运用3种统计学方法构建的顽固性产后尿潴留的风险预测模型的分析比较被引量：7

Analysis and comparison of risk prediction models for protracted postpartum urinary retention constructed by three statistical methods

在线阅读下载PDF

导出

摘要目的:比较列线图、决策树和随机森林模型对顽固性产后尿潴留(PUR)的预测效能。方法:回顾性分析2020年1月至2021年12月在钦州市第一人民医院经阴道分娩的380例产妇的临床资料,根据产妇分娩2 d后有无顽固性PUR分为顽固性PUR组和对照组。筛选顽固性PUR的影响因素,运用列线图、决策树和随机森林模型建立顽固性PUR风险预测模型,采用受试者工作特征(ROC)曲线评价3种统计学模型的预测效能。结果:列线图模型的准确度为0.833,灵敏度为0.933,特异度为0.759,召回率为0.900,精确率为0.806,ROC曲线下面积(AUC)为0.902,均高于决策树模型(0.754、0.767、0.741、0.767、0.766、0.763)和随机森林模型(0.771、0.839、0.757、0.800、0.774、0.884)。结论:基于列线图模型在预测阴道分娩产妇发生顽固性PUR中具有较好的预测效果和稳定性,其预测效能优于决策树预测模型和随机森林模型。 Objective: To compare the effectiveness of nomogram, decision tree and random forest model in predicting the risk of protracted postpartum urinary retention(PUR). Methods: The clinical data of 380 women who delivered vaginally at the First People’s Hospital of Qinzhou from January 2020 to December 2021 were retrospectively analyzed, and the women were divided into recalcitrant PUR group and control group according to the presence or absence of protracted PUR at 2 d after delivery. The influencing factors for protracted PUR were screened, and risk prediction models for protracted PUR were established using nomograph, decision tree and random forest models, and the predictive efficacy of the three statistical models was evaluated by using receiver operating characteristic(ROC) curves. Results: The nomograph model had an accuracy of 0.833, sensitivity of0.933, specificity of 0.759, recall of 0.900, precision of 0.806, and area under the ROC curve(AUC) of 0.902, all of which were higher than those of the decision tree model(0.754, 0.767, 0.741, 0.767, 0.766, and 0.763, respectively) and the random forest model(0.771, 0.839, 0.757, 0.800, 0.774, and 0.884, respectively). Conclusion:The nomogram-based model has better prediction effect and stability in predicting protracted PUR after vaginal delivery, and its predictive efficacy is better than the factors decision tree model and random forest model.

作者邓连方梁洁卢丹黄秋杰何水 Deng Lianfang;Liang Jie;Lu Dan;Huang Qiujie;He Shui(Department of Obstetrics,First People’s Hospital of Qinzhou,Qinzhou 535000,China)

机构地区钦州市第一人民医院产科

出处《广西医科大学学报》 CAS 2022年第9期1442-1447,共6页 Journal of Guangxi Medical University

基金广西壮族自治区卫生和计划生育委员会自筹经费科研课题(No.Z20180259)。

关键词产后尿潴留风险预测列线图决策树随机森林影响因素 postpartum urinary retention risk prediction nomogram decision tree random forest influencing factors

分类号 R714.64 [医药卫生—妇产科学]