数据挖掘与模型构建在预测重症手足口病中的应用

Application of data mining and model construction in prediction of severe hand-foot-mouth disease

摘要:

目的 探讨数据挖掘与模型构建在预测重症手足口病方面的价值。方法 回顾性分析郑州大学 第五附属医院2016 年6 月-2017 年10 月收治的838 例手足口病患儿的临床资料，使用SPSS Statistics 23.0 统 计软件进行数据的预处理和分析，使用SPSS Modeler 18.0 软件进行模型构建和评估。根据总体精确性对所有 算法进行筛选，选取最优算法，配置模型参数，输出分类树模型，评估模型的预测性能。结果 经过自动分类 器筛选，最终确定C&R 算法最佳。模型共纳入3 个解释变量：易惊、呕吐及肢体抖动。使用错分矩阵计算 后，模型的预测正确率为91.17%，敏感性为84.36%，特异性为96.25%。ROC 曲线下面积为0.903[（95% CI ： 0.878，0.927），P =0.000]。结论 决策树模型在预测手足口病方面有一定的优势，模型预测精确度较高，对临 床疾病诊疗有一定的辅助价值。

Abstract:

Objective To explore the value of data mining and model construction in predicting severe hand-foot-mouth disease (HFMD). Methods A retrospective analysis was performed on the clinical data of 838 children with HFMD treated in the Fifth Affiliated Hospital of Zhengzhou University from June 2016 to October 2017. SPSS Statistics 23.0 was used for data preprocessing and statistical analysis, while SPSS Modeler 18.0 was used for modeling and evaluation. The model parameters were configured to output classification tree model and assess predictive performance when the optimal algorithm was screened from all algorithms based on overall accuracy. Results C&R algorithm was finally determined to have better accuracy by the automatic classifier screening. The model included three explanatory variables: shock, vomiting and limb shaking. The prediction accuracy of the model was 91.17%, with the sensibility of 84.36% and the specificity of 96.25%. The area under the ROC curve was 0.903 (95% CI: 0.878, 0.927) (P < 0.05). Conclusions Decision tree model has some advantages in the prediction of hand, foot and mouth disease and has high prediction accuracy. The model has a supplementary value in clinical diagnosis and treatment of the disease.