Feasibility of large language models for assessing and coaching surgeons’ non-technical skills

摘要

This study demonstrates Large Language models (LLMs) to assess and coach surgeons on their non-technical skills, traditionally evaluated through subjective and resource-intensive methods. Llama 3.1 and Mistral effectively analyzed robotic-assisted surgery transcripts, identified exemplar and non-exemplar behaviors, and autonomously generated structured coaching feedback to guide surgeons' improvement. Our findings highlight the potential of LLMs as scalable, data-driven tools for enhancing surgical education and supporting consistent coaching practices.