基于大模型的多阶段出院小结质控方法研究

Research on Quality Control Method of Multi-stage Discharge Summary Based on Large Model

目的为提高医生书写医疗文书的质量、效率和一致性,利用大模型探索多阶段出院小结质量节点自动生成方法,为其质控奠定基础。方法梳理出院小结内容与其他电子病历数据内容的字段映射关系,设计基于大模型ChatGLM2-6B的多阶段质控模板生成方案,并构建面向出院小结的质量节点自动生成任务数据集,采用指令数据构建、P-Tuning v2微调、Prompt构建等技术,训练并调优该模型。结果基于大模型的多阶段质控模板生成方法,能够有效捕获其他电子病历数据的关键信息,自动生成格式规范且结构完整的出院小结模板内容,并保证生成结果连贯、可读,满足相关质控要求。结论根据医疗文书质控要求,利用大模型自动生成医疗文书质控模板是可行的。此模型的设计可为同类应用提供技术框架,以推进医疗信息技术的发展。

Objective In order to improve the quality,efficiency and consistency of doctors'written medical documents,we explored the use of large model technology to build a multi-stage discharge summary quality node automatic generation method to lay the foundation for its quality control.Methods Sort out the field mapping relationship between discharge summary content and other electronic medical record data content,design a multi-stage quality control template generation scheme based on ChatGLM2-6B,and build a quality node for patient discharge summary to automatically generate task data sets,using command data construction,P-Tuning v2 fine-tuning,Prompt construction and other technologies are used to train and tune the model.Results The multi-stage quality control template generation method based on large models can effectively capture key information of other electronic medical record data,automatically generate discharge summary template content that conforms to format specifications and has a complete structure,and ensures that the generated results are coherent and readable,meeting relevant quality control requirements.Conclusion According to the medical document quality control requirements,it is feasible to use large model technology to automatically generate medical document quality control templates.This model design can provide a technical framework for the construction of similar medical large models and jointly promote the development of future medical information technology.