Logistic模型联合ROC曲线法和Bayes判别函数对新型冠状病毒肺炎严重程度的鉴别

Application of Logistic model combined with ROC curve and Bayes discriminant function in the diagnosis of severity of COVID-19

目的使用两种统计方法(Logistic模型联合ROC曲线法和Bayes判别函数法)对新型冠状病毒肺炎(简称新冠肺炎,corona virus disease 2019,COVID-19)患者严重程度进行预测,以期辅助临床对于患者分型早期诊断。方法收集538名确诊病例的人口学相关信息、临床资料及流行病学调查资料等,计数资料使用例数(%)表示,采用Logistic回归分析模型进行单因素和多因素回归分析;采用ROC曲线法找到最佳临界值;采用Bayes判别法对研究对象进行分类。结果 Logistic模型联合ROC曲线法的总体预测准确率分别为0.682,重症预测正确率为0.784,轻症预测正确率为0.662;Bayes判别法总体预测准确率分别为0.703,重症预测正确率为0.705,轻症预测正确率为0.702。结论 Logistic模型联合ROC曲线法与Bayes判别分析在鉴别COVID-19临床严重程度诊断中均有较高的正确率,且各有优势,两种方法均有一定的应用价值。

Objective corona virus disease 2019(COVID-19) patients were predicted by two statistical methods(Logistic model combined with ROC curve and Bayes discriminant function), in order to assist the clinical classification of patients for early diagnosis. Methods Demography related information, clinical data and epidemiological investigation data of 538 confirmed cases was collected. Number of cases(%) was used to describe categorical data. Logistic model is used for single factor and multi factor regression analysis;ROC curve method is used to find the optimum critical point;Bayes discriminant method is used to classify the subjects. Results The overall prediction accuracy of the Logistic combined ROC curve method was 0.682, the prediction accuracy of severe symptoms cases was 0.784, and that of mild symptoms cases was 0.662. The overall prediction accuracy of Bayes discriminant method was 0.703, the prediction accuracy of severe symptoms cases was 0.705, and that of mild symptoms cases was 0.702. Conclusions Logistic regression analysis model combined ROC curve method and Bayes discriminant analysis both have high accuracy in the diagnosis of clinical severity of COVID-19, and each has its own advantages. Both of them have certain application value.