基于随机森林算法的儿童重症腺病毒肺炎临床预测模型研究

A study on establishing a clinical predictive model of severe adenovirus pneumonia in children based on random forest algorithm

目的基于随机森林算法构建儿童重症腺病毒肺炎(severe adenovirus pneumonia,SAP)的临床预测模型,并对其进行验证。方法采用观察性研究设计,回顾性分析2019年1月至2021年1月天津市儿童医院收治的542例腺病毒肺炎患儿的临床、实验室及影像学资料。将研究对象随机分为训练集和验证集(8∶2)。训练集通过随机森林算法筛选SAP的预测因子建立预测模型,并通过列线图将预测模型可视化表达。在验证集中利用受试者工作特征(ROC)曲线和敏感性、特异性、误判率、混淆矩阵对其进行验证。结果训练集患儿439例,其中重症型187例(42.60%),验证集患儿103例,其中重症型44例(42.71%)。训练集中单核细胞百分比(M%)、PLT、AST、IL-6、热峰、肺部大片炎性实变、肺部斑片状阴影是影响SAP的独立预测因子。模型区分度验证发现训练集和验证集的ROC曲线下面积分别为0.95(95%CI:0.92~0.98)和0.92(95%CI:0.82~0.99)。训练集的准确度、灵敏度、特异性、阳性预测值和阴性预测值分别为0.994、1.000、0.987、0.998、1.000;验证集的分别为0.752、0.990、0.514、0.945、0.857。结论该预测模型具有较好的判别能力,早期的临床及血液学指标有助于提高儿童SAP的识别和筛选,具有一定的临床价值。

Objective To construct a clinical predictive model of severe adenovirus pneumonia(SAP)in children using random forest and verify it.Methods The clinical,laboratory and imaging data of 542 children with adenovirus pneumonia treated in Tianjin Children′s Hospital from January 2019 to January 2021 were analyzed retrospectively.The research object was randomly divided into training dataset and validation dataset(8∶2).The training dataset screened the predictors of SAP of pneumonia through random forest and established a prediction model,and the prediction model was expressed visually by the nomogram.In the validation dataset,the receiver operating characteristic curve(ROC)and sensitivity,specificity,error rate and confusion matrix were used to validate it.Results A total of 439 children were in the training dataset,and 187 cases(42.60%)of the training data was divided as severe type.A total of 103 children were in validation dataset,and 44 cases(42.71%)of the validation dataset was divided as severe type.The percentage of monocytes(M%),PLT,AST,IL-6,the peak of body temperature,pulmonary inflammation of the consolidation and patchy shadowing were independent predictors of SAP in children.The area under the ROC curve of the training dataset and the validation dataset was 0.95(95%CI:0.92~0.98)and 0.92(95%CI:0.82~0.99),respectively.The accuracy,sensitivity,specificity,positive predictive value and negative predictive value of the training dataset were 0.994,1.000,0.987,0.998,1.000 and in validation dataset were 0.752,0.990,0.514,0.945 and 0.857,respectively.Conclusion The predictive model has good discriminant ability,and the early clinical and hematological indexes are helpful to improve the identification and screening of SAP in children.