A Clinical Decision Support System for Prediction of Postpartum Hemorrhage in Vaginal Birth

Abstract

Objective The morbidity and mortality caused by postpartum hemorrhage has been increased since 2016 in China, in addition, promoting vaginal delivery is an important task in China currently. This study aimed to develop a clinical decision support system (CDSS) to predict postpartum hemorrhage among vaginal delivery women. Design: A retrospective cohort study. Methods We performed a retrospective analysis of medical records among 1587 vaginal delivery women, who had visited the obstetrics clinic at the Third Affiliated Hospital of Zhengzhou University from 2018 to 2020, these women then were randomly divided into a training set (70%), a validation set (15%) and a test set (15%). We adopted a univariate logistic regression model to select the significant features (P < 0.01). Afterward, we trained several artificial neural networks and binary logistic regression to predict the postpartum hemorrhage, the neural networks included multi-layer perceptron (MLP), back propagation (BP) and radial basis function (RBF). In order to compare and identify the most accurate network, we used the confusion matrix and the receiver operating characteristic (ROC) curve. We finally developed a clinical decision support system based on the most accurate network. All statistical analyses were performed by IBM SPSS (version 20), and MATLAB 2013b software was applied to develop the clinical decision support system. Results Initially, 45 potential variables were addressed by the univariate logistic regression, 16 significant predictors were then selected to enter the binary logistic regression and neural networks (P-value < 0.01). After validation, the best performing model was the multi-layer perceptron network with the highest discriminative ability (AUC 0.862, 95% CI 0.838–0.887). Followed by the back propagation model (AUC 0.866; 95% CI 0.842–0.890), the logistic regression model (AUC 0.856; 95% CI 0.832–0.880). The radial basis function model (AUC 0.845; 95% CI 0.820–0.870) had lower discriminative ability. Conclusion In summary, in terms of predicting postpartum hemorrhage, the multi-layer perceptron network performed better than the back propagation network, logistic regression model, and radial basis function network. The developed clinical decision support system based on the multi-layer perceptron network is expected to promote early identification of postpartum hemorrhage in vaginal delivery women, thereby improve the quality of obstetric care and the maternal outcome.

A novel oppositional binary crow search algorithm with optimal machine learning based postpartum hemorrhage prediction model

“…The ML based methods that integrate HER derived predictors can increase symptoms-based screening practices by recognizing the higher risk population as a fundamental necessity for preventive interference before the onset of PPD. Yang et al [18] adapted a univariate LR method for selecting the important features (P < 0.01). Then , they trained numerous ANN and binary LR methods for predicting postpartum hemorrhaging the NN method includes RBF, MLP, and BP.…”