基于神经网络模型的电子病历文本提取及质量缺陷分析  

Text Extraction and Quality Defect Analysis of Electronic Medical Record Based on TextCNN

作　　者：曹新志 沈君姝 袁雪 王泽川 胡欣 马存宁 王至诚 王杰[1] CAO Xinzhi;SHEN Junshu;YUAN Xue;WANG Zechuan;HU Xin;MA Cunning;WANG Zhicheng;WANG Jie(Information Center,Jiangsu Province Hospital on Integration of Chinese and Western Medicine,Nanjing Jiangsu 210028,China;Department of Equipment,Jiangsu Province Hospital on Integration of Chinese and Western Medicine,Nanjing Jiangsu 210028,China)

机构地区：[1]江苏省中西医结合医院信息中心,江苏南京210028 [2]江苏省中西医结合医院设备科,江苏南京210028

出　　处：《中国医疗设备》2022年第4期98-101,共4页China Medical Devices

基　　金：江苏省医院协会医院管理创新研究课题(JSYGY-3-2021-483,JSYGY-3-2020-505)。

摘　　要：目的探讨通过神经网络模型分析电子病历文本提取及质量缺陷的效果。方法抽取我院2020年1月1日至2021年10月31日归档的电子病历9万份,其中门诊病历7万份,住院病历2万份。使用疾病诊断相关分类系统(Diagnosis Related Groups,DRGs)对病历分组,统计提取及质量缺陷,用自然语言提取临床有用信息,运用用于文本分类任务的卷积神经网络(Text Convolutional Neural Networks,TextCNN)构建出质量缺陷分类模型。结果纳入研究的9万份电子病历中,有16685份电子病历存在质量缺陷,质量缺陷发生率是18.54%。人工复检发现存在质量缺陷病历问题主要包括对病情和医嘱变化或检查报告等病程记录不全、部分检查报告缺失不全等。TextCNN模型的准确率、敏感度、特异度、阳性及阴性预测值均高于人工复审(P<0.05)。采用TextCNN模型分析后DRGs筛选出问题病案发生率显著下降(χ^(2)=16.830,P<0.05)。结论神经网络模型在电子病历文本提取和筛查质量缺陷病历中的效果较好,可用于临床推广以提高病历质量。Objective To explore the effect of text extraction and quality defect analysis of electronic medical record based on neural network model.Methods A total of 90000 electronic medical records filed in our hospital from January 1,2020 to October 31,2021 were selected,including 70000 outpatient medical records and 20000 inpatient medical records.Diagnosis related groups(DRGs)system was used to group medical records,statistical extraction and quality defects.Natural language was used to extract clinical useful information,and text convolutional neural network(TextCNN)was used to build quality defect classification model.Results Among the 90000 electronic medical records included in the study,16685 electronic medical records had quality defects,and the incidence of quality defects was 18.54%.After manual reexamination,the problems of medical records with quality defects mainly include incomplete records of changes in the condition and doctor’s orders,incomplete filling of inspection reports,etc.The accuracy,sensitivity,specificity,positive and negative predictive values of TextCNN model were higher than those of manual review(P<0.05).After the analysis of TextCNN model,the incidence of problem cases screened by DRGs decreased significantly(χ^(2)=16.830,P<0.05).Conclusion The neural network model has a good effect on the text extraction and quality defect screening of electronic medical record,and can be used in clinical promotion to improve the quality of medical record.

关 键 词：神经网络模型 电子病历 文本提取 质量缺陷 

分 类 号：R197.3[医药卫生—卫生事业管理]