Quality and Reliability of Social Media Videos on Weight-Loss Interventions for Obstructive Sleep Apnea: A Multi-Platform Cross-Sectional Study Authors: Rexidanmu Hudabai, Jiaomin Jian, Peng Yang, Yan Yi, Yingxiang Xu, Aiqing Peng, Jingyu Yao, Xianzhen Chen Published: 2026 ABSTRACT AbstractPurposeThis study aimed to evaluate the quality and reliability of social media videos addressing weight loss for OSA across three major platforms (YouTube, TikTok, Bilibili) and to identify key determinants of trustworthy content.MethodsWe conducted a cross-platform content analysis of publicly available videos on weight loss for OSA from YouTube, TikTok, and Bilibili, representing major video-sharing platforms with different regional audiences and content formats. The top 100 default search results from each platform were screened using predefined criteria, yielding 235 eligible videos. Video and uploader characteristics, engagement metrics, and quality scores were extracted. Videos were assessed using PEMAT, VIQI, GQS, mDISCERN, and JAMA benchmarks. The primary outcome was overall video quality assessed by the Global Quality Score (GQS), while other assessment tools were considered secondary outcomes. For TikTok, we used English-language search terms (“obstructive sleep apnea and weight loss” and “OSA and weight loss”) to capture content comparable with YouTube. The Chinese-language TikTok counterpart (Douyin) was not included because it operates under a distinct platform ecosystem with different content algorithms, regional audience, and moderation policies.ResultsYouTube videos scored significantly higher in understandability, actionability, and overall quality compared with TikTok and Bilibili. Content from verified institutions and healthcare professionals consistently received higher ratings across all tools. Engagement metrics (likes, shares, comments) showed weak correlations with information quality. In regression models, production quality (VIQI) and source transparency (JAMA benchmarks) emerged as the strongest predictors of content reliability, while video duration had minimal effect.ConclusionWithin digital health environments, popular engagement does not equate to informational quality for OSA-related weight-loss content. Platform- and uploader-level characteristics—particularly verification status and production standards—are key markers of reliability. These findings underscore the need for platform-level governance that prioritizes verified health sources, algorithmic transparency, and integrated digital health literacy efforts to safeguard patients navigating lifestyle advice for OSA online.Keywords: obstructive sleep apnea, social media, public health, Youtube, TikTok, Bilibili Introduction Obstructive sleep apnea (OSA) is a common sleep-related breathing disorder characterized by recurrent collapse of the upper airway during sleep, leading to apneas and hypopneas.1,2 OSA is closely associated with hypertension, cardiovascular disease, metabolic syndrome, cognitive impairment, as well as male sexual dysfunction and reduced quality of life, making it a significant global public health issue.3–6 Weight management and weight loss therapy are widely recognized as essential non-surgical interventions for OSA. These strategies can improve upper airway patency and sleep quality, while concurrently reducing the risk of OSA-related complications and alleviating the associated healthcare and economic burden.7,8 Weight-loss interventions are a cornerstone of OSA management, but guidance in this area is often complex and individualized. Online platforms are replete with diet fads, rapid-weight-loss strategies, off-label therapies, and commercially driven advice that may not be evidence-based. Patients seeking OSA-related weight-loss information are therefore exposed to a high risk of encountering misleading or unverified content. The combination of medical complexity, variable patient adherence, and the prevalence of attention-grabbing digital content increases susceptibility to misinformation, justifying a systematic evaluation of the quality and reliability of social media videos in this domain. In the setting of increasingly constrained healthcare resources and limited clinical consultation time, patients are progressively seeking health-related information beyond traditional medical encounters. In parallel, the rapid expansion of the internet and social media has fundamentally reshaped health information–seeking behaviors.9 Video-sharing and short-form platforms, such as YouTube, TikTok, and Bilibili, have become integral components of the digital health ecosystem, providing highly accessible and engaging modalities for health communication.10,11 However, in contrast to peer-reviewed medical literature, these platforms typically function without standardized professional oversight, raising substantial concerns regarding the accuracy, completeness, and reliability of the medical information disseminated. The dissemination of low-quality or misleading health content has increasingly been recognized as a major public health challenge in the digital environment, with potential adverse implications for patient perceptions, self-management practices, and clinical decision-making.12 Previous studies evaluating health-related content on social media have consistently demonstrated suboptimal informational quality across a broad range of medical topics. Using standardized assessment instruments, including the Global Quality Scale (GQS), the modified DISCERN (mDISCERN) tool, and the Video Information and Quality Index (VIQI), these studies have shown that viewer engagement metrics—such as view counts, likes, and comments—do not reliably reflect scientific accuracy or educational value.13–15 Moreover, substantial heterogeneity in information quality has been observed across platforms, disease-specific domains, and content creator backgrounds, underscoring the critical need for topic-specific and platform-sensitive evaluations of online health information.16–18 Despite the expanding body of literature on health communication via social media, no study to date has systematically evaluated the quality, reliability, and educational value of videos focusing on weight-loss management strategies for OSA across multiple platforms. Given that weight management represents a cornerstone of both clinical treatment and patient self-management in OSA,7,19 understanding how such information is presented and disseminated in the online environment is particularly important. Therefore, this study was guided by the following focused research question: among publicly available YouTube, TikTok, and Bilibili videos addressing weight-loss interventions for OSA, what are the levels of understandability, actionability, overall quality, reliability, and transparency, and which platform- and uploader-level characteristics are associated with higher-quality and more reliable content? By addressing this question, the study aimed to provide evidence to inform patients, healthcare professionals, content creators, and platform governance, thereby contributing to improved digital health communication in sleep medicine. The findings are expected to provide evidence to inform patients, healthcare professionals, content creators, and platform governance, thereby contributing to improved health communication within the digital health landscape. Accordingly, the present study was guided by the following focused research question: among publicly available YouTube, TikTok, and Bilibili videos addressing weight-loss interventions for OSA, what are the levels of understandability, actionability, overall quality, reliability, and transparency, and which platform- and uploader-level characteristics are associated with higher-quality and more reliable content? Materials and Methods Search Strategy and Data Collection This study was initiated on December 10, 2025, and searches were performed over three consecutive days (December 10–12, 2025) on YouTube, Bilibili, and TikTok. To reduce the impact of personalized recommendations, we created new anonymous accounts and cleared browsing histories before each search. Given the primary language used on each platform, we employed English keywords (“obstructive sleep apnea and weight loss”, “OSA and weight loss”) for YouTube and TikTok, and the corresponding equivalent keywords in Chinese for Bilibili. On each platform, we took the top 100 videos returned by the default relevance‑based ranking for initial screening. This cutoff was selected based on pilot observations indicating that videos beyond the first 100–150 results showed increased redundancy and lower topical relevance. Therefore, the top 100 videos were considered representative of the content most likely encountered by typical users while allowing feasible in-depth quality assessment. Because the primary user language and search environment differed across platforms, the resulting comparisons were interpreted as cross-platform and cross-language comparisons rather than as isolated platform effects. Because social media platforms dynamically update their recommendation algorithms and search rankings, the retrieved videos represent a snapshot of available content at a single time point rather than a permanent representation of online information. We excluded videos that met any of the following criteria: (1) non-Chinese or non-English content; (2) duration shorter than 20 seconds; (3) content unrelated to the topic of obstructive sleep apnea (OSA); (4) duplicate or highly similar content; (5) purely commercial promotional or advertisement videos; (6) videos published within the past 7 days. (Supplementary Material 1 for details; the study flowchart is presented in Figure 1). Video Data Collection, Screening, and Analysis Flowchart. Video Classification and Assessments of Video Quality Videos that passed the initial screening underwent a two-phase evaluation. Phase One: Extraction of Video and Uploader Characteristics Basic video features (eg, duration, upload date, views, and engagement metrics) and uploader information (account type, verification status, follower count, etc.) were extracted. An engagement index and a daily average view rate were subsequently calculated (see Supplementary Material 1 for details). Phase Two: Content Classification and Quality Assessment In the second stage, video content classification and quality assessment were independently performed by two uniformly trained researchers with clinical medical backgrounds and more than three years of clinical experience. Both researchers completed standardized training covering video content analysis and health information quality evaluation and were proficient in all assessment tools. Prior to formal evaluation, 20 pre-test videos (not included in the official sample) were independently scored for calibration. Discrepancies were discussed item-by-item and resolved in accordance with official scale guidelines, with all pre-test results reviewed and verified by a senior clinician with over ten years of clinical experience. Formal evaluation commenced only when the inter-rater discrepancy was controlled within 20% of the full score of each scale. Content classification was conducted based on a pre-established coding framework consisting of eight thematic categories with standardized operational definitions and keyword examples (Supplementary Material 1). Video health information quality was assessed using five validated evaluation tools: the Patient Education Materials Assessment Tool for Understandability (PEMAT-U) and Actionability (PEMAT-A),20 the revised Video Information Quality Index (VIQI) covering four dimensions (accuracy, flow, completeness, and presentation),18 the Global Quality Score (GQS),21 the modified DISCERN (mDISCERN) scale comprising five binary items (clarity, authoritative citation, impartiality, supplementary information, and uncertainty statement) with a total score ranging from 0 to 5, and the JAMA benchmark criteria for health information quality and reliability evaluation.22 PEMAT-A contains four actionable items. Items were scored only when corresponding operational recommendations were provided in videos; non-applicable items were marked as not applicable and excluded from scoring. Samples with all four items non-applicable were recorded as missing data. All 235 included videos contained at least one actionable health recommendation, resulting in no missing PEMAT-A data. Disagreements regarding item applicability were adjudicated by a third senior evaluator. Final scores were calculated as the mean value of the two independent evaluators. A third senior researcher resolved inconsistencies and made final decisions if scoring discrepancies exceeded 20% of the full score of any assessment tool or if disagreements occurred in thematic classification (Supplementary Material 1). Inter-rater reliability was evaluated for all 235 videos. Weighted Kappa (κw) was applied for ordinal scales (GQS and mDISCERN), and two-way mixed-effects single-measure intraclass correlation coefficients (ICC 2,1) were calculated for continuous scales (PEMAT, VIQI, and JAMA). The reliability results were as follows: GQS (κw = 0.82, 95% CI: 0.74–0.90), mDISCERN (κw = 0.85, 95% CI: 0.78–0.92), PEMAT-U (ICC = 0.89, 95% CI: 0.84–0.93), VIQI (ICC = 0.86, 95% CI: 0.80–0.91), and JAMA (ICC = 0.91, 95% CI: 0.87–0.94). All reliability coefficients exceeded 0.80, indicating excellent inter-rater agreement. For thematic classification reliability, the preliminary test on 30 videos yielded an overall agreement rate of 93.3% with a Cohen’s κ of 0.91 (95% CI: 0.78–1.00). For the formal classification of 235 videos, 209 videos were consistently categorized by both evaluators, with an agreement rate of 88.94%. The remaining 26 discrepant cases were finally adjudicated by a senior clinician. Predefined Research Question and Outcomes The primary outcome of this study was overall video quality, assessed using the Global Quality Score (GQS), which provides an overall evaluation of the educational value and quality of health-related videos. Secondary outcomes included video understandability and actionability assessed using PEMAT-U and PEMAT-A, information reliability assessed using mDISCERN, transparency assessed using the JAMA benchmark criteria, VIQI scores, platform-level differences, uploader characteristics, engagement indicators, and associations between engagement metrics and video quality or reliability (Supplementary Material 2). Statistical Analysis All statistical analyses were performed using IBM SPSS Statistics 27.0. For descriptive analysis, continuous variables were presented as the median and interquartile range (IQR). The Shapiro–Wilk test was routinely applied to evaluate the distribution of continuous variables, and all variables showed a non-normal distribution (P < 0.001); accordingly, non-parametric statistical methods were adopted for all subsequent analyses. For comparative analyses across YouTube, TikTok, and Bilibili, the Kruskal–Wallis H-test was used for overall group comparisons, with χ2 and p values reported. Significant overall results were further examined using Dunn’s post-hoc pairwise test, and the Benjamini-Hochberg (BH) method was employed for multiple testing correction. The Mann–Whitney U-test was used for two-group comparisons, and BH correction was applied only when the number of comparisons was ≥ 3. Comparisons of uploader types (≥ 3 groups) were also performed using the Kruskal–Wallis H-test with Dunn’s post-hoc test and BH correction. Spearman’s rank correlation coefficient (ρ) was used to assess the associations between video quality scores and audience engagement metrics. Effect sizes with 95% confidence intervals (CIs) were calculated and reported for all key group comparisons. For Kruskal–Wallis tests with three or more groups, effect sizes were calculated using eta-squared (η2[H]) from the H-statistic, using the formula η2[H] = (H - k + 1)/(n - k), where H is the chi-square statistic, k is the number of groups, and n is the total sample size. Effect sizes were categorized as small [0.01, 0.06), moderate [0.06, 0.14), or large [0.14, 1.00] based on established thresholds. Exploratory stepwise regression analyses were conducted based on the Akaike Information Criterion (AIC) for variable selection to identify variables associated with video quality and reliability. Given the conceptual overlap among quality assessment instruments, regression findings were interpreted as descriptive rather than causal. All regression models reported unstandardized coefficients (B), standard errors (SE), standardized coefficients (β), 95% CIs, t-values, p-values, and collinearity statistics (tolerance and variance inflation factor, VIF). All statistical tests were two-tailed, and statistical significance was set at P < 0.05. Results Basic Characteristics and Topic Distribution of Videos A total of 235 videos meeting the inclusion criteria were analyzed (Bilibili: n=71; TikTok: n=73; YouTube: n=91). The results revealed significant differences in key metrics across the three platforms (Table 1). TikTok had the highest median likes (117) and interaction index (3.57). The differences in interaction index between platforms were statistically significant (adjusted P < 0.001 for pairwise comparisons; η2 [H] = 0.14, 95% CI: 0.07–0.25). Although Bilibili achieved the highest median view count (2278), it registered the lowest interaction index (1.13). YouTube’s engagement metrics were relatively moderate. It is important to acknowledge that interaction indices are based on engagement features specific to each platform (for instance, “likes” on YouTube or TikTok as opposed to “coins” on Bilibili), and thus may not be directly comparable across different platforms. Table 1Characteristics of Videos Related to OSA and Weight Loss on YouTube, Bilibili, and TiktokVariablesBilibili (n = 71, Median (IQR))TikTok (n = 73, Median (IQR))YouTube (n = 91, Median (IQR))χ2pViews2278.00 (673.00–13,500.00)2804.00 (911.00–8723.00)751.00 (172.00–6589.50)11.50.003Likes43.00 (6.00–476.00)117.00 (32.00–551.00)10.00 (2.00–132.00)22.9<0.001Coins6.00 (0.00–28.00)––––Favorites36.00 (4.00–256.00)25.00 (4.00–83.00)–0.1650.685Shares23.00 (3.00–289.00)27.00 (4.00–101.00)–0.0000.997Video duration134.00 (74.00–369.50)59.00 (41.00–86.00)100.00 (60.00–255.00)38.9<0.001Days Since Release770.00 (299.00–1230.00)345.00 (122.00–859.00)378.00 (310.50–1080.00)11.20.004Comments5.00 (0.00–50.50)9.00 (2.00–33.00)0.00 (0.00–7.00)20.6<0.001Interaction Index1.13 (0.75–2.50)3.57 (1.81–8.79)1.59 (0.80–3.22)35.0<0.001View Rate4.50 (1.19–33.84)8.55 (3.66–49.60)1.34 (0.29–19.62)20.1<0.001 Characteristics of Videos Related to OSA and Weight Loss on YouTube, Bilibili, and Tiktok Video duration exhibited distinct platform‑specific patterns: Bilibili hosted the longest videos (median duration: 134 seconds), TikTok the shortest (59 seconds), and YouTube fell in between (100 seconds) (Table 1). Content focus also varied: Bilibili was primarily dominated by “Basic Knowledge of OSA” (25.4%) and “Multidisciplinary Comprehensive Treatment” (22.5%); TikTok focused on “Pharmacotherapy” (23.3%) and “Treatment Efficacy and Prognosis” (21.9%); while YouTube contained the highest proportion of videos mentioning “Bariatric Surgery” (14.3%) (Figure 2A). Regarding production style, although monologue narration was a common format across platforms, YouTube and Bilibili offered a greater diversity of styles, including documentaries, animations, and PPT-style presentations (Figure 2B). Content analysis of videos on YouTube, Bilibili, and TikTok. (A) Main topic types. (B) Video style types. Characteristics of Video Uploaders The distribution of uploader types and their influence varied markedly across platforms (Table 2). YouTube was predominantly led by Verified Group (35.2%), whereas Unverified Individuals (57.5%) constituted the majority on TikTok. Bilibili presented an intermediate profile, with Verified Individuals (45.1%) forming the core group. Overall, content from authoritative sources accounted for the highest proportion on YouTube (exceeding 60%) (Figure 3A). Table 2Characteristics of Video Uploaders About OSA and Weight Loss on YouTube/Bilibili/TikTokVariableUnverified GroupUnverified IndividualVerified GroupVerified Individualχ2P valueBilibili6231032Followers1669.00 (442.00–2001.25)2703.00 (370.50–9932.50)135,500.00 (13,868.25–372,000.00)8010.50 (927.50–118,250.00)9.140.028Video Posting Frequency0.15 (0.13–0.24)0.18 (0.03–0.53)2.20 (0.68–3.38)0.40 (0.18–0.74)15.5<0.001Total Videos147.00 (45.50–356.50)132.00 (70.00–333.00)1432.50 (449.50–5458.25)457.00 (210.50–1107.50)18.7<0.001Total Video Views89500.00 (75,500.00–118,500.00)375,000.00 (17,000.00–2,115,000.00)5,397,270.00 (1,516,500.00–60,041,000.00)2,448,000.00 (323,000.00–13,750,750.00)13.6<0.001Total Likes of All Videos1058.50 (635.25–2440.25)4297.00 (1157.00–77,000.00)389,500.00 (19,250.00–1,935,222.00)47,500.00 (4702.00–350,250.00)12.5<0.001TikTok042427Follower–2564.00 (534.50–32,750.00)1,061,700.00 (169,508.00–2,075,000.00)34,100.00 (1219.00–50,400.00)7.990.018Video Posting Frequency–0.67 (0.06–1.06)1.29 (0.93–1.37)0.78 (0.33–1.23)3.050.217Total Videos–762.00 (236.00–1378.75)23,348.50 (11,450.50–82,489.00)1246.00 (352.00–2222.50)7.160.028Total Likes of All Videos–20,000.00 (1231.75–220,100.00)91,105,000.00 (24,854,600.00–160,340,000.00)243,100.00 (6666.00–412,150.00)10.5<0.001YouTube0353224Follower–10,300.00 (306.50–39,450.00)148,000.00 (3322.50–1,832,500.00)3615.00 (1072.50–32,925.00)13.8<0.001Video Posting Frequency–0.00 (0.00–0.01)2.54 (0.00–23.70)0.01 (0.00–0.44)12.40.002Total Videos–336.00 (128.00–1351.50)5529.00 (393.75–46,039.00)364.50 (196.25–883.50)13.9<0.001Total Video Views–1,564,954.00 (108,621.50–9,106,759.50)74,955,119.50 (1,031,081.00–1,229,285,481.75)471,831.00 (219,926.50–4,600,306.75)16.4<0.001 Characteristics of Video Uploaders About OSA and Weight Loss on YouTube/Bilibili/TikTok Uploader characteristics on YouTube, Bilibili, and TikTok. (A) Heat map: Distribution of uploader types per platform. (B) Scatter plot: Number of subscribers. The long vertical line represents the median, and the short vertical line represents the interquartile range. ns (not significant) between the indicated groups. YouTube uploaders also had the highest median number of subscribers, indicating that its content creators generally possessed a broader viewer base. In contrast, Bilibili uploaders had the lowest median subscriber count, though some accounts maintained a cohort of highly engaged followers within specific niches (Figure 3B). Video Quality Assessment This study employed five assessment tools—PEMAT, VIQI, GQS, mDISCERN, and JAMA—to conduct a multidimensional evaluation of video quality. The results revealed significant differences in content attributes across various platforms and uploader types. Video Understandability and Actionability The evaluation results of video understandability, actionability, and total scores indicated that both platform source and uploader type influenced video quality (Figure 4). At the platform level, YouTube videos achieved the highest median score for PEMAT Understandability (PEMAT-U) with a concentrated score distribution, reflecting superior understandability and stable quality (Figure 4A). TikTok presented moderate and tightly distributed PEMAT-U scores, whereas Bilibili exhibited the widest score range with poorer quality consistency (η2 [H] = 0.16, 95% CI: 0.09–0.25). YouTube also outperformed the other two platforms in PEMAT Actionability (PEMAT-A) and overall PEMAT total scores (Figures 4B and C). Comparative analysis of video understandability (PEMAT-U) and Actionability (PEMAT-A) of OSA and weight loss Videos. (A) Patient Education Materials Assessment Tool-Understandability scores of videos across YouTube, TikTok and Bilibili. (B) Patient Education Materials Assessment Tool-Actionability scores of videos across YouTube, TikTok and Bilibili. (C) Patient Education Materials Assessment Tool-Total scores of videos across YouTube, TikTok and Bilibili. (D) Patient Education Materials Assessment Tool-Understandability scores of videos across different sources. (E) Patient Education Materials Assessment Tool-Actionability scores of videos across different sources. (F) Patient Education Materials Assessment Tool-Total scores of videos across different sources. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001 and ns (not significant) between the indicated groups. Stratified analysis by uploader type showed that videos released by verified institutions had the best understandability, which may be attributed to standardized team-based production procedures. In contrast, videos from unverified individuals showed greater score fluctuation and inconsistent quality (Figure 4D). For actionability, verified institutions obtained the highest scores, while videos from non-professional creators generally scored lower (Figure 4E). Total score comparison further confirmed that videos published by verified institutions achieved significantly better quality than those from other uploader categories (Figure 4F). Information Quality and Production Standard The influence of uploader type on video information quality (VIQI) varied across platforms (Table 3). On Bilibili, no significant differences were observed in the total VIQI score or its sub-dimension scores among different uploader types (P > 0.05). On TikTok, significant between-group differences were found in all dimensions and the total score (P < 0.05), except for VIQI-3 (P = 0.082), indicating that verification status significantly affected video quality on this platform. On YouTube, all VIQI metrics demonstrated highly significant between-group differences (P< 0.01), suggesting that whether the uploader was verified had a pronounced impact on the assessment of video quality (Table 3). Table 3Quality Comparison Between the Videos Uploaded by Different SourcesVariableUnverified GroupUnverified IndividualVerified GroupVerified Individualχ2P valueBilibiliVIQI10.5 (8–16.75)12 (9–15)10.5 (9.25–13.5)10.5 (8–13)1.900.593VIQI-13 (2–4.75)4 (3–4.5)3 (3–3.75)3 (2–4)1.770.622VIQI-23 (2–4.75)4 (2.5–4)3 (3–3.75)3 (2–3.25)2.560.464VIQI-32 (2–2.75)2 (2–3)2.5 (2–3)2 (2–3)2.050.561VIQI-42.5 (2–3.75)3 (2–4)2 (2–3)2.5 (2–3)1.990.575TikTokVIQI-7 (6–10)14.5 (12.75–15.25)10 (9–13)13.01<0.001VIQI-1-2 (2–3)4.5 (3.5–5)3 (2–3)10.350.006VIQI-2-2 (1–3)4 (3.5–4)3 (3–4)16.94<0.001VIQI-3-1.5 (1–2)2.5 (2–3)2 (1–3)5.010.082VIQI-4-2 (1–3)3.5 (3–4)3 (2–3)9.910.007YouTubeVIQI-12 (11–14)17.5 (15.75–19)14 (13–17)36.36<0.001VIQI-1-3 (3–3)5 (4–5)3 (3–4)27.75<0.001VIQI-2-3 (3–4)5 (5–5)5 (5–5)54.04<0.001VIQI-3-3 (2–3.5)3 (3–4)3 (2–4)9.350.009VIQI-4-3 (3–3)5 (4–5)4 (3–4)30.50<0.001 Quality Comparison Between the Videos Uploaded by Different Sources Overall Quality and Reliability For the primary outcome, GQS scores demonstrated significant differences across platforms and uploader categories. Secondary analyses evaluated reliability using mDISCERN and transparency using JAMA benchmark criteria. Bilibili showed the widest distribution of GQS scores, with some videos achieving higher ratings but a greater concentration in the lower score range (Figure 5A). Regarding reliability, mDISCERN scores were generally low across all platforms. However, the Kruskal–Wallis H-test revealed significant differences in mDISCERN scores among the three platforms (H = 21.3, η2[H] = 0.0833). TikTok scored significantly lower than YouTube (P < 0.001) and Bilibili (P = 0.0238), while Bilibili and YouTube were comparable (P = 0.160) (Figure 5B). From the perspective of uploader type, videos from verified individuals and verified organizations generally received higher GQS scores (Figure 5C). In terms of reliability scores, verified organizations performed the best, while videos from individual sources exhibited more dispersed and polarized score distributions. This may be directly associated with the diverse educational and professional backgrounds of individual uploaders (Figure 5D). Global Quality Score and modified DISCERN for OSA and weight loss Videos. (A) Comparison of Global Quality Scores between YouTube, TikTok and Bilibili videos. (B) Comparison of modified DISCERN scores between YouTube, TikTok and Bilibili videos. (C) Global Quality Scores across different video sources. (D) Modified DISCERN scores across different video sources. *P < 0.05, **P< 0.01, ***P< 0.001, ****P< 0.0001 and ns (not significant) between the indicated groups. Transparency Assessment using the JAMA benchmark criteria revealed that TikTok and Bilibili scored higher than YouTube on the “Authorship” attribute, which is likely attributable to their structured verification systems for healthcare content creators. However, all platforms showed substantial deficiencies in the “Attribution” and “Disclosure of potential conflicts of interest” domains (Figure 6A). Regarding video source, verified uploaders were more likely to provide declarations concerning authorship and content statements (Figure 6B). JAMA Score for OSA and weight loss Videos. (A) Comparison of JAMA Scores between YouTube, TikTok and Bilibili videos. (B) JAMA Scores across different video sources. **P < 0.01, ****P < 0.0001 and ns (not significant) between the indicated groups. Correlation and Regression Analysis Spearman correlation analysis revealed strong consistency among video engagement indicators (likes, comments, favorites, and shares), with correlation coefficients ranging from 0.86 to 1.00, indicating that users tended to engage in multiple interactive behaviors on the same video (Figure 7). Creator follower counts were positively correlated with all video engagement metrics (r = 0.58–0.66). Overall video quality (GQS) was positively correlated with content reliability (mDISCERN), information transparency (JAMA), and all subdimensions of VIQI (r = 0.58–1.00), suggesting that higher-quality videos generally exhibit greater reliability and standardization. In contrast, weak correlations were observed between engagement indicators and scores of GQS, mDISCERN, JAMA, and VIQI, demonstrating that public engagement was not positively associated with informational quality or reliability. Video popularity may therefore be driven more by non-quality factors, such as title attractiveness and emotional resonance. Video understandability (PEMAT-U) was positively correlated with GQS (r = 0.71) and mDISCERN (r = 0.72), while video actionability (PEMAT-A) was correlated with all VIQI subdimensions (r = 0.48–0.63), indicating that well-produced videos with sufficient visual materials are more comprehensible and operable for viewers. Correlation matrix of variables assessing the characteristics, quality, understandability, and actionability of OSA and weight loss videos.DIS:mDISCERN; GQS: Global Quality Score; PEMAT-A: Patient Education Materials Assessment Tool-Actionability; PEMAT-U: Patient Education Materials Assessment Tool-Understandability. *P < 0.05, **P< 0.01, ***P< 0.001 and ns (not significant) between the indicated groups. Exploratory stepwise regression analyses were performed to identify variables associated with GQS and mDISCERN scores. For GQS, the final model incorporated VIQI score, video duration, JAMA score, PEMAT-A score, and PEMAT-U score, with an adjusted coefficient of determination (R2) of 0.622, explaining 62.2% of the total variance (Table 4). VIQI score showed the strongest positive association with GQS scores. (β=0.397, P < 0.001), followed by JAMA score (β=0.216, P < 0.001), whereas video duration showed a negligible and non-significant predictive effect (β=0.070, P = 0.100) (Tables 5 and 6). For mDISCERN prediction, the final model achieved an adjusted R2 of 0.657, accounting for 65.7% of the variance. JAMA score was the primary predictive factor (β=0.410, P < 0.001), and VIQI, PEMAT-U, and PEMAT-A scores also exerted significant predictive effects, while video duration remained non-significant (Table 7). Collinearity diagnostics showed a maximum VIF of 3.8 and a minimum tolerance of 0.26 among the VIQI, JAMA, and PEMAT indicators, indicating low-to-moderate multicollinearity. These values remained within commonly accepted thresholds, suggesting that collinearity was unlikely to materially affect the interpretation of the regression models. Table 4Stepwise Regression Analysis Predicting Global Quality Scores of OSA and weight Loss on TikTok, Bilibili and YouTubeModelPredictors IncludedR2Adjusted R2SE of the EstimateF ChangeP valueANOVAFP1GQS ~ VIQI + duration0.5780.5740.655158.8850.000158.885< 0.0012GQS ~ VIQI + duration + JAMA0.6140.6090.62821.3350.000122.319< 0.0013GQS ~ VIQI + duration + JAMA + PEMAT_A0.6240.6170.6216.1810.01495.342< 0.0014GQS ~ VIQI + duration + JAMA + PEMAT_A + PEMAT_U0.6300.6220.6174.0650.04578.103< 0.001 Stepwise Regression Analysis Predicting Global Quality Scores of OSA and weight Loss on TikTok, Bilibili and YouTube Table 5Stepwise Regression Coefficients, Statistical Significance, and Collinearity Assessment for the Prediction of Global Quality Scores of OSA and Weight LossModel and PredictorsUnstandardized CoefficientStandardized CoefficienttP valueCollinearity StatisticsB (95% CI)SEToleranceVif1 Constant0.749 (0.488, 1.010)0.133–5.6240.000–– VIQI0.175 (0.154, 0.196)0.0110.72216.2570.0000.9231.084 Duration0.014 (0.003, 0.025)0.0060.1112.5030.0130.9231.0842 Constant0.668 (0.416, 0.921)0.129–5.1890.000–– VIQI0.129 (0.101, 0.157)0.0140.5339.0290.0000.4802.084 Duration0.011 (0.001, 0.022)0.0050.0902.1060.0360.9121.096 JAMA0.264 (0.152, 0.376)0.0570.2724.6190.0000.4822.0743 Constant0.508 (0.229, 0.788)0.143–3.5620.000–– VIQI0.117 (0.087, 0.146)0.0150.4827.7790.0000.4272.344 Duration0.010 (0.000, 0.021)0.0050.0811.9070.0580.9051.104 JAMA0.237 (0.125, 0.350)0.0580.2444.1230.0000.4652.149 PEMAT_A0.006 (0.001, 0.011)0.0030.1272.4860.0140.6321.5834 Constant0.374 (0.067, 0.681)0.157–2.3900.018–– VIQI0.096 (0.061, 0.131)0.0180.3975.3160.0000.2903.448 Duration0.009 (−0.002, 0.019)0.0050.0701.6530.1000.8911.122 JAMA0.210 (0.094, 0.325)0.0590.2163.5620.0000.4402.274 PEMAT_A0.005 (−0.001, 0.010)0.0030.0911.6900.0920.5621.780 PEMAT_U0.009 (0.000, 0.017)0.0040.1582.0160.0450.2623.815 Stepwise Regression Coefficients, Statistical Significance, and Collinearity Assessment for the Prediction of Global Quality Scores of OSA and Weight Loss Table 6Stepwise Regression Analysis Predicting Modified DISCERN of OSA and Weight Loss on TikTok, Bilibili and YouTubeModelPredictors IncludedR2Adjusted R2SE of the EstimateF ChangeP valueANOVAFP1MDISCERN ~ VIQI + duration0.5180.5140.693124.7940.000124.794< 0.0012MDISCERN ~ VIQI + duration + JAMA0.6320.6270.60771.3630.000132.216< 0.0013MDISCERN ~ VIQI + duration + JAMA + PEMAT_A0.6540.6480.58914.9710.000108.902< 0.0014MDISCERN ~ VIQI + duration + JAMA + PEMAT_A + PEMAT_U0.6640.6570.5826.6030.01190.565< 0.001 Stepwise Regression Analysis Predicting Modified DISCERN of OSA and Weight Loss on TikTok, Bilibili and YouTube Table 7Stepwise Regression Coefficients, Statistical Significance, and Collinearity Assessment for the Prediction of Modified DISCERN Scores of OSA and Weight LossModel and PredictorsUnstandardized CoefficientStandardized CoefficienttP valueCollinearity StatisticsB (95% CI)SEToleranceVif1 Constant0.656 (0.380, 0.932)0.141–4.6570.000–– VIQI0.161 (0.139, 0.184)0.0110.67314.1950.0000.9231.084 Duration0.016 (0.004, 0.028)0.0060.1292.7150.0070.9231.0842 Constant0.514 (0.270, 0.758)0.124–4.1280.000–– VIQI0.081 (0.053, 0.108)0.0140.3365.8360.0000.4802.084 Duration0.011 (0.001, 0.022)0.0050.0912.1870.0300.9121.096 JAMA0.467 (0.358, 0.575)0.0550.4868.4480.0000.4822.0743 Constant0.278 (0.012, 0.543)0.135–2.0500.041–– VIQI0.062 (0.034, 0.090)0.0140.2604.3770.0000.4272.344 Duration0.010 (0.000, 0.020)0.0050.0781.9090.0580.9051.104 JAMA0.427 (0.320, 0.534)0.0550.4447.8220.0000.4652.149 PEMAT_A0.009 (0.005, 0.014)0.0020.1893.8690.0000.6321.5834 Constant0.116 (−0.173, 0.406)0.148–0.7870.432–– VIQI0.037 (0.004, 0.071)0.0170.1562.1980.0290.2903.448 Duration0.008 (−0.002, 0.018)0.0050.0651.5960.1120.8911.122 JAMA0.394 (0.285, 0.503)0.0560.4107.0930.0000.4402.274 PEMAT_A0.007 (0.002, 0.012)0.0030.1452.8390.0050.5621.780 PEMAT_U0.011 (0.003, 0.019)0.0040.1922.5700.0110.2623.815 Stepwise Regression Coefficients, Statistical Significance, and Collinearity Assessment for the Prediction of Modified DISCERN Scores of OSA and Weight Loss Sensitivity Analysis We performed a supplementary sensitivity analysis using a composite quality score (sum of standardized z-scores of PEMAT-U, PEMAT-A, VIQI, and JAMA) to predict GQS and mDISCERN. Results were consistent with the main models. The composite score was the strongest predictor for both GQS (adjusted R2= 0.616, β= 0.764, P < 0.001) and mDISCERN (adjusted R2= 0.635, β= 0.775, P < 0.001), while duration was not significant in either model (P = 0.137 and P = 0.124, respectively). These findings support the robustness of our primary regression results. Full details are provided in Supplementary material 1. Discussion This multi-platform cross-sectional analysis demonstrates that although weight loss interventions for obstructive sleep apnea (OSA) are widely discussed and highly engaged with on social media platforms, the informational quality, reliability, and transparency of such content remain inconsistent and frequently suboptimal. Critically, audience engagement metrics were only weakly associated with evidence-based quality indicators, indicating that highly engaged videos are not necessarily those with the highest informational quality or reliability.23,24 These findings suggest that patients who rely on social media engagement metrics alone may have difficulty distinguishing higher-quality videos from lower-quality videos, whereby misleading or low-quality weight management content may distort patient understanding of OSA pathophysiology, therapeutic effectiveness, and the role of lifestyle modification in long-term disease control.12,25–27 Substantial platform-level heterogeneity was observed in content composition, uploader characteristics, production formats, and quality outcomes. YouTube, which hosted the highest proportion of verified institutional and healthcare professional creators, demonstrated superior understandability, actionability, and overall quality scores for weight loss-related OSA content. In contrast, TikTok and Bilibili were dominated by non-verified individual creators and short-form videos optimized for rapid engagement, yet often exhibited weaker evidentiary grounding and limited transparency.28 These findings suggest that content quality differs substantially across platforms and may be associated with differences in uploader composition, video format, and verification status, frequently privileging emotional salience and narrative appeal over scientific accuracy and completeness.29,30 Uploader verification status emerged as one of the strongest determinants of informational quality and reliability. Videos produced by certified institutions and healthcare professionals consistently achieved higher scores across PEMAT, VIQI, GQS, mDISCERN, and JAMA benchmarks, supporting established health communication theory emphasizing source credibility and institutional endorsement as central to trustworthy health information dissemination.13,31,32 Nevertheless, non-verified creators accounted for a substantial proportion of highly engaged weight loss-related content, particularly on TikTok and Bilibili, this means that some highly visible or highly engaged videos were produced by sources without clear professional verification. Similar patterns have been reported in other domains of preventive health and chronic disease management, including obesity prevention, vaccination, and cancer screening, where non-professional creators often achieve high reach despite limited evidentiary rigor.25,33 In the context of OSA—a condition closely linked to obesity and requiring sustained behavioral modification, weight management adherence, and shared clinical decision-making—such disparities may contribute to unrealistic expectations regarding weight loss efficacy, underutilization of effective therapies, and delayed engagement with formal sleep medicine services.34–36 Engagement indicators—including likes, comments, shares, and saves—were strongly intercorrelated but only weakly associated with informational quality and reliability measures. Our findings showed that engagement indicators were only weakly correlated with quality and reliability scores. Therefore, patients may have difficulty identifying trustworthy content based solely on popularity metrics such as likes, comments, or shares.37 Large-scale studies of online information diffusion demonstrate that emotionally evocative or sensationalized narratives propagate more rapidly and widely than factual content, irrespective of scientific validity.12,26,27 Within weight loss–related OSA content ecosystems, our qualitative observations suggest that some content may overemphasize rapid or extreme weight reduction strategies, dietary misinformation, or commercially driven interventions. However, these specific categories were not systematically coded or quantified in our analysis. Our findings further highlight the central role of content design quality and transparency in shaping informational value. Strong associations were observed between PEMAT understandability and actionability scores and global quality and reliability indicators. Regression analyses identified VIQI production quality and JAMA transparency benchmarks as the strongest predictors of both overall quality and reliability, whereas video duration exerted minimal influence.38 These results suggest that optimizing visual presentation, narrative structure, and evidence disclosure may yield greater improvements in public health communication outcomes than increasing content length or prioritizing platform-specific stylistic conventions. This interpretation aligns with digital health literacy and knowledge translation frameworks emphasizing cognitive accessibility, usability, and trustworthiness as prerequisites for effective lifestyle and behavioral health interventions.32,38,39 From a clinical perspective, This study provides a cross-sectional snapshot of the current social media landscape surrounding weight loss interventions for OSA. Across 235 videos from YouTube, TikTok, and Bilibili, content produced by verified institutions and healthcare professionals consistently demonstrated higher quality and reliability than content from non-verified sources. Importantly, engagement metrics showed limited ability to identify trustworthy information. For sleep clinicians, these findings suggest that patients may frequently encounter highly viewed but suboptimal educational content online. Clinicians should therefore be prepared to discuss online information with patients and recommend reliable, evidence-based resources when counseling individuals about weight management strategies for OSA.40 Potential strategies include strengthening medical content verification pathways, implementing transparency and credibility labeling systems, and recalibrating recommender algorithms to increase the visibility of evidence-based content produced by trusted institutions and professional organizations.12,31 Concurrently, healthcare systems, sleep medicine societies, and obesity prevention agencies should adopt proactive digital engagement strategies, producing platform-adapted, evidence-informed educational content capable of competing effectively within attention-driven social media environments. Such efforts may be particularly impactful in addressing obesity-related sleep disorders, where early lifestyle intervention can substantially reduce disease burden and healthcare costs. Moreover, our findings underscore digital health literacy as a complementary and scalable public health intervention. As individuals increasingly seek guidance on weight loss and sleep health through social media platforms, strengthening their capacity to critically appraise online content—including evaluating source credibility, evidence transparency, and commercial bias—may mitigate misinformation-related harms and enhance patient agency in weight management and sleep disorder prevention.32,38 Improving patients’ ability to evaluate online health information may represent an important complementary strategy. Clinicians should encourage patients to consider source credibility, transparency, and evidence support when seeking information about weight loss interventions for OSA on social media platforms.31 These findings also have practical implications for patient counseling in sleep clinics. Because patients with OSA may increasingly encounter weight-loss advice through social media before or between clinical visits, clinicians should proactively ask about online information sources during counseling. Patients should be reminded that high engagement metrics, such as likes, comments, shares, or views, do not necessarily indicate medical accuracy or suitability for their individual condition. Sleep clinicians may also help patients evaluate online content by encouraging them to check whether the uploader is a verified healthcare professional or institution, whether evidence sources are cited, and whether the advice aligns with established OSA and obesity-management guidelines. In addition, clinicians can direct patients toward trustworthy educational resources and emphasize that weight-loss strategies for OSA should be individualized, sustainable, and integrated with evidence-based therapies such as positive airway pressure therapy, oral appliance therapy, behavioral weight management, pharmacotherapy, or bariatric surgery when clinically appropriate. This study has several strengths, including its multi-platform scope, application of multiple validated assessment instruments, and use of regression modeling to identify structural determinants of content quality and reliability. Because only English-language TikTok content was analyzed, findings may not generalize to Chinese-language short-form video platforms such as Douyin. Differences in content, algorithms, and audience characteristics may affect cross-platform comparisons. However, several limitations warrant consideration. First, the cross-sectional design precludes causal inference regarding the effects of exposure to weight loss–related OSA content on patient knowledge, behavioral change, or clinical outcomes. Second, our dataset represents an algorithmic snapshot of search results at the time of data collection; social media platforms dynamically adjust content based on algorithms, user engagement, and temporal trends, which may limit reproducibility. Third, platform and language were partly confounded in the present design. YouTube and TikTok were searched using English keywords, whereas Bilibili was searched using Chinese keywords. Therefore, observed platform-level differences may reflect combined influences of platform, language, regional creator ecosystems, cultural norms, healthcare systems, audience preferences, and search-term performance, rather than platform architecture alone. Future studies should adopt parallel multilingual search strategies and perform language- or region-stratified analyses. Fourth, although standardized tools were employed and two reviewers independently scored videos with adjudication of discrepancies, stepwise regression was used to explore predictors of video quality and reliability; however, this approach has recognized limitations, including model instability, data-driven variable selection, and potential overfitting. Accordingly, regression results should be interpreted as exploratory rather than confirmatory. Future studies could also integrate automated natural language processing and machine learning approaches to enable scalable surveillance of weight loss misinformation and real-time detection of high-risk content.41–43 Conclusion In conclusion, this cross-sectional analysis of publicly available social media videos found substantial variation in the quality, reliability, and transparency of weight loss–related OSA content across YouTube, TikTok, and Bilibili. In this sample, videos from verified uploaders and those with higher production quality and transparency scores tended to receive higher quality and reliability ratings, whereas audience engagement metrics were only weakly associated with evidence-based quality indicators. Importantly, this study evaluated online video content only and did not measure patient-level outcomes, such as patient knowledge, behavioral change, treatment adherence, or clinical effects. Therefore, the findings should not be interpreted as evidence that exposure to these videos directly influences patient decision-making or health outcomes. Rather, they suggest that clinicians, researchers, and platforms may need to pay closer attention to source credibility, transparency, and content quality when considering the role of social media in OSA-related weight-management communication. Future studies should examine how exposure to such content affects patient understanding, treatment choices, and clinical outcomes. Acknowledgments The authors would like to express their gratitude to the video uploaders for their contributions to public health. Funding Statement This study was supported by the Natural Science Foundation of Guangdong Province, China (grant No. 2025A1515012592). Abbreviations OSA, Obstructive sleep apnea; GQS, Global Quality Scale; Mdiscern, modified DISCERN; VIQI, Video Information and Quality Index; JAMA, Journal of the American Medical Association; PEMAT, Patient Education Materials Assessment Tool; PEMAT-U, Patient Education Materials Assessment Tool-understandability; PEMAT-A, Patient Education Materials Assessment Tool-actionability. Data Sharing Statement The data sets generated during and/or analyzed during this study are available from the corresponding author on reasonable request. Ethical Approval This study did not involve human biological specimens, animal subjects, or identifiable clinical datasets. All analyzed data were publicly available on YouTube, TikTok, and Bilibili at the time of collection and did not include private user information or require active user interaction. Therefore, individual informed consent was not required. The study analyzed only publicly accessible videos on YouTube, TikTok, and Bilibili at the time of data collection. Data were collected through manual viewing and recording of publicly posted content; no private accounts, password-protected content, direct user interaction, automated scraping, or API-based collection methods were used. Uploader-level information, including account names and follower counts, was aggregated or anonymized, and only summary-level data are reported. The study was conducted with reference to the publicly available terms, community guidelines, and content policies of each platform, without making legal interpretations regarding permission requirements. The study was reviewed by the Ethics Committee of the Fifth Affiliated Hospital of Sun Yat-sen University, which confirmed that formal ethical approval was not required because the study used publicly available online content and did not involve human biological specimens, animal subjects, identifiable clinical datasets, or direct interaction with users. Data collection was limited to passive observation and manual recording of publicly accessible video-level and uploader-level information. No private accounts, password-protected content, direct user interaction, automated scraping, or API-based data extraction was involved. To protect uploader privacy, uploader-level data, including account names and follower counts, were aggregated or anonymized, and only summary-level results are reported. No individual uploader identifiers are disclosed in the manuscript. The study was conducted with reference to the publicly available terms of service, community guidelines, and content policies of the respective platforms, without making legal determinations regarding platform permission requirements. Author Contributions Rexidanmu Hudabai, Jiaomin Jian, and Peng Yang are joint first authors. Specific contributions: Rexidanmu Hudabai: conceptualization, methodology, formal analysis and investigation, writing-original draft preparation, resources, Jiaomin Jian: conceptualization, methodology, formal analysis and investigation, writing-original draft preparation, resources, Peng Yang: conceptualization, methodology, formal analysis and investigation, writing-original draft preparation, resources. Yan Yi: conceptualization, methodology, formal analysis and investigation, writing-original draft preparation, Yingxiang Xu: conceptualization, methodology, formal analysis and investigation, writing-original draft preparation. Aiqing Peng: conceptualization, methodology, writing-original draft preparation. Jingyu Yao: conceptualization, writing-review and editing, supervision. Xianzhen Chen: conceptualization, writing-review and editing, funding acquisition, supervision. All authors gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work. Disclosure The authors report no conflicts of interest in this work. References Associated Data Data Availability Statement The data sets generated during and/or analyzed during this study are available from the corresponding author on reasonable request.