基于机器学习模型对缺血性脑卒中住院期间死亡风险的预测  

Prediction of mortality risk during hospitalization for ischemic stroke based on machine learning models

作　　者：王小曼 游一鸣 韩梦琦 程峙娟 俞鹏飞 刘建模 罗颢文 刘松 涂江龙[5] 易应萍[1] WANG Xiao-man;YOU Yi-ming;HAN Meng-qi;CHENG Zhi-juan;YU Peng-fei;LIU Jian-mo;LUO Hao-wen;LIU Song;TU Jiang-long;YI Ying-ping(Medical Big Data Research Center,The Second Affiliated Hospital of Nanchang University,Nanchang,Jiangxi 330006,China;不详)

机构地区：[1]南昌大学第二附属医院医疗大数据研究中心,江西南昌330006 [2]南昌大学公共卫生学院,江西省预防医学重点实验室,江西南昌330000 [3]青岛市市立医院放射科 [4]西安交通大学附属红会医院科教部 [5]南昌大学第二附属医院神经内科,江西南昌330000 [6]南昌大学第二附属医院科技处

出　　处：《现代预防医学》2024年第19期3457-3462,3482,共7页Modern Preventive Medicine

基　　金：科技部国家重点研发计划(2020YFC2002901);国家自然科学基金地区科学基金项目(81960609);江西省重点研发计划重点项目(揭榜挂帅)(20223BBH80013);江西省科技厅项目(20203BBGL73129);江西省应用研究培育计划(20212BAG70029);南昌大学第二附属医院院内资助项目(2021efyB03)。

摘　　要：目的使用logistic回归与极端梯度提升(extreme gradient boosting,XGboost)、支持向量机(support vector machine,SVM)、随机森林三种机器学习算法对缺血性脑卒中患者住院期间死亡风险进行预测,并比较几种模型的预测效果。方法回顾性收集2018年1月-2019年12月南昌大学第二附属医院收治的4038例缺血性脑卒中患者资料,按照7∶3的比例分为训练集和验证集,使用Python进行随机森林、SVM、XGboost三种机器学习模型的训练与验证,并进行logistic回归分析,通过准确性、召回率、F1-score和受试者工作特征曲线下面积(area under the curve,AUC)等指标比较各模型对缺血性脑卒中患者住院期间死亡风险预测的效果。结果三种机器学习模型及logistic回归模型对缺血性脑卒中患者住院期间死亡风险预测的效果:按准确率从低到高为logistic、随机森林、SVM、XGboost,分别为90.63%、92.36%、92.62%、93.52%;按F1-score从低到高为随机森林、SVM、logistic、XGboost,分别为87.32%、88.41%、89.04%、90.01%;按AUC从低到高为SVM、logistic、随机森林、XGboost,分别为0.93、0.94、0.96、0.97。结论建立的XGboost模型对缺血性脑卒中患者住院期间死亡风险预测具有良好的效果,可为医生提供良好的临床辅助决策。Objective To predict the mortality risk during hospitalization for patients with ischemic stroke using three machine learning algorithms:Logistic Regression,Extreme Gradient Boosting(XGBoost),Support Vector Machine(SVM),and Random Forest,and to compare the predictive performance of these models.Methods A retrospective study was conducted to collect data of 4038 ischemic stroke patients admitted to The Second Affiliated Hospital of Nanchang University from January 2018 to December 2019.The data were divided into training and validation sets at a 7:3 ratio.Random Forest,SVM,and XGBoost models were trained and validated using Python,and Logistic Regression analysis was performed.The models were compared based on accuracy,recall,F1-score,and the Area Under the Curve(AUC)of the Receiver Operating Characteristic(ROC)curve.Results The predictive performance of the four models for mortality risk during hospitalization for ischemic stroke patients,ranked by accuracy from lowest to highest,was as follows:Logistic Regression(90.63%),Random Forest(92.36%),SVM(92.62%),and XGBoost(93.52%).According to the F1-score,the ranking was as fllows Random Forest(87.32%),SVM(88.41%),Logistic Regression(89.04%),and XCBoost(90.01%).According to AUC,the models ranked as follows:SVM(0.93),Logistic Regression(0.94),Random Forest(0.96),and XGBoost(0.97).Conclusion The established XGBoost model demonstrated excellent predictive performance for mortality risk during hospitalization for ischemic stroke patients,providing valuable clinical decision support for physicians.

关 键 词：缺血性脑卒中 住院 死亡风险 机器学习 预测 

分 类 号：R743.3[医药卫生—神经病学与精神病学]