基于大模型的DRG医保结算清单智能生成系统探索与应用  

Exploration and Application of an Intelligent DRG Medical Insurance Settlement Form Generation System Based on Large Language Models

作　　者：刘晓坤[1] 朱卫国[1] 李乃适[1] 崇伟峰 韩丁[1] LIU Xiaokun;ZHU Weiguo;LI Naishi;CHONG Weifeng;HAN Ding(Peking Union Medical College Hospital,Beijing 100730,China)

机构地区：[1]中国医学科学院北京协和医院,北京市100730 [2]云知声智能科技股份有限公司,北京市100096

出　　处：《中国卫生信息管理杂志》2025年第1期20-25,共6页Chinese Journal of Health Informatics and Management

基　　金：中国医学科学院医学与健康科技创新工程重大协同创新项目“医学知识管理与智能化知识服务关键技术研究”(2021-I2M-1-056);国家自然科学基金项目“数智技术赋能多层次医疗保障下的公立医院运营机制”(72441018)。

摘　　要：目的研究基于大语言模型的医保结算清单智能生成系统,以解决传统医保结算清单填报耗费人力、缺乏专业人才、填报质量不高等痛点,从而提高工作效能。方法以定制化大语言模型为基座,通过继续预训练、指令微调和人类对齐等手段构建医保结算清单智能生成系统,利用思维链、搜索增强、样例学习等技术,实现病历语义理解、国际疾病分类(ICD)编码匹配,以及主诊断、主手术和其他诊断、其他手术自动选择,生成医保结算清单,并经人工审核优化性能。结果某医院1322份疾病诊断相关分组(DRG)结算病例验证结果显示,智能生成清单的入组准确率达到93.6%,逐步接近人类编码专家入组准确率98.6%。结论基于大模型的医保结算清单智能生成系统显著提升了清单生成效能,降低了人力物力等成本。未来,该系统在医保支付方式改革和基金监管中具有广泛应用前景,可为医院、医保、医政等部门提供强有力的技术支撑。Objective This study aims to address key challenges in traditional medical insurance settlement form completion,including high labor costs,shortage of specialized personnel,and low data quality.It explores an intelligent settlement form generation system based on large language models(LLMs)to improve operational efficiency.Methods A customized large language model was developed as the foundation,utilizing continued pretraining,instructionfine-tuning,and human alignment techniques to construct a medical insurance-specific model.Advanced methodologies such as chain-of-thought reasoning,search augmentation,and example-based learning were employed to enable semantic understanding of medical records,ICD code matching,and automated selection of primary diagnoses,primary procedures,and secondary diagnoses and procedures.The generated settlement forms were further refined through human review to optimize performance.Results Validation using 1,322 DRG settlement cases from a hospital demonstrated that the intelligent system achieved a grouping accuracy of 93.6%,gradually approaching the human coding expert's accuracy rate of 98.6%for enrollment.Conclusion The intelligent settlement form generation system based on LLMs significantly improves the efficiency of form generation and reduces labor and resource costs.This system shows broad application potential in medical insurance payment reforms and fund supervision,providing robust technical support for hospitals,insurance agencies,and health administration departments.

关 键 词：大语言模型 DRG支付 医疗保障基金结算清单 智能生成 

分 类 号：R-39[医药卫生] R319