This study adopted a cross-sectional design. A comprehensive video search was performed on October 15, 2025, on the TikTok platform (Douyin, the mainland Chinese version). To minimize recommendation bias caused by platform algorithms, browser history and cookies were cleared before the search, which was conducted without logging into any user account. The following Chinese keyword was used: “年龄相关性黄斑变性.” The personalized recommendation filter was disabled during the search to obtain the most objective results possible.

To ensure analytical feasibility while capturing the most relevant and frequently recommended content for this keyword, we predefined the sample and limited screening to the first 200 videos ranked under “comprehensive sorting.” This sample size balances coverage of the AMD-related information space on TikTok with the workload of duplicate rating and detailed scoring for each clip. After preliminary screening of these 200 videos, those meeting the eligibility criteria were included in the final analysis.

The inclusion criteria were as follows: (1) the video content was directly related to the etiology, symptoms, diagnosis, treatment, or prevention of AMD; (2) the video was presented in Chinese; and (3) the video was fully accessible for playback. The exclusion criteria were as follows: (1) videos irrelevant to the topic; (2) duplicate content; (3) videos of insufficient quality for evaluation; (4) advertisements created solely for product promotion; and (5) user comments or non–primary video content. Two researchers with medical backgrounds independently screened the videos, and any discrepancies were resolved through discussion with a third senior investigator who made the final decision.

For each included video, a standardized data extraction form was designed, and information was independently collected by two researchers. The extracted data included: (1) basic video characteristics, such as URL, upload date, duration, number of likes, comments, favorites, and shares; and (2) information on the video uploader. Based on the uploader’s profile, video content, and verification status, uploaders were categorized into four groups:

In practice, accounts were coded as Non-Profit when the profile explicitly represented an institution (e.g., hospital or health agency name in the account title, official logo, institutional verification badge, or descriptions such as “official account of XX Hospital”). Accounts were coded as Medical when they primarily highlighted a named clinician and used personal pronouns (e.g., “I,” “my clinic”), even if a hospital affiliation was listed in the profile. In the Chinese context, public hospitals and governmental agencies were treated as Non-Profit organizations, whereas private clinics and commercial health companies were categorized as For-Profit. Ambiguous accounts were jointly discussed by the two raters, and disagreements were resolved by a third senior ophthalmologist.

To ensure the scientific rigor of the evaluation, this study adopted the Age-Related Macular Degeneration Preferred Practice Pattern issued by the American Academy of Ophthalmology (AAO) as the content evaluation standard (21). The content completeness was assessed according to the educational video evaluation framework proposed by Goobie et al. (22), covering six core aspects of AMD: (1) definition, (2) classification, (3) symptoms, (4) risk factors, (5) diagnostic methods, and (6) treatment and management.

Video quality was evaluated using two validated tools: The DISCERN instrument, which consists of 16 questions, is designed to assess the quality of consumer health information regarding treatment options and is considered the gold standard for evaluating the reliability of health information. Each question is rated on a 5-point scale, with a total score ranging from 16 to 80, where higher scores indicate better quality (23, 24). The Patient Education Materials Assessment Tool for Audiovisual Materials (PEMAT-A/V) consists of 17 items divided into two domains: Understandability and Actionability, rated as “agree,” “disagree,” or “not applicable” (25, 26). Prior to the formal evaluation, all raters received standardized training. A random sample of 20 videos was selected for a pilot evaluation to ensure inter-rater reliability, and the intraclass correlation coefficient (ICC) among raters was subsequently calculated.

All statistical analyses were performed using IBM SPSS Statistics version 27.0 (IBM Corp., Armonk, NY, United States). The Shapiro–Wilk test was employed to examine the normality of data distributions. Continuous variables were summarized as mean ± standard deviation (mean ± SD) for approximately normally distributed variables and as median (P25, P75) for non-normally distributed variables. Inter-rater consistency between the two evaluators was assessed using the intraclass correlation coefficient (ICC). Differences in quality scores among the four uploader groups were first examined using one-way analysis of variance (ANOVA). When the homogeneity of variance assumption was violated, Welch’s ANOVA or the Kruskal–Wallis test was used instead. Where global tests were statistically significant, pairwise post hoc comparisons were conducted using Bonferroni-adjusted t-tests (for ANOVA) or Dunn–Bonferroni tests (for Kruskal–Wallis) to control for type I error. Correlations between video characteristics and quality scores were examined using Pearson’s correlation coefficient for approximately normally distributed variables and Spearman’s rank correlation coefficient for non-normally distributed variables. In particular, engagement metrics, which showed markedly skewed distributions, were analyzed using Spearman’s rank correlations. All statistical tests were two-sided, and a P < 0.05 was regarded as statistically significant.

A total of 200 videos were initially retrieved through the preliminary search. After removing three duplicate videos, 197 videos remained for preliminary screening. Screening based on titles and video content led to the exclusion of one video that was either unrelated to the topic of AMD or purely promotional. Ultimately, a total of 196 videos were included in the final analysis.

Among the 196 included videos, the distribution of video uploaders was as follows: The Medical group uploaded the largest number of videos (n = 163, 83.2%), followed by the Non-Profit group (n = 15, 7.7%), and both the Non-Medical and For-Profit groups contributed nine videos each (4.6%). Details are shown in Figure 1.

Regarding engagement metrics, the mean number of likes was 2762.04 ± 10845.78, with a median of 325.00 (101.00, 996.25). The mean number of comments was 157.25 ± 561.55, with a median of 20.00 (8.00, 66.25). The mean number of favorites was 1113.52 ± 5106.81, with a median of 104.00 (34.25, 379.25). The mean number of shares was 1296.00 ± 7428.63, with a median of 68.50 (16.00, 287.00), and the mean video duration was 118.14 ± 146.46 s, with a median of 71.00 (47.00, 131.50). Details are provided in Table 1.

We analyzed the thematic content of 196 included videos, noting that many of them covered multiple topics simultaneously. As shown in Figure 2 and Table 2, the most frequently mentioned topic was Management (n = 151, 22.6%), followed by Symptoms (n = 146, 21.9%). In contrast, Classification (n = 73, 10.9%) and Diagnosis (n = 78, 11.7%) were the least frequently addressed. Definition and Risk Factors accounted for 18.6% (n = 124) and 14.4% (n = 96), respectively. These findings indicate that the videos tend to focus more on symptoms and management, which are of primary concern to patients, while professional definitions and diagnostic aspects are less frequently discussed.

Regarding content sources, the Medical group was the predominant contributor across all thematic categories. For example, within the most popular Management topic, 126 videos (83.44%) originated from the Medical group, far exceeding other sources. Similarly, the Medical group also played a dominant role in the Symptom and Definition categories. The Non-Medical group showed a relatively sparse distribution of video content across topics. Videos from the For-Profit group primarily focused on Management, accounting for 77.78% of their uploads, while the Non-Profit group covered all six thematic aspects, with a minimum coverage rate of 66.67%, as shown in Figure 3.

For Reliability, Treatment choice, DISCERN tool scores, Overall quality score, Understandability, and Actionability, the ICCs between two raters were 0.904, 0.843, 0.839, 0.947, 0.913, and 0.869, respectively (P < 0.001) (Figure 4). The mean Reliability score for all included videos was 24.88 ± 4.39, with a median of 25.00, while the mean Treatment choice score was 19.58 ± 3.02, with a median of 20.00. The median Overall quality score of the videos was 3.00, with a mean of 2.96 ± 0.93, indicating a moderate overall quality. The DISCERN tool score had a median of 48.00 and a mean of 47.42 ± 7.27, also within the moderate range. However, the videos performed well in terms of Understandability (median = 84.62%), but scored slightly lower in Actionability (median = 75.00%) (Table 1). Significant differences among groups were observed in Reliability, Treatment choice, and DISCERN tool scores (P < 0.05) (Figures 5A,B,D). For Overall quality score, no significant difference was found between the Non-Medical and For-Profit groups (P = 0.762), whereas other intergroup comparisons revealed significant differences (P < 0.001) (Figure 5C).

In the intergroup comparison of Understandability, significant statistical differences were observed between the Medical and Non-Medical groups (P = 0.002), the Medical and Non-Profit groups (P < 0.001), the Non-Medical and Non-Profit groups (P < 0.001), and the For-Profit and Non-Profit groups (P < 0.001), whereas no significant difference was found between the Non-Medical and For-Profit groups (P = 0.192) (Figure 5E). In terms of numerical trends, the Non-Profit group exhibited the highest Understandability scores, followed by the Medical group, while the For-Profit group showed the lowest scores.

Regarding Actionability, the Non-Profit group again achieved the highest scores, followed by the Medical group, with the For-Profit group scoring the lowest. In the intergroup comparison of Actionability, significant differences were also found between the Medical and Non-Medical groups (P = 0.036), the Medical and For-Profit groups (P = 0.004), the Medical and Non-Profit groups (P = 0.019), the Non-Medical and Non-Profit groups (P < 0.001), and the For-Profit and Non-Profit groups (P < 0.001), whereas no significant difference was detected between the Non-Medical and For-Profit groups (P = 0.667) (Figure 5F).

Significant differences in video quality were observed across different uploader types. We compared video quality among four categories of uploaders. Results showed that videos uploaded by Non-Profit organizations achieved the highest scores across all quality dimensions, with a mean Reliability score of 32.13 ± 1.85, mean Treatment choice score of 23.13, median Overall quality score of 4.00, mean DISCERN tool scores of 59.13 ± 2.50, median PEMAT-A/V Understandability of 84.62%, and median PEMAT-A/V Actionability of 75.00%, all significantly higher than those of other groups (P < 0.001). The Medical group, as the dominant uploader category, ranked second in overall video quality, with generally above-average scores: median Reliability 25.00, median Treatment choice 20.00, median Overall quality score 3.00, median DISCERN tool scores 48, median Understandability 84.62%, and median Actionability 75.00%. In contrast, videos from the Non-Medical and For-Profit groups demonstrated the lowest quality. Notably, the For-Profit group had a median Overall quality score of only 1.00 and a median DISCERN tool score of just 22.00 (Table 3).

Regarding the proportion of Reliability scores, the Non-Profit and Medical groups accounted for 32.9 and 26.0%, respectively, whereas the Non-Medical and For-Profit groups contributed less, at only 18 and 11.1%. For Treatment choice, Non-Profit and Medical groups accounted for 30.3 and 26.4%, whereas Non-Medical and For-Profit groups accounted for only 17.3 and 14.0%. For Understandability, the Non-Profit and Medical groups contributed 25.4 and 23.1%, whereas the Non-Medical and For-Profit groups contributed only 20.6 and 18.9%. For Actionability, the Non-Profit and Medical groups accounted for 26.6 and 23.0%, while the Non-Medical and For-Profit groups contributed only 20.0 and 18.5%. Across these four dimensions, the Non-Profit and Medical groups consistently achieved higher performance levels (Figure 6).

Significant positive correlations were generally identified among all video quality evaluation metrics (P < 0.05). DISCERN tool scores exhibited a strong positive correlation with Reliability (r = 0.78, P < 0.05) and moderate correlations with both the Overall quality score (r = 0.51, P < 0.05) and Treatment choice (r = 0.51, P < 0.05). This suggests that the DISCERN tool maintains strong alignment with the Reliability, Overall quality score, and Treatment choice domains when assessing the quality of information. Reliability demonstrated a moderate positive correlation with the Overall quality score (r = 0.30, P < 0.05), implying that higher reliability is generally accompanied by a higher overall quality rating; however, this relationship was weaker than the correlation between the DISCERN tool and Reliability. PEMAT-A/V revealed a weak correlation between Understandability and Actionability (r = 0.17, P < 0.05). Both metrics showed only weak-to-moderate positive correlations with DISCERN tool scores (r = 0.20 and r = 0.28, P < 0.05), suggesting that although PEMAT-derived Understandability and Actionability are related to DISCERN tool scores to some extent, their associations are weaker than those observed between DISCERN and other quality dimensions (Figure 7).

Reliability was moderately and positively associated with user engagement metrics, including likes, comments, favorites, and shares (r = 0.48–0.54, P < 0.05). Similarly, DISCERN tool scores were moderately correlated with these engagement metrics (r = 0.40–0.44, P < 0.05). However, correlations between PEMAT-A/V Understandability and Actionability and engagement metrics were relatively weak. Video duration showed a mild positive correlation with likes, comments, favorites, and shares (r = 0.23–0.30, P < 0.05). Moreover, video duration demonstrated weak positive correlations with Reliability (r = 0.26) and DISCERN tool scores (r = 0.18) (P < 0.05), but was not significantly correlated with Treatment choice (r = −0.051), Overall quality score (r = −0.055), Understandability (r = −0.073), or Actionability (r = −0.045) (P > 0.05) (Figure 7).