EV71型手足口病重症早期预警模型的建立  

Establishment of early prediction model of severe EV71 hand-foot-mouth disease

作　　者：陈晓瑾 陈海哨 周红萍 CHEN Xiaojin;CHEN Haishao;ZHOU Hongping(Department of Pharmacy,Hangzhou Children's Hospital,Hangzhou 310014,China;不详)

机构地区：[1]杭州市儿童医院药剂科,310014 [2]杭州市儿童医院感染科,310014

出　　处：《浙江医学》2022年第23期2499-2503,2508,共6页Zhejiang Medical Journal

基　　金：杭州市科技发展计划项目(20150633B19)。

摘　　要：目的探讨基于遗传算法优化的误差反向传播(GABP)神经网络构建的预警模型在手足口病重症早期预测中的作用,为重症患儿早期识别提供参考。方法以2014年1月-2016年12月杭州市儿童医院收治的EV71型重症和普通型手足口病患者为对象建立预测模型并进行模型验证。采用MATLAB R2016b构建误差反向传播(BP)模型,计算危险因素的平均影响值(MIV),选取MIV排名前20位的因素优化BP神经网络结构。计算均方根误差,以最小值作为遗传算法最优解,优化BP神经网络的权值和阈值,构建GABP模型。采用五折交叉验证法对模型的性能进行验证,以AUC、特异度和敏感度为指标,比较GABP模型与BP模型的预测性能。结果影响手足口病重症化的前20位危险因素依次为惊跳天数、通用型病毒载量、最高体温、EV71病毒载量、IgM、精神差、肌酸激酶同工酶、FPG水平、易惊、咳嗽、发热(≥37.4℃)天数、颈强直、散居/托幼、IgA、性别、呼吸节律不齐、高热(≥39.0℃)天数、发病至入院时间、手足抖动、体质量。BP模型最终的网络结构为20→40→1,网络性能评价显示GABP模型达到预设均方根误差的步数为100步,所需时间24 s,五折交叉验证中GABP每组模型的AUC均显著高于BP模型,灵敏度和特异度也均优于BP模型。结论GABP神经网络能提高模型训练效率和预测精准度,可作为EV71型手足口病重症早期预警的参考工具。Objective To establish an early prediction model of enterovirus 71(EV71)hand-foot-mouth disease(HFMD)based on genetic algorithms(GA)and back propagation(BP)neural networks.Methods The clinical data of severe and mild EV71 HFMD children within 3 days of onset,who were admitted to Hangzhou Children’s Hospital from January 2014 to December 2016 were collected.A BP model was established by MATLAB software and the input variables were ranked by the mean impact values(MIV)of the risk factors.Top 20 factors were selected to optimize the structure of BP model.GA was optimized by calculating the minimum value of root mean squared error,as well as the weights and thresholds of the BP network to establish a GABP models.The 5-fold cross validation was performed to evaluate GABP and BP models according to the parameters of ROC curve.Results The top 20 risk factors for the severe progression of HFMD were as follows:duration of startle,total viral loading,the highest body temperature,EV71 viral loading,IgM level,poor spirit,creatine kinase enzyme(CK-MB)level,FPG,panic tendency,cough,duration of fever(≥37.4℃),neck rigidity,in home care,IgA level,gender,respiratory rhythm,duration of high fever(≥39.0℃),days from onset to admission,shaking of hands and feet,body weight.The final network structure of BP model was 20→40→1.The network performance of GABP model was more efficient than single BP model,iterative step was 100 steps and the time cost was 24 seconds in GABP models under the minimum value of root mean squared error.The 5-fold cross validation of GABP models showed better predictive performance than BP models according to the AUC,sensitivity and specificity.Conclusion BP neural network optimized by genetic algorithm could improve the performance efficiency and accuracy of the prediction model.It is a feasible method for early identification of progression of EV71 HFMD.

关 键 词：遗传算法 BP神经网络 预测模型 手足口病 重症化 

分 类 号：R725.1[医药卫生—儿科] R181.3[医药卫生—临床医学]