大语言模型病历质控与病程记录生成评估方法研究  

The Research on the Evaluation Method of Medical Record Quality Control and Progress Note Generation Using Large Language Models

作　　者：周文粲 陈洁[1] 冯艳芳 刘伟佳 刘丽红[1] ZHOU Wencan;CHEN Jie;FENG Yanfang;LIU Weijia;LIU Lihong(Peking University People’s Hospital,Beijing 100032,China)

机构地区：[1]北京大学人民医院,北京市100032

出　　处：《中国卫生信息管理杂志》2025年第2期163-170,共8页Chinese Journal of Health Informatics and Management

摘　　要：目的探索大语言模型在病历内涵质控与病程记录生成任务中的应用效果,通过合理的实验设计与评价指标,验证其在医疗场景中的技术可行性,研究有效的评估方法。方法设计包含内涵质控和病程记录生成任务的验证集,确保数据多样性与实际应用场景匹配。内涵质控任务通过思维链优化构建提示词,通过对比生成结果与专家质控结果计算准确率;病程记录生成任务采用小样本学习机制生成内容,由专家修正后基于ROUGE、BERTScore指标及关键信息覆盖率进行综合评估。结果基于验证集的评估结果表明,大语言模型在内涵质控任务中准确率较高,在病程记录生成任务中能够有效生成符合医疗文书规范要求的病历文本。结论本研究验证了大语言模型在电子病历应用中的可行性,结合有效的评估体系,为智慧医疗场景下的AI落地和发展提供了参考框架和实证支持。Objective To explore the application of large language models in the tasks of medical record quality control and progress note generation,and to verify their technical feasibility and research effective evaluation methods through reasonable experimental design and evaluation indicators.Methods A validation set containing tasks for both EMR quality control and progress note generation was designed to ensure data diversity and alignment with real-world clinical scenarios.EMR quality control task utilized chain-of-thought optimization for prompt construction,with accuracy evaluated by comparing generated results against expert-reviewed quality control outcomes.The progress note generation task employed a few-shot learning mechanism,where generated content was revised and scored by experts.Comprehensive evaluation are conducted based on ROUGE,BERTScore metrics,and key information coverage rate.Results The evaluation based on the validation set shows that the large language model ensured a high accuracy in the EMR quality control task and can effectively generate progress notes that complies with the standards of medical documentation.Conclusion This study verifies the feasibility of the large language model in the application of electronic medical records,and combined with an effective evaluation system,it provided a new technical reference framework and empirical support for the implementation and development of AI in intelligent medical scenarios.

关 键 词：大语言模型 内涵质控 病程记录 人工智能 电子病历 

分 类 号：R197.323.1[医药卫生—卫生事业管理] R319[医药卫生—公共卫生与预防医学]