决策树C5.0与Logistic回归模型对产后腹直肌分离预测性能的比较研究

Decision tree C5.0 versus Logistic regression model in predicting postpartum diastasis recti abdominis

目的比较决策树C5.0与Logistic回归模型对产后腹直肌分离的预测效果。方法选取产后复查的产妇476例作为研究对象。采用问卷调查法获取产妇的一般资料;采用电刺激治疗仪评估盆底肌电值;采用腹部触诊法判断腹直肌分离程度。将所有数据按照3∶2的比例建立训练集与测试集,运用决策树C5.0及Logistic回归建立产后腹直肌分离的风险预测模型;采用准确度、灵敏度、特异度、约登指数、阴性预测值、阳性预测值和受试者工作特征(ROC)曲线的曲线下面积(AUC)对模型的预测性能进行比较。结果在训练集中,决策树C5.0和Logistic回归模型的准确度分别为96.94%、72.45%,灵敏度分别为98.92%、86.02%,特异度分别为93.52%、49.07%,阳性预测值分别为96.34%、74.42%,阴性预测值分别为98.06%、67.09%,约登指数分别为92.44%、35.10%,AUC分别为0.962、0.675;训练集中,决策树C5.0和Logistic回归模型的AUC比较,差异有统计学意义(P<0.05)。在测试集中,决策树C5.0与Logistic回归模型的准确率分别为81.50%、62.43%,灵敏度分别为88.35%、82.52%,特异度分别为71.43%、32.86%,阳性预测值分别为81.98%、64.39%,阴性预测值分别为80.65%、56.10%,约登指数分别为59.78%、15.38%,AUC分别为0.799、0.577;测试集中,决策树C5.0和Logistic回归模型的AUC比较,差异有统计学意义(P<0.05)。结论决策树C5.0对产后腹直肌分离的预测效能优于Logistic回归模型。

Objective To compare the effects of decision tree C5.0 and Logistic regression model in predicting postpartum diastasis recti abdominis.Methods A total of 476 puerperas with postpartum re-examination were selected as research objects.A questionnaire survey method was used to obtain general information of puerperas;an electrical stimulation therapy device was used to assess the electromyography values of the pelvic floor muscles;abdominal palpation method was used to determine the degree of diastasis recti abdominis.All the data were divided into training set and test set at a ratio of 3∶[KG-*2/5]2,and the decision tree C5.0 and Logistic regression were used to establish the risk prediction models for postpartum diastasis recti abdominis;the accuracy,sensitivity,specificity,Youden index,negative predictive value,positive predictive value and area under curve(AUC)of the receiver operating characteristic(ROC)curve were used to compare the prediction performance of the models.Results In the training set,the accuracies of decision tree C5.0 and Logistic regression model were 96.94%and 72.45%respectively,the sensitivities were 98.92%and 86.02%respectively,the specificities were 93.52%and 49.07%respectively,the positive predictive values were 96.34%and 74.42%respectively,the negative predictive values were 98.06%and 67.09%respectively,the Youden index values were 92.44%and 35.10%respectively,and the AUC values were 0.962 and 0.675 respectively;there was a significant difference in AUC between decision tree C5.0 and Logistic regression model in the training set(P<0.05).In the test set,the accuracies of decision tree C5.0 and Logistic regression model were 81.50%and 62.43%respectively,the sensitivities were 88.35%and 82.52%respectively,the specificities were 71.43%and 32.86%respectively,the positive predictive values were 81.98%and 64.39%respectively,the negative predictive value were 80.65%and 56.10%respectively,the Youden index values were 59.78%and 15.38%respectively,and the AUC values were 0.799 and 0.577 respectively;there was a significant difference in AUC between decision tree C5.0 and Logistic regression model in the test set(P<0.05).Conclusion Decision tree C5.0 is better than Logistic regression model in predicting postpartum diastasis recti abdominis.