基于机器学习法的急诊留观患者分流研究

Study on Flow of Emergency Observation Patients Based on Machine Learning Method

目的通过建立急诊留观患者分流预测模型,促进急诊医疗服务管理水平提升。方法通过急诊一体化信息系统收集某三级综合医院2018年7月-2020年6月急诊留观区临床资料,应用机器学习法(BP神经网络和CART分类决策树)建立预测模型,使用SPSS 20.0统计软件和SPSS modeller 18.0统计软件进行数据分析和拟合评估。结果BP神经网络和CART分类决策树模型的分类准确率均达75%以上。急诊留观患者收治病房的影响因素主要有疾病谱、诊断个数、首诊科室、年龄。疾病谱为神经系统或循环系统的患者,收治病房比例相对较高;多系统疾病、诊断个数超过4个或≥80岁的患者,收治病房比例较低。结论应优化急诊重点病种救治流程,通过多学科协作优化床位内部分配和收治机制,加强互联网+医联体协作,改善院外分流措施。

Objective To establish the prediction model of emergency observation patient flow to improve the emergency medical service.Methods Clinical data from the emergency observation units of a tertiary general hospital were collected from July 2018 to June 2020 through the integrated emergency management information system.The prediction model was constructed using the machine learning method(BP artificial neural network and CART classification decision tree).The SPSS 20.0 software and SPSS modeller 18.0 software were used for data analysis and fit evaluation..Results The classification accuracy of BP artificial neural network and CART classification decision tree method was more than 75%.The main factors predicting the admission ward of the patients under observation were disease spectrum,diagnosis number,receiving medical department and age.Patients with neurological or circulatory disease spectrum had a relatively higher proportion of admitted to the ward.The proportion of patients admitted to the ward was lower in patients with multi-system diseases or diagnoses exceeding 4 or older than 80 years old.Conclusion The emergency treatment process for key diseases should be optimized.The internal allocation and admission mechanism of hospital beds should be optimized through multidisciplinary collaboration.Hospital should make initiatives to strengthen the collaboration of internet plus medical association and improve the measures of out-of-hospital diversion.