基于深度神经网络的剖宫产术后出血临床预测模型  

Clinical prediction model for postpartum hemorrhage after cesarean section based on deep neural network

作　　者：贾俊荣 JIA Junrong(Department of Obstetrics and Gynecology,Mudan People's Hospital,Shandong Province,Heze274000,China)

机构地区：[1]山东省菏泽市牡丹人民医院妇产科,山东菏泽274000

出　　处：《中国当代医药》2025年第1期30-33,共4页China Modern Medicine

摘　　要：目的分析剖宫产产后出血(PPH)的影响因素,筛选PPH的最佳预测特征,建立准确预测发生PPH的临床预测模型。方法选取菏泽市牡丹人民医院收治的210例剖宫产产妇为研究对象,采用随机抽样法,将210例产妇以7∶3的比例划分为训练集147例和测试集63例。根据PPH的临床表现,将产妇分为PPH组和非PPH组。回顾性分析产妇的临床资料,利用随机森林(RF)算法,筛选出PPH的最佳预测特征,构建基于最佳预测特征的深度神经网络预测模型。采用AUC为指标,评估深度神经网络模型的预测性能。结果随机森林算法筛选的PPH的最佳预测特征是前置胎盘、妊娠高血压、血小板异常、多胎、多羊水、妊娠糖尿病、胎盘植入、胎盘早剥、骨盆软产道异常和年龄。神经网络模型在训练集和测试集上的AUC值分别为0.918和0.902。结论本研究建立的基于深度神经网络的剖宫产PPH临床预测模型,具有良好的预测性能,能够为剖宫产PPH的临床风险作出有效评估。Objective To analyze the influencing factors of postpartum hemorrhage(PPH)after cesarean section,identify the best predictive features of PPH,and develop a clinical model to predict the occurrence of PPH accurately.Methods A total of 210 puerpera with cesarean section admitted to Mudan People's Hospital were selected as the research objects,and they were divided into training dataset(147 cases)and testing dataset(63 cases)according to random sampling method.Based on the clinical manifestations of PPH,puerpera were categorized into PPH group and non-PPH group.Retrospective analysis of clinical data of puerpera was applied,and the random forest(RF)algorithm was firstly used to identify the best predictive features of PPH,and then the deep neural network was developed based on the best predictive features.Finally,the AUC was used as the index to evaluate the performance of the deep neural network model.Results The best predictive features of PPH identified by the RF algorithm includes ten categories,i.e.,placenta previa,gestational hypertension,platelet abnormalities,multiple births,polyhydramnios,gestational diabetes,placenta accreta,placenta abruption,pelvic soft birth canal abnormalities and age.The value of the AUC of the deep neural network model for PPH during cesarean delivery were 0.918 and 0.902 for training dataset and testing dataset,respectively.Conclusion The proposed clinical prediction model for PPH after cesarean section proposed in this research has a satisfactory accuracy,and it can effectively evaluate the clinical risk of PPH after cesarean section.

关 键 词：产后出血 剖宫产 深度神经网络 临床预测模型 

分 类 号：R71[医药卫生—妇产科学]