Development of early warning models for postpartum hemorrhage following cesarean section in parturi-ents with different types of placenta previa using machine learning

LIN Chaoying

doi:10.3969/j.issn.1671-8631.2025.11.002

Abstract：Objective:To develop early warning models utilizing machine learning techniques for assessing the risk of postpartum hemorrhage following cesarean section in parturients with varying types of placenta previa.Methods A prospective selection was made of 77 pregnant women with different types of placenta previa who were admitted to the obstetrics department of ganzhou maternal and child health hospital from January 1,2022 to November 30,2023.These cases were divided into two groups based on the occurrence of massive postpartum hemorrhage:the non-massive post-partum hemorrhage group and the massive postpartum hemorrhage group.Additionally,according to a 7∶3 ratio for the training set and validation set,33 cases of pregnant women with different types of placenta previa admitted from De-cember 1,2023 to October 1,2024,were selected for the validation set.Relevant patient data were collected,and inde-pendent predictors were identified through univariate analysis and multivariate logistic regression analysis.Subsequent-ly,logistic regression,decision tree(DT),and back-propagation neural network(BPNN)models were constructed.The predictive performance of these models was evaluated using the area under the receiver operating characteristic(ROC)curve(AUC)and confusion matrix-related indicators(accuracy,precision,recall,F1 score).External valida-tion was also conducted.Results:Eight independent predictors of postpartum hemorrhage were identified through uni-variate and multivariate logistic regression analyses:gravidity and parity,history of induced abortion,metabolic diseases during pregnancy,history of hysteroscopy,scarred uterus,placenta accreta,placenta adhesion,and type of placenta pre-via.The BPNN model exhibited superior AUC values(0.902,0.892)in both the training and validation sets compared to the Logistic and DT models(0.833,0.820 and 0.884,0.852).The BPNN model also outperformed the Logistic and DT models in terms of confusion matrix-related indicators in both sets.In the BPNN model,the importance ranking of independent influencing factors for massive postpartum hemorrhage in cesarean delivery patients with placenta previa is as follows:type of placenta previa＞placenta accreta＞placenta adherence＞scarred uterus＞metabolic diseases during pregnancy＞gravidity and parity＞history of hysteroscopy surgery＞history of induced abortion.Conclusion:The BPNN model demonstrates better performance than the Logistic and DT models in predicting postpartum hemorrhage following cesarean section in parturients with placenta previa,providing improved risk assessment for patients and effectively guiding clinical decision-making.