梯度增强机算法预测重症手足口病的应用价值  

Gradient boosting machine algorithms in prediction of severe hand-foot-mouth disease

作　　者：王斌[1] 冯慧芬[2] 王芳[3] 黄平[1] 秦新华 赵保玲 赵敬 易佳音 WANG Bin;FENG Hnifen;WANG Fang;HUANG Ping;QIN Xinhua;ZHAO Baoling;ZHAO Jing;YI Jiayin(Department of Gastroenterology,Fifth Affiliate Hospital of Zhengzhou University,Zhengzhou 450052,Henan Province,China;Department of Infection,Fifth Affiliate Hospital of Zhengzhou University,Zhengzhou 450052,Henan Province,China;Department of Infectious Diseases.Children's Hospital of Zhengzhou University,Zhengzhou 450052,Henan Province,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 [3]郑州大学附属儿童医院感染科,河南郑州450052

出　　处：《解放军医学院学报》2018年第11期959-963,977,共6页Academic Journal of Chinese PLA Medical School

基　　金：国家自然科学基金项目(81473030);河南省医学科技攻关普通项目(201403130);河南省卫生系统出国研修项目(2015065)~~

摘　　要：目的探讨梯度增强机算法模型在预测重症手足口病方面的应用价值。方法收集郑州大学附属儿童医院2017年5-12月住院部诊治的手足口病患儿资料。使用R软件(V3.4.3)进行资料分析,分别构建梯度增强机模型和Logistic回归模型,并对两者的模型预测性能进行比较。结果共纳入1 137例手足口病患儿,平均年龄为2.0±1.4岁,其中男581例,女556例。对于梯度增强机和Logistic模型,其预测正确率分别为82.1%和76.4%,ROC曲线下面积分别为0.813(95%CI:0.796～0.829)和0.752(95%CI:0.693～0.731)。输出梯度增强机的预测变量重要性,前3位分别为白细胞计数、肠道病菌71(EV71)结果和中性细胞比率。结论梯度增强机模型可以用于预测重症手足口病,且相比于传统Logistic算法具有一定的优越性。Objective To explore the value of gradient boosting machine (GBM)algorithms in predicting severe hand-foot-mouth disease (HFMD).Methods The medical data of children with HFMD admitted to Children's Hospital of Zhengzhou University from May to December in 2017 were collected and analyzed by the R software version 3.4.3.GBM and logistic regression model was built respectively and compared for the predictive performance.Results A total of 1137 children with HFMD were enrolled,included 581 males and 556 females.The average age was 2.0±1.4 years.For GBM and logistic model,the prediction accuracy rates were 82.1%,76.4% and the area under the ROC curves were 0.813(95%CI,0.796-0.829),and 0.752(95%CI,0.693-0.731),respectively. In GBM analysis,the top three predictors were white blood cell count,EV71 and neutrophil ratio.The top three risk factors in the logistic model were blood glucose,EV71 and heart rate.Conclusion GBM model can be used to predict severe HFMD and it performs better than traditional logistic regression algorithm.

关 键 词：重症手足口病 梯度增强机 数学模型 

分 类 号：R195[医药卫生—卫生统计学]