A Review of Large Language Models (LLMs) for Surgical Training and Evaluations of Performance

This review examines large language models (LLMs) in surgical training, education, and performance evaluation. A literature search across PubMed, Cochrane Library, Scopus, and IEEE Xplore identified 23 relevant studies from 1,159 records. The review shows that LLMs can support surgical knowledge assessment, procedural reasoning, intraoperative decision-making, postoperative feedback, patient communication, and simulation-based education. Despite this, important limitations remain, including hallucinations, privacy risks, bias, and inconsistent clinical reliability. Future research may investigate multi-institutional studies, standardized benchmarks, multimodal integration, and rigorous evaluation of safety in surgical training.