Abstract
Introduction:
Craniofacial injuries from racquetball sports in the United States remain high, despite guidelines for protective equipment. While national injury data describe injury patterns in squash, badminton, and tennis, they fail to provide actionable, age- and sport-specific management, treatment, or preventive strategies.
Methods:
Using the National Electronic Injury Surveillance System (NEISS) database, this study evaluated ChatGPT-4o’s ability to risk-stratify patients and provide preventive strategies based on demographic characteristics, injury mechanisms, and patient needs. Standardized clinical vignettes reflected craniofacial injuries requiring stratification and counseling. AI-generated responses were scored using the validated DISCERN criteria by 2 board-certified plastic surgeons. Readability was evaluated with the Flesch-Kincaid grade level, and specificity was rated on a 5-point Likert scale by 2 independent medical student reviewers.
Results:
DISCERN score was 32.5/75, with a mean reliability score of 2.9/5, and a treatment quality score of 1.4/5. Readability averages an 11th-grade readability across sports. Specificity ratings indicated moderately high specificity (3.9–4/5).
Discussion/Conclusion:
While ChatGPT4-o can provide accessible, structured information, its performance in this study demonstrated moderate reliability, low treatment guidance quality, a reading level above AMA recommendations, and moderately high specificity. These findings underscore the need for cautious integration of AI tools in patient education and clinical decision-making. As LLMs evolve, there is potential for risk stratification and injury prevention tools to improve. Careful development and validation will be integral to ensure safe and effective clinical use, as well as HIPAA compliance, lack of bias, and accurate information.