The Reliability and Quality of Videos as Guidance for Gastrointestinal Endoscopy: A Cross-sectional Study (Preprint)

BACKGROUND Gastrointestinal endoscopy represents a useful tool for the diagnosis and treatment of gastrointestinal diseases. Video platforms for spreading endoscopy-related knowledge may help patient understand the pros and cons of endoscopy on the premise of ensuring accuracy. However, videos with misinformation may lead to adverse consequences. Therefore, this study aims to evaluate the quality of gastrointestinal endoscopy-related videos on YouTubeTM and to assess whether large language models (LLMs) can help patients obtain information from videos more efficiently. OBJECTIVE This study aims to evaluate the quality of gastrointestinal endoscopy-related videos on YouTubeTM and to assess whether large language models (LLMs) can help patients obtain information from videos more efficiently. METHODS Methods: We collected information from YouTubeTM videos about three commonly used gastrointestinal endoscopes (gastroscopy, colonoscopy and capsule endoscopy) and assessed their quality (rated by the modified DISCERN Tool, mDISCERN), reliability (rated by JAMA), and recommendation (rated by Global Quality Score, GQS). We tasked LLM with summarizing the video content and assessed it from three perspectives: accuracy, completeness and readability. RESULTS A total of 167 videos were included. According to the indicated scoring, the quality, reliability and recommendation of the three gastrointestinal endoscopy-related videos on YouTubeTM were overall unsatisfactory, and the quality of the videos released by patients was particularly poor. Capsule endoscopy yielded significantly lower GQS than did gastroscopy and colonoscopy. LLM-based summaries yielded accuracy scores of 4 (4-5), completeness scores of 4 (4-5), and readability scores of 2 (1-2). CONCLUSIONS The quality of gastrointestinal endoscope-related videos currently on YouTubeTM is poor. Moreover, additional regulatory and improvement strategies are needed in the future. LLM may be helpful in generalizing video-related information, but there is still room for improvement in its ability.