摘要
目的探索数据挖掘技术在预测住院天数上的应用,为医院管理提供辅助数据支持。方法采集某三级甲等医院2015―2016年病案首页数据,通过单因素分析结合专家建议筛选特征属性,使用支持向量机对住院天数进行分类预测实验。结果筛选出10个属性作为实验特征属性,住院天数分为极短期(1天),短中期(2至14天),中长期(15至28天),长期(28天以上)。在四分类预测中,极短期,短中期及长期住院患者预测效果较好;二分类预测中,短中期与长期住院患者预测效果较好。结论预测结果可以为医院前置综合管理提供决策支持,如病区医疗资源分配、床位周转、异常住院天数人群干预等。
Objective To explore the application of data mining technology in predicting the inpatient days and to provide auxiliary data support for hospital management.Methods The first page data of medical records in a Grade-A tertiary hospital from 2015 to 2016 was collected.Through single factor analysis combined with expert recommendations,the characteristic attributes were selected.The support vector machine was used to classify and predict the inpatient days.Results Ten attributes were selected as the experimental characteristic attributes.The length of hospital stay was divided into very extreme short-term(1 day),short-term(2-14 days),medium-and long-term(15-28 days),and long-term(28 days and above).In the fourcategory prediction,the prediction effect of extreme short-term,short-term and long-term hospitalized patients is better.In the bipartite prediction,the prediction effect of short-term and long-term hospitalized patients is better.Conclusion The prediction results can provide decision support for the comprehensive management of the hospital,such as the allocation of medical resources in the ward,bed turnover,and intervention for people with abnormal hospitalization days.
作者
庞震
孙静
李佩佳
张欣阳
石勇
杨宇飞
PANG Zhen;SUN Jing;LI Pei-jia;ZHANG Xin-yang;SHI Yong;YANG Yu-fei(Xiyuan Hospital CACMS;Peking University Third Hospital;Ningbo Development Planning Research Institute;CAS Research Center on Fictitious Economy&Data Science)
出处
《医院管理论坛》
2020年第10期22-24,27,共4页
Hospital Management Forum
