基于环境气象因素的机器学习模型预测呼吸系统疾病急诊量  

Machine learning models based on environmental meteorological factors to predict emergency visits for respiratory diseases

作　　者：张宇[1] 万爽 向准 王觅也 ZHANG Yu;WAN Shuang;XIANG Zhun;WANG Miye(Information Center,West China Hospital of Sichuan University,Chengdu 610041,Sichuan Province,China;Medical Management Service Guidance Center,National Health Commission)

机构地区：[1]四川大学华西医院信息中心,成都610041 [2]国家卫生健康委医疗管理服务指导中心

出　　处：《中国数字医学》2023年第7期40-45,共6页China Digital Medicine

基　　金：国家重点研发计划项目-健康体检大数据云服务平台构建(2020YFC2003404)。

摘　　要：目的:探索环境气象因素与呼吸系统疾病急诊量的关系,为医院合理分配急诊医疗资源提供参考。方法:收集成都市2015年1月1日至2018年12月31日逐日环境污染资料、气象资料以及某三级甲等公立医院呼吸系统疾病急诊量。构建多个机器学习模型,遍历预测窗口和预测算法,探索最优模型以分析环境气象因素对呼吸系统疾病人群的影响,并预测呼吸系统疾病急诊量。结果:机器学习模型遍历结果表明7天预测窗口下的RF模型为最佳模型,验证集决定系数(R2)为0.55,均方根误差(RMSE)为6.50。Pearson相关性系数显示呼吸系统疾病急诊量和环境污染因素呈正相关,相关系数绝对值为0.50左右;与气温呈负相关,相关系数绝对值为0.75左右。Embedder特征选择显示,相比于环境污染因素,气温更易引起呼吸系统疾病急诊量发生变化。结论:基于环境气象因素的机器学习模型准确性较高,在测试集上拟合效果较好,模型具备一定的泛化能力。Objective To explore the correlation between environmental meteorological factors and the number of emergency visits for respiratory diseases,and provide reference for the rational allocation of emergency medical resources in hospitals.Methods The daily environmental pollution data,meteorological data and the number of emergency visits of respiratory diseases in a Grade-A tertiary public hospital in Chengdu from January 1,2015 to December 31,2018 were collected.In order to analyze the influence of environmental meteorological factors on the population of respiratory diseases and to predict the emergency cases for respiratory diseases,multiple machine learning models were constructed,and the prediction windows and algorithms were traversed.Results The results of the machine learning model traversing showed that the RF model under the 7-day prediction window was the best model,2 with an validation set R of 0.55 and an RMSE of 6.50.The Pearson correlation coefficient showed that the number of emergency visits for respiratory diseases was positively correlated with environmental pollution factors,the absolute value of correlation coefficient was about 0.50,while it was negatively correlated with air temperature and the absolute value of correlation coefficient was about 0.75.The Embedder's feature selection showed that air temperature was more likely to cause changes in the number of emergency visits for respiratory diseases than environmental pollution factors.Conclusion The model based on environmental and meteorological factors has high accuracy and good performance on the test set,and it has certain generalization ability.

关 键 词：机器学习 呼吸系统疾病 环境污染 气象 

分 类 号：R195[医药卫生—卫生统计学] R714.253[医药卫生—卫生事业管理]