极端梯度上升模型在预测临床重症手足口病中的应用价值  

Value of extreme gradient boosting model in predicting clinical severe hand-foot-mouth disease

作　　者：赵敬 冯慧芬[2] 王斌[1] 黄平[1] ZHAO Jing;FENG Hui-fen;WANG Bin;HUANG Ping(Department of Gastroenterology,Fifth Affiliated Hospital of Zhengzhou University,Zhengzhou,Henan 450052,China)

机构地区：[1]郑州大学第五附属医院消化内科,河南郑州450052 [2]郑州大学第五附属医院感染科,河南郑州450052

出　　处：《中国临床研究》2019年第10期1323-1326,共4页Chinese Journal of Clinical Research

基　　金：国家自然科学基金(81473030);河南省卫生系统出国研修项目(2015065)~~

摘　　要：目的探讨极端梯度上升模型(XGBoost)和Logistic回归模型在临床重症手足口病(HFMD)预测中的应用对比。方法回顾性收集郑州市某医院2017年3月至11月期间住院部收治的HFMD患儿872例的临床资料,其中轻症488例,重症384例。使用R344软件进行所有资料的分析,分别构建XGBoost和Logistic回归模型,比较两种模型对重症HFMD的预测效果。结果在XGBoost模型中,输出变量重要性中前三位分别为:白细胞计数、年龄和心率,其对重症HFMD总体预测准确性为92.4%,ROC曲线下面积为0.952(95%CI:0.931~0.967)。Logistic回归模型总体预测准确性为801%,ROC曲线下面积为0.848(95%CI:0.833~0.866)。模型评估显示XGBoost模型的预测效果明显优于Logistic回归模型。结论XGBoost模型可以用于预测重症HFMD,相比于传统模型,具有较高的准确性和诊断价值。Objective To compare the application of extreme gradient boosting(XGBoost) model and Logistic regression model in the prediction of severe hand-foot-mouth disease(HFMD).Methods Clinical data of 872 children with HFMD admitting to hospital from March 2017 to November 2017 were collected retrospectively,including mild cases(n=488) and severe cases(n=384).R 3.4.4 software was used for all data analysis,XGBoost and Logistic regression models were constructed respectively.The predictive effects of two models on severe HFMD were compared.Results In XGBoost model,the first three most important output variables were white blood cell count,age and heart rate.The overall accuracy of predicting severe HFMD was 92.4%,and the area under receiver operating characteristic curve(AUC) was 0.952(95% CI:0.931-0.967).In Logistic regression model,the overall accuracy of predicting severe HFMD was 80.1%,and AUC was 0.848(95% CI:0.833-0.866).Model evaluation showed that XGBoost model was superior to Logistic regression model in prediction effect.Conclusion XGBoost model can be used to predict severe HFMD disease and has a higher precision and diagnostic value compared with traditional model.

关 键 词：手足口病 重症 极端梯度上升模型 LOGISTIC回归模型 预测 

分 类 号：R512.5[医药卫生—内科学]