This study applied the Chinese keyword “腺样体肥大” (“adenoid hypertrophy”) to screen relevant videos from Bilibili, TikTok, and Xiaohongshu. We included the top 100 videos sorted by overall comprehensive relevance. The 3 platforms were chosen on practical grounds. All 3 platforms are highly popular within China’s short-video market, although each has a distinct content orientation. Bilibili prioritizes knowledge-sharing posts, TikTok delivers content mainly through algorithmic recommendation, whereas Xiaohongshu centers on lifestyle-based user-generated content. We limited the sample to the top 100 videos from each platform ranked by comprehensive relevance, as prior research suggests that users rarely browse beyond the first several pages of search results and are most influenced by highly ranked content [14,27]. To minimize potential bias introduced by platform-specific recommendation algorithms, newly created accounts were used to perform all searches. The search was conducted between April 8, 2025, and April 15, 2025, and included all videos available on each platform at the time of retrieval, regardless of upload date. Duplicate videos, silent videos, advertisements, and content unrelated to the topic were excluded. Videos presented in multiple parts were treated as a single entry. For each included video, the following characteristics were recorded and analyzed: source platform; video duration; number of likes, collections, comments, and shares; uploader identity; verification status of the uploader (as indicated by the platform); and video production style. Two reviewers independently screened all retrieved videos based on predefined inclusion and exclusion criteria. Discrepancies were resolved through discussion; if consensus could not be reached, a third reviewer adjudicated the final decision.

We evaluated the completeness, understandability, actionability, reliability, and overall quality of the included videos. On the basis of the 2025 Chinese Expert Consensus on Adenoid Hypertrophy [4] the 2023 German S2k Guideline for the Diagnosis and Treatment of Adenoid Hypertrophy [28], and previously published methods for video content analysis [29,30], our research team developed an integrity scoring framework. The framework was further refined through consultation with two otolaryngology professors (refer to the Acknowledgments section for details). The framework covered 6 domains: definition, symptoms and complications, risk factors, diagnosis, treatment or management, and outcomes. Each domain was scored on a 5-point scale (0‐2) using predefined criteria: 0=“no content,” 0.5=“minimal content,” 1=“some content,” 1.5=“most content,” and 2=“extensive content” (Multimedia Appendix 1).

Understandability and actionability were assessed using the Patient Education Materials Assessment Tool for Audiovisual Materials (PEMAT-A/V) [31]. This instrument includes 17 items, with 13 items evaluating understandability and 4 items assessing actionability. Each item was rated as “agree” (1 point), “disagree” (0 points), or “not applicable.” Scores for understandability (Patient Education Materials Assessment Tool–Understandability [PEMAT-U]) and actionability (Patient Education Materials Assessment Tool–Actionability [PEMAT-A]) were calculated as follows: (total score÷total possible score)×100, with higher scores indicating better performance (Multimedia Appendix 2).

A modified version of DISCERN (mDISCERN) instrument was used to assess the reliability of the video content. The mDISCERN is a well-established and widely used tool for evaluating the quality of health information in both research and clinical settings [20,32]. Video reliability was evaluated based on 5 criteria: clarity, traceability, balance, relevance, and robustness. Each item was scored as 1 (“yes”) or 0 (“no”), yielding a total score ranging from 0 to 5, with higher scores indicating greater reliability (Multimedia Appendix 3).

The Global Quality Scale (GQS) was used to assess overall video quality. The GQS is commonly applied to evaluate the quality of online health information and uses a 5-point scale, where 1 indicates poor quality and 5 indicates excellent quality [33,34] (Multimedia Appendix 4). These instruments were selected to assess complementary dimensions of video quality. The PEMAT-A/V evaluates understandability and actionability from a patient education perspective, the mDISCERN assesses the reliability and transparency of health information, and the GQS provides an overall global assessment of educational quality.

Prior to formal assessment, the 2 primary reviewers (CH and YZ) jointly evaluated the first 20 videos to calibrate scoring criteria. Both are registered nurses with 8 years of otorhinolaryngology experience and have received training in all assessment instruments used in this study (content completeness scoring, PEMAT-A/V, mDISCERN, and GQS). A third nurse (WJ), a head nurse with 30 years of experience, resolved any disagreements. All videos were independently evaluated by 2 reviewers. During this process, any statements that contradicted established clinical guidelines were recorded verbatim and thematically categorized as misleading (Multimedia Appendix 5). Discrepancies were initially resolved through discussion; if consensus could not be reached, a third reviewer adjudicated the final score. Interrater agreement was assessed using Cohen kappa, interpreted according to the Landis and Koch criteria [35]. The κ values ranged from 0.77 to 0.993 across all measures: content completeness (κ=0.835-0.923), PEMAT-U (κ=0.789-0.993), PEMAT-A (κ=0.933-0.975), mDISCERN (κ=0.77-0.961), and GQS (κ=0.873). These results indicate acceptable-to-excellent interrater agreement.

Uploader verification status (“certification”) refers to platform-designated credentials indicating that an account has undergone official verification, typically confirming its affiliation with a licensed health care professional, medical institution, or recognized organization.

Statistical analyses and data visualization were performed using SPSS (version 26; IBM Corp) and R (version 4.5.3; R Foundation for Statistical Computing). Continuous variables were assessed for normality; normally distributed data are presented as mean (SD), whereas nonnormally distributed data are presented as median (IQR). Categorical variables are presented as frequencies and percentages. For comparisons between 2 independent groups (eg, medical vs nonmedical uploaders), the Mann-Whitney U test was used for continuous variables and the chi-square test was used for categorical variables. For comparisons among the 3 platforms (Bilibili, TikTok, and Xiaohongshu), the Kruskal-Wallis H test was applied. When significant differences were identified, post hoc pairwise comparisons were performed using Mann-Whitney U tests with Bonferroni correction. Spearman rank correlation analysis was used to examine associations between video characteristics (duration and engagement metrics, including likes, comments, shares, and collections) and quality-related scores (content completeness, reliability, PEMAT-U, PEMAT-A, and GQS). To identify independent predictors of overall video quality (GQS), a stepwise linear regression analysis was performed with candidate variables including video duration, log-transformed engagement metrics, uploader type, platform (dummy-coded with TikTok as the reference), PEMAT-U, and PEMAT-A. A 2-tailed P<.05 was considered statistically significant.

This study used only publicly available videos from Bilibili, TikTok, and Xiaohongshu. No personally identifiable information was collected, and there was no direct interaction with platform users. According to Article 32 of the Measures for Ethical Review of Life Science and Medical Research Involving Human Subjects (the official Chinese policy document measures for ethical review of life science and medical research involving human participants, Guowei Kejiao Fa, 2023, No 4) [36], research that uses lawfully obtained publicly available data without harming human subjects or involving sensitive personal information or commercial interests may be exempt from formal ethics review. Therefore, ethics approval was not sought for this study.

As shown in Figure 1, a total of 220 videos on AH were included based on the predefined inclusion and exclusion criteria. Of these, 90 (40.9%) videos were from Bilibili, 63 (28.6%) from TikTok, and 67 (30.5%) from Xiaohongshu. Videos on Bilibili were significantly longer than those on TikTok and Xiaohongshu (P<.001). Xiaohongshu videos had significantly higher numbers of likes, collections, and comments than videos on Bilibili and TikTok (all P<.001), whereas no significant difference in the number of shares was observed among the 3 platforms (P=.09). Detailed characteristics are shown in Table 1.

The source and production characteristics of the videos differed significantly across platforms (all P<.001; Table 2). Medical professionals accounted for most videos on Bilibili (n=69, 76.7%) and TikTok (n=60, 95.2%), but for less than half of those on Xiaohongshu (n=29, 43.3%). Verification rates were high on Bilibili (n=90, 100%) and TikTok (n=60, 95.2%), but low on Xiaohongshu (n=30, 44.8%). Solo narration was the most common format across platforms (n=126, 57.3%). Xiaohongshu had a markedly higher proportion of animation- or action-based videos (n=22, 32.8%) than the other 2 platforms (n=5, ≤5.6%).

As shown in Figure 2, treatment or management and symptoms and complications were the most frequently addressed topics, with 29.6% (n=65) and 28.6% (n=63) of videos, respectively, providing at least some relevant information. In contrast, definitions, risk factors, diagnostic approaches, and outcomes were rarely covered.

We further evaluated content completeness across platforms and uploader types. Completeness scores were generally low (median total score of 1.5-2 out of 12). As shown in Table 3, overall Kruskal-Wallis H tests revealed no significant differences for definitions or symptoms (both P>.05), but significant differences for risk factors (H=8.00; P=.018), diagnosis (H=8.99; P=.01), treatment (H=9.09; P=.011), outcomes (H=25.40; P<.001), and total completeness (H=18.52; P<.001). Post hoc comparisons with Bonferroni correction showed that Bilibili outperformed TikTok in risk factors (adjusted P=.028), diagnosis (adjusted P=.011), and outcomes (adjusted P=.001) and outperformed Xiaohongshu in treatment (adjusted P=.014) and outcomes (adjusted P<.001). Bilibili also had higher total completeness than both TikTok (adjusted P=.002) and Xiaohongshu (adjusted P=.001). No significant differences were observed between medical and nonmedical uploaders for any completeness domain (all P>.05).

In addition, we identified misleading or non–evidence-based information in some videos (Multimedia Appendix 4). A total of 9 videos were clearly misleading. Examples included recommendations of tuina, dietary therapy, or herbal remedies as substitutes for surgery; claims that “90% obstruction can be cured by Chinese herbs”; and exaggerated statements about the consequences of not undergoing surgery (eg, “surgery must be performed immediately, otherwise growth and intelligence will be affected”). It should be noted that observations regarding specific content characteristics were derived from qualitative review, rather than from formal quantitative comparisons. In our dataset, most misleading videos were uploaded by nonprofessional creators. Among the 9 videos that remained accessible, 5 (55.5%) were uploaded by medical professionals (pediatricians or otolaryngologists).

As shown in Table 4, all 5 quality indicators differed significantly across platforms (all P<.05). For content completeness, post hoc comparisons with Bonferroni correction showed that Bilibili outperformed both TikTok (adjusted P<.001) and Xiaohongshu (adjusted P<.001). For actionability (PEMAT-A), Xiaohongshu scored higher than TikTok (adjusted P=.011). No other pairwise comparisons reached statistical significance. The median GQS score was 2 across platforms, with a significant but small difference (P=.041). Figure 3 provides a visual comparison.

As shown in Figure 4, videos uploaded by medical professionals had significantly higher understandability scores than those uploaded by nonmedical individuals (Z=−3.22; P =.001). No significant differences were found for completeness, actionability, reliability, or overall quality (all P>.05).

Spearman correlation analysis (Figure 5) showed that video duration was strongly positively correlated with content completeness (ρ=0.64; P<.001), but not with other quality indicators (all ρ≤0.2). Weak but statistically significant correlations were observed between shares and both completeness (ρ=0.16; P=.02) and actionability (ρ=0.18; P=.006). As shown in the heat map, engagement metrics (likes, collections, comments, and shares) were highly correlated with each other (ρ>0.80; P<.001), and understandability (PEMAT-U) was strongly associated with reliability (mDISCERN) and overall quality (GQS; ρ=0.58 and 0.64, respectively; P<.001). No other meaningful associations were found.

Stepwise linear regression analysis was performed to identify independent predictors of overall video quality (GQS; Table 5). Candidate variables included PEMAT-U, PEMAT-A, video duration (seconds), log-transformed engagement metrics (likes, shares, comments, and collections), uploader type (medical professional vs nonmedical professional), and platform (dummy-coded with TikTok as the reference). The stepwise procedure (entry criterion P≤.05; removal criterion P≥.10) yielded 4 models. The final model (model 4) was statistically significant (adjusted R²=0.461; F_4,215=47.91; P<.001) and included 4 predictors: PEMAT-U (β=.573; P<.001), PEMAT-A (β=.230; P<.001), video duration (β=.166; P=.001), and uploader type (β=.148; P=.005). No other variables entered the model. These results indicate that understandability was the strongest predictor of perceived video quality, followed by actionability and video length, while uploader type had a smaller but significant effect. Engagement metrics and platform did not contribute independently after accounting for content-related characteristics.

This study provides a systematic evaluation of AH-related videos across 3 major Chinese short-video platforms using multiple validated instruments. Overall video quality was suboptimal, with substantial variability across platforms. Bilibili showed a significant advantage only in content completeness; for understandability and reliability, overall differences were observed, but pairwise comparisons did not reach statistical significance after Bonferroni correction. Xiaohongshu had higher actionability than TikTok, yet no platform performed well across all dimensions. Most engagement metrics were unrelated to content quality, except for weak positive correlations between shares and completeness and actionability. Stepwise linear regression further identified understandability (PEMAT-U) as the strongest independent predictor of overall quality (GQS), followed by actionability (PEMAT-A), video duration, and uploader type; engagement metrics and platform did not enter the model. Notably, a substantial proportion of misleading videos originated from medical professionals, in contrast to the pattern observed in previous YouTube-based research [8]. These findings collectively highlight the critical role of content clarity and actionability, the limitations of engagement metrics as quality proxies, and the need for better content oversight regardless of uploader credentials.

Stepwise regression identified understandability (PEMAT-U) as the strongest independent predictor of overall quality (GQS), followed by actionability (PEMAT-A), video duration, and uploader type. Engagement metrics and platform did not enter the final model, further confirming that popularity and platform differences are unreliable proxies for content quality.

This finding aligns with 2 recent studies. A study on adolescent sexuality education also found PEMAT-U to be the strongest positive predictor of GQS (β=.485), surpassing actionability, video duration, and uploader type [37]. Similarly, an analysis of uterine fibroid videos reported that the “completeness score,” “source,” and “PEMAT scores” were significant predictors of video quality, confirming that content characteristics outweigh simple popularity or uploader identity in determining educational value [38].

Our finding that engagement metrics were not significant predictors aligns with the “credibility-evidence gap” documented in a broad evaluation of physician-generated health claims [39]. In that study, 62.5% of medical claims rested on “very low” or “no evidence,” yet low-evidence videos received 34.6% more views than high-evidence ones, directly mirroring our observation that engagement metrics do not predict quality. This evidence suggests that current algorithm recommendations, which place substantial weight on user engagement, may unintentionally prioritize low-quality content. Consequently, platform algorithms should be adjusted to promote comprehensible, actionable content rather than simply rewarding videos that generate high volumes of likes, comments, or shares.

Misinformation has been reported in health-related videos on platforms such as YouTube and TikTok across various conditions, including atopic dermatitis, ophthalmology, and oral health [14,15,40]. Similarly, a YouTube-based study on AH found that 21.5% of videos were misleading, with the vast majority uploaded by nonprofessional sources [8].

Consistent with prior reports [8], most misleading videos in our dataset came from nonprofessional creators. However, even verified professional accounts can occasionally produce misleading content. One possible explanation is that some clinicians use simplified or absolute language to increase engagement or that traditional Chinese medicine promotions sometimes blur the boundary between evidence-based and unverified claims. Thus, misleading information can originate from both nonprofessional and professional sources, and because viewers trust verified accounts, misinformation from the latter is especially concerning.

Short-video platforms have become a common source of health information for parents of children with conditions such as AH. Our findings indicate that content quality is generally low, and neither professional credentials nor engagement metrics reliably predict actual quality. In fact, understandability was the strongest independent predictor of overall quality, and misleading videos appeared from both nonprofessional and verified professional sources.

These results point to several actionable directions. For clinicians and health educators, improving the clarity and actionability of video content should take priority over simply displaying professional titles. Oversimplified or absolute statements—even from experts—can be misleading. For platform developers, the weak correlation between engagement metrics and quality suggests that ranking algorithms based heavily on likes, shares, or comments may inadvertently promote low-quality content. Incorporating evidence-based indicators such as content completeness and understandability into ranking systems would be a more effective strategy. For regulators, account verification alone is insufficient. Content-based review mechanisms are essential to counter misinformation, regardless of whether the uploader is a medical professional.

This study has several limitations. First, it only included Chinese short videos related to AH on Bilibili, TikTok, and Xiaohongshu. This leads to selection bias, which limits the external validity of the results and makes them unrepresentative of content quality on other platforms and in different cultural contexts. Second, although 4 validated scales were used in the study, the inherent flaws and systematic biases of these scales cannot be completely avoided. Third, this study was a cross-sectional survey conducted at a specific time point. As online short-video content is frequently updated, edited, and deleted, the research conclusions may lack long-term stability. Future research can expand the sampling scope to cover more platforms and multilingual groups. It may also be feasible to adopt automated content analysis methods and longitudinal research designs to further explore the impact of online health information on users’ health cognition, behavioral habits, and long-term health outcomes.

This study evaluated the overall quality of educational videos related to AH on Bilibili, TikTok, and Xiaohongshu. The results showed that the overall quality of such short videos was relatively low. Videos on Bilibili had higher content completeness, while Xiaohongshu performed better than TikTok in actionability. Interaction metrics (likes, shares, and comments) exhibited weak or even no correlation with video content quality. This implies that current platform algorithms may inadvertently promote the spread of low-quality educational content. Stepwise regression analysis indicated that understandability (PEMAT-U) was the strongest predictor of overall video quality, followed by actionability, video duration, and uploader type. These findings suggest that easy-to-understand content with practical guiding value matters more than creators’ professional qualifications or video length. Misleading information was identified in videos released by both nonprofessional creators and medical professionals. Simple account verification alone cannot ensure the authenticity and reliability of online educational content. On the basis of the findings, medical practitioners are advised to produce popular science content that is easy to understand and practically informative, short-video platforms should optimize recommendation algorithms that overly depend on interaction data, and relevant regulatory departments should improve the review and supervision systems focused on content quality.