MACHINE LEARNING PREDICTIVE MODELS FOR POSTPARTUM HEMORRHAGE: A SYSTEMATIC REVIEW AND META-ANALYSIS

Background Postpartum hemorrhage (PPH) is a leading cause of maternal mortality worldwide. Early risk stratification may enable the implementation of preventive measures and facilitate timely management. Machine learning (ML) models offer potential for predicting PPH by capturing complex risk patterns. Objectives To assess the predictive performance of ML models applied at admission or before delivery for predicting PPH in singleton or twin pregnancies. Search Strategy We systematically searched PubMed, Embase, Web of Science, and Google Scholar without date restrictions. Only English-language studies were considered. Selection Criteria Eligible studies included observational designs or clinical trials that developed or validated supervised ML models to predict PPH. Exclusion criteria were conference abstracts and studies without original data or performance metrics. Data Collection and Analysis Two reviewers independently screened studies and extracted data using instruments based on the CHARMS and TRIPOD+AI frameworks. The risk of bias and applicability were assessed using the PROBAST tool. A random-effects meta-analysis was used to estimate the pooled area under the curve (AUC) with 95% confidence intervals (CIs). Heterogeneity was quantified using the I ² statistic. Main Results Twenty-four studies met the inclusion criteria. The pooled AUC was 0.83 (95% CI: 0.78–0.88), indicating good discriminatory performance for PPH. However, 13 studies were rated as having a high risk of bias, and 7 raised significant concerns regarding applicability. Heterogeneity was substantial ( I ² = 99.8%). Conclusions ML models show promise in predicting PPH but are limited by the methods employed, inconsistent outcome definitions, and limited validation.