Do ChatGPT and Gemini’s Recommendations Align With Established Guidelines for Hand and Upper Extremity Surgery?

Abstract

BACKGROUND: The use of large language models (LLMs) such as ChatGPT and Gemini in clinical settings has surged, presenting potential benefits in reducing administrative workload and enhancing patient communication. However, concerns about the clinical accuracy of these tools persist. This study evaluated the concordance of ChatGPT and Gemini's recommendations with American Academy of Orthopedic Surgeons (AAOS) clinical practice guidelines (CPGs) for carpal tunnel syndrome, distal radius fractures, and glenohumeral joint osteoarthritis. METHODS: ChatGPT (version 4o) and Gemini (version 1.5 Flash) were queried using structured text-based prompts aligned with AAOS CPGs. The LLMs' outputs were analyzed by blinded reviewers to determine concordance with the guidelines. Concordance rates were compared across models, topics, and guideline strength using descriptive statistics and McNemar's test. The transparency of responses, including source citation, was also assessed. RESULTS: < .0001). CONCLUSIONS: Despite modest concordance rates, both models exhibited significant limitations, including variability across topics and guideline strengths, as well as insufficient citation transparency. These findings highlight the challenges in integrating LLMs into clinical practice and emphasize the need for further refinement and evaluation before adoption in hand surgery.

Ähnliche Arbeiten