决策树模型在手足口病合并脑膜脑炎重症化危险因素中的应用

目的 探讨C5.0模型结合Logistic回归分析（以下简称C5.0组合模型）在手足口病（hand，foot and mouth disease，HFMD）合并脑膜脑炎重症化影响因素分析中的应用价值。方法 整群抽取2015年4月~2017年7月河南郑州某医院收治的324例HFMD患儿，应用SPSS 21.0进行Logistic回归分析，应用SPSS Modeler 18.0建立C5.0模型，利用单因素Logistic回归分析结果建立C5.0组合模型，比较三种方法的差异。结果 重症化危险因素结果分别为:C5.0组合模型显示心率≥ 140次/min、血糖升高、手足抖动与烦躁不安是其危险因素；C5.0模型显示心率≥ 140次/min、血糖升高、意识障碍、手足抖动和呕吐是其危险因素；Logistic回归模型显示心率≥ 140次/min、手足抖动、血糖升高、呕吐、烦躁不安、中性细胞比率升高是其危险因素。三种模型的灵敏度、特异度、约登指数与受试者工作特征曲线（receiver operating characteristic，ROC）曲线下面积分别如下:C5.0组合模型依次为95.7%，94.2%，0.90，0.946；C5.0模型依次为80.7%，88.3%，0.69，0.845；Logistic回归模型依次为98.0%，70.0%，0.68，0.840。结论 C5.0模型结合单因素Logistic回归在预测HFMD合并脑膜脑炎重症化危险因素方面价值优于C5.0模型和Logistic回归模型。

Application of decision tree in the analysis and prediction of risk factors of severe hand,foot and mouth disease combined with meningocephalitis

Objective To explore the risk factors in the severity progress of children's hand-foot-mouth disease (HFMD) combined with meningocephalitis by the model of C5.0 mixing with the Logistic regression (C5.0 combined models for short). Methods The method of cluster sampling was used to recruit the 324 cases of HFMD children in Zhengzhou Children's Hospital from April 2015 to July 2017. SPSS 21.0 was used to conduct the analysis of Logistic regression. SPSS Modeler 18.0 was used to establish the C5.0 model, SPSS Modeler 18.0 was used to establish the C5.0 combined models based on single Logistic regression analysis. The three models were compared by the differences of the results. Results The results of the three models were as follows:The C5.0 combined models showed heart rate above 140 time per minute, elevated blood sugar, shake of hands and feet and dysphoria. The C5.0 model showed heart rate above 140 time per minute, disorder of consciousness, elevated blood sugar, shake of hands and feet and vomiting. The Logistic regression model showed heart rate above 140 time per minute, shake of hands and feet, elevated blood sugar, vomiting, dysphoria and the ratio of neutrophils. The sensitivity, specificity, Youden index and the area under the ROC of the three models were as follows:the C5.0 combined models were 95.7%, 94.2%, 0.90, 0.946, the model of C5.0 were 80.7%, 88.3%, 0.69 and 0.845 respectively, the Logistic regression model were 98.0%, 70.0%, 0.68 and 0.840 respectively. Conclusions The model of C5.0 combined with the single Logistic regression is better than the C5.0 model and the Logistic regression model which are used to predict the risk factors in the severity progress of HFMD combined with meningocephalitis.