Quality, reliability, and transparency of algorithm-exposed top acute leukemia-related videos on Douyin and Bilibili

Short-video platforms increasingly disseminate health information. This study evaluated the content, quality, reliability, and transparency of acute leukemia (AL)-related videos on Douyin and Bilibili. On August 27, 2025, algorithm-ranked search results were sequentially screened, and the first 100 eligible videos from each platform were included, reflecting prominently exposed eligible content rather than the full AL-related video population. Video characteristics, engagement metrics, uploader identity, and content categories were recorded. Quality, reliability, and transparency were assessed using the Global Quality Score (GQS), modified DISCERN, and Journal of the American Medical Association (JAMA) benchmark criteria. Group differences were analyzed using nonparametric tests, and associations between engagement and quality were examined using Spearman correlation and multivariable ordinal logistic regression. Inter-rater reliability was assessed using weighted and unweighted Cohen’s kappa. A total of 200 videos were analyzed. Median scores were 3.00 for GQS, 1.50 for modified DISCERN, and 1.50 for Journal of the American Medical Association. Clinical manifestations and treatment were commonly covered, whereas prevention and etiology were underrepresented. Videos uploaded by blood disease-related experts had significantly higher quality and reliability scores. On Douyin, engagement metrics were negatively associated with GQS in correlation analyses, but this association was no longer significant after adjustment for uploader type, video duration, and days since upload (adjusted odds ratio = 0.97, 95% confidence interval: 0.92–1.02, P = .18). Inter-rater reliability was substantial to almost perfect. Algorithm-exposed AL-related videos on Douyin and Bilibili showed suboptimal quality and reliability. Uploader identity, rather than engagement, was more closely associated with information quality. Findings are limited to prominently exposed videos retrieved on one date.