BP神经网络在精神分裂症患者住院费用研究中的应用  

Application of BP neural network in the study of hospitalization expense of schizophrenic patients

作　　者：郭正军 董娇[1] 姚丰菊[1] 王玉杰[1] 王慧颖 王传升[1] GUO Zhengjun;DONG Jiao;YAO Fengju;WANG Yujie;WANG Huiying;WANG Chuansheng(The Second Affiliated Hospital of Xinxiang Medical University,Xinxiang,Henan 453002,China)

机构地区：[1]新乡医学院第二附属医院,河南新乡453002

出　　处：《河南预防医学杂志》2022年第11期820-825,共6页Henan Journal of Preventive Medicine

基　　金：河南省医学科技攻关计划联合共建项目(LHGJ20200531)。

摘　　要：目的利用BP神经网络对河南省精神分裂症患者住院费用进行预测,探索其影响因素,并分析模型预测性能。方法收集2020年河南省某三级甲等精神专科医院第一出院诊断为精神分裂症的患者资料,利用SPSS 26.0软件对BP神经网络模型和多重线性回归模型进行拟合,并对2021年1月数据进行测试。计算均方误差(mean square error,MSE),均方根误差(root mean square error,RMSE),平均绝对误差(mean absolute error,MAE)以及平均相对误差(average relative error,MRE)对模型预测性能进行评价。结果根据病人资料和住院费用信息,将住院费用分为:治疗费、检查费、护理费、药费、床位费、无抽搐电痉挛治疗(modified electra convulsive therapy,MECT)费和其他,各类费用的平均费用占总费用平均费用的比例分别为:41.95%、25.94%、13.89%、9.58%、4.01%、3.72%和0.91%。经过BP神经网络模型拟合,影响精神分裂症患者住院费用前三位因素为住院天数、MECT、年龄。BP神经网络模型预测精神分裂症患者住院费用相关误差(MSE、RMSE、MAE、MRE)均低于多重线性回归模型,对2021年1月数据进行预测,BP神经网络模型平均绝对误差小于多重线性回归(2227.08<2403.60),MRE亦优于多重线性回归模型(11.38%<12.48%)。结论BP神经网络拟合河南省精神分裂症患者住院费用性能较优,相关部门可以针对不同病症精神分裂症患者、不同婚姻状况和年龄患者等情况分类给予不同政策帮助。Objective To predict hospitalization expenses of schizophrenia patients using BP neural network,explore the influencing factors,and analyze the prediction performance.Methods The data of the patients with first discharged diagnosis as schizophrenia in a class-A tertiary psychiatric hospital in Henan province in 2020 were collected as the study content.The BP neural network model and multiple linear regression model were fitted by IBM SPSS 26,and the data in January 2021 were used for the model test.Mean square error(MSE),root mean square error(RMSE),mean absolute error(MAE)and average relative error(MRE)were used to assess the prediction performance of the model.Results According to the patient data and hospitalization expenses information,the hospitalization expenses were divided into treatment expenses,inspection expenses,nursing expenses,medicine expenses,bed expenses,MECT expenses and other expenses,and the average expenses of each kind of expenses accounted for the average total expenses were 41.95%,25.94%,13.89%,9.58%,4.01%,3.72%and 0.91%,respectively.After BP neural network model fitting,the top three factors affecting hospitalization expenses of schizophrenia patients were hospitalization days,MECT and age.The relative errors(MSE,RMSE,MAE,MRE)of BP neural network model in predicting hospitalization expenses of schizophrenic patients were lower than those of multiple linear regression model.The data in January 2021 were predicted.The average absolute error of BP neural network model was less than that of multiple linear regression model(2227.08<2403.60),and MRE was also better than that of multiple linear regression model(11.38%<12.48%).Conclusion The fitting performance of BP neural network for hospitalization expenses of patients with schizophrenia in Henan Province is better.Relevant departments can give different policy assistance according to the classification of patients with different diseases,different marital status and age.

关 键 词：BP神经网络 精神分裂症 住院费用 预测性能 影响因素 

分 类 号：R195.1[医药卫生—卫生统计学]