sarcopenia, but it has several limitations. First, focusing solely on Chinese-language content limits the generalizability of the findings, as Douyin and the international TikTok platform differ in their target audiences, operational mechanisms, and content moderation.^[41,46] Second, although the assessment tools employed are commonly used across various health topics, most were not explicitly designed for short health videos. Consequently, their application to health science short videos presents limitations. Future studies should develop and validate quality assessment instruments tailored to short-form health videos to improve the validity and consistency of research findings. Third, restricting retrieval to the top 100 results for each keyword under the platform’s default “Comprehensive” ranking may have overrepresented highly ranked and engagement-favored videos while underrepresenting less visible content. Future studies should compare samples retrieved using alternative sorting options, such as “Most Recent,” to evaluate the robustness and generalizability of the findings. Fourth, we did not assess potential confounding between video duration and uploader type using a multivariable model. Future studies should use multivariable analyses to determine whether video duration remains associated with quality after adjustment for uploader type and other relevant video characteristics.

5. Conclusions

In the analyzed sample of sarcopenia-related videos on Douyin, overall information quality, reliability, and source transparency were suboptimal. Although medical professionals are the primary content creators, significant opportunities for improvement remain. Furthermore, video duration was positively correlated with GQS and mDISCERN scores but not with JAMA scores, while user engagement metrics were not significantly associated with any of the 3 quality measures. These findings suggest that video popularity should not be regarded as a reliable indicator of information quality. Users should therefore exercise caution when getting health information from short-video platforms and verify it against authoritative medical sources. Platform operators should strengthen content-quality review and provide clearer verification of professional credentials to support a more reliable, scientifically sound digital health information environment.

Acknowledgments

The authors would like to acknowledge the video uploaders for their contributions to public health.

Author contributions

Conceptualization: Ting Zhang, Jinli Guo.

Data curation: Ting Zhang, Jiabei Wu.

Formal analysis: Ting Zhang.

Investigation: Ting Zhang, Jiabei Wu.

Methodology: Ting Zhang, Jiabei Wu.

Project administration: Jinli Guo.

Resources: Jinli Guo.

Software: Ting Zhang.

Supervision: Jinli Guo.

Validation: Ting Zhang.

Visualization: Ting Zhang.

Writing – original draft: Ting Zhang.

Writing – review & editing: Jiabei Wu, Jinli Guo.