基于病案首页的DRG低风险组患者出院风险预测模型研究  

Study on Hospital Discharge Risk Prediction Model of DRG Low-Risk Group Patients Based on the Home Page of Medical Records

作　　者：袁筱祺 高玮 YUAN Xiaoqi;GAO Wei(Medical Department,Shanghai First People’s Hospital,Shanghai 200080,China;Shanghai Eye Disease Prevention Center,Shanghai 200040,China)

机构地区：[1]上海市第一人民医院医务处,上海200080 [2]上海市眼病防治中心,上海200040

出　　处：《中国医疗设备》2025年第3期96-101,共6页China Medical Devices

基　　金：国家青年基金项目(82203742);上海交通大学中国医院发展研究院医院管理建设项目(CHDI-2019-B-14)。

摘　　要：目的探讨疾病诊断相关分组(Diagnosis Related Group,DRG)中低风险组患者死亡的独立危险因素,利用径向基神经网络模型构建风险预测模型,以期降低DRG低风险组患者死亡率,提高院内患者医疗安全质量。方法选取上海市某三甲医院2023年1—8月50344条病案首页数据,根据出院情况分为治愈组、未愈组和死亡组,通过单因素分析筛选出重要风险因素,作为径向基神经网络的分析变量,构建风险预测模型。采用曲线下面积(Area Under Curve,AUC)及模型预测准确度、敏感度及特异性等评估模型的预测效能。结果径向基神经网络模型整体预测准确度为98.69%。其中,患者出院情况为治愈的AUC=0.829,95%CI:0.826~0.832,约登指数最大值为0.5823,敏感度为76.29%,特异性为82.06%。患者出院情况未愈的未愈的AUC=0.825,95%CI:0.822~0.828,约登指数最大值为0.5779,敏感度为76.84%,特异性为80.95%。患者出院情况死亡的AUC=0.600,95%CI:0.596~0.605,约登指数最大值为0.2009,敏感度为44.99%,特异性为75.10%。结论对于DRG低风险组患者风险客观化预测中,径向基神经网络模型中治愈模型的预测性能较优。输血反应、患者年龄、住院天数、住院总费用为影响DRG低风险组出院情况的重要独立危险因素,本研究的模型可以为DRG低风险组患者出院情况恶化和干预提供理论依据。Objective To investigate the independent risk factors of death in the low-risk group of diagnosis related group(DRG),and construct a risk prediction model by using the radial basis neural network model,in order to reduce the death rate of patients in the low-risk group of DRG and improve the quality of medical safety of patients in hospital.Methods A total of 50344 pieces of home page data of medical records in a gradeⅢ-A hospital in Shanghai from January to August 2023 were selected and divided into cured group,non-cured group and death group according to discharge conditions.Important risk factors were screened out through single factor analysis and used as analysis variables of radial basis neural network to build a risk prediction model.The area under curve(AUC)and the accuracy,sensitivity and specificity of the model were used to evaluate the prediction efficiency of the model.Results The overall prediction accuracy of radial basis function neural network model was 98.69%.Among them,for cured group,the AUC 0.829,95%CI:0.826-0.832,the maximum Jorden index was 0.5823,the sensitivity was 76.29%,and the specificity was 82.06%.For non-cured group,the AUC was 0.825,95%CI:0.822-0.828,the maximum Jorden index was 0.5779,the sensitivity was 76.84%,and the specificity was 80.95%.For death group,the AUC was 0.600,95%CI:0.596-0.605,the maximum Jorden index was 0.2009,the sensitivity was 44.99%,and the specificity was 75.10%.Conclusion For the risk objective prediction of DRG low-risk patients,the radial basis function neural network model of the cure model has better predictive performance.Blood transfusion response,patient age,length of stay and total hospitalization cost are important independent risk factors affecting the discharge status of DRG low-risk group.The model of this study can provide theoretical basis for the deterioration of discharge status and intervention of DRG low-risk group.

关 键 词：径向基神经网络 疾病诊断相关分组 低风险死亡组 风险预测模型 独立危险因素 准确度 敏感度 特异性 

分 类 号：R197.3[医药卫生—卫生事业管理]