随机森林与Logistic回归模型对子宫内膜癌患者加速康复外科术后早期出院预测的比较

Comparison of random forest and Logistic regression models for predicting early discharge after enhanced recovery after surgery for patients with endometrial cancer

目的应用随机森林和Logistic回归分别构建子宫内膜癌患者加速康复外科术后早期出院预测模型,比较2种模型预测效果。方法采用便利抽样法,选取2019年1月-2021年12月在某三级甲等医院接受妇科加速康复外科手术并符合纳排标准的子宫内膜癌患者328例,按照7∶3的比例随机分配至建模组和验证组,运用随机森林和Logistic回归构建子宫内膜癌患者加速康复外科术后早期出院预测模型,采用准确度、灵敏度、特异度、阳性预测值、阴性预测值、约登指数和受试者工作特征曲线下面积对2种模型的性能进行比较。结果建模组中,随机森林与Logistic回归准确度为1.000,0.896、灵敏度为1.000,0.833、特异度为1.000,0.915、阳性预测值为1.000,0.750、阴性预测值为1.000,0.942、约登指数为1.000,0.729、AUC为1.000,0.950;验证组中,随机森林与Logistic回归准确度为0.969,0.888、灵敏度为0.960,0.750、特异度为0.973,0.943、阳性预测值为0.923,0.840、阴性预测值为0.986,0.904、约登指数为0.933,0.693、AUC为0.940,0.900。结论随机森林模型对子宫内膜癌患者加速康复外科术后早期出院的预测性能优于Logistic回归模型。

Objective To construct prediction models for early discharge of endometrial cancer patients after enhanced recovery after surgery(ERAS)with random forest and Logistic regression,respectively,and compare the prediction effects of the 2 models.Methods Using convenience sampling,328 patients with endometrial cancer who underwent ERAS and met the inclusion and exclusion criteria in a tertiary grade-A hospital from January 2019 to December 2021 were randomly assigned to model group and validation group according to the ratio of 7∶3,and the random forest and Logistic regression were used to construct the prediction models of early discharge after ERAS for patients with endometrial cancer.The performance of the 2 models was compared in terms of accuracy,sensitivity,specificity,positive predictive value,negative predictive value,Jorden index,and AUC of ROC.Results In the model group,the accuracy of random forest and Logistic regression was 1.000,and 0.896;sensitivity 1.000,and 0.833;specificity1.000,and 0.915;positive predictive value 1.000,and 0.750;negative predictive value 1.000,and 0.942;Jorden index 1.000,and 0.729and AUC 1.000,and 0.950 respectively;in the validation group,the accuracy of random forest and Logistic regression was 0.969,and0.888;sensitivity 0.960,and 0.750;specificity 0.973,and 0.943;positive predictive value 0.923,and 0.840;negative predictive value0.986,and 0.904;Jorden’s index 0.933,0.693,and AUC 0.940,and 0.900,respectively.Conclusion Random forest model outperforms Logistic regression model in predicting early discharge after enhanced recovery after surgery for endometrial cancer patients.