随着大语言模型（large language models,LLMs）在医学诊断、科研与教育等领域的广泛应用，其卓越的生成与推理能力已显著展现优势。然而，医学领域对伦理、隐私保护及模型准确性的严格标准，也使LLMs的实际应用面临严峻挑战。尽管个体预后或诊断的多变量模型透明报告(transparent reporting of a multivariable prediction model for individual prognosis or diagnosis，TRIPOD)+人工智能（artificial intelligence，AI）为预后或诊断预测模型提供了报告规范，但其在生成式人工智能研究中的适用性仍显不足。本文解读了在TRIPOD+AI基础上扩展形成的TRIPOD-LLM报告指南，系统梳理了其在模型构建、验证、任务适应性及人类监督等方面的报告要素，为提升医学领域LLMs研究的透明度、规范性与可复现性提供了参考。

With the widespread application of large language models (LLMs) in medical diagnosis,scientific research,education,and other fields,their outstanding generative and reasoning capabilities have demonstrated significant advantages.However,the strict standards for ethics,privacy protection,and model accuracy in the medical field have also posed severe challenges to the practical application of LLMs.Although the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) + Artificial Intelligence (AI) provides reporting norms for prognosis or diagnosis prediction models,its applicability in generative artificial intelligence research is still insufficient.This article interprets the TRIPOD-LLM reporting guideline,which is extended based on TRIPOD+AI.It systematically sorts out its reporting elements in terms of model construction,validation,task adaptability,and human supervision,providing an important reference for improving the transparency,standardization,and reproducibility of LLM research in the medical field.

R47

中央高水平医院临床科研业务（2022-PUMCH-B-130）