ARTIFICIAL INTELLIGENCE PREDICTION MODELS FOR POSTPARTUM HEMORRHAGE: A SYSTEMATIC REVIEW AND META-ANALYSIS

Abstract

Background: Postpartum hemorrhage (PPH) is a leading cause of maternal mortality globally, with the highest burden in low- and middle-income countries, including regions of Southeast Asia. Given the limited predictive accuracy of traditional risk assessment models, artificial intelligence (AI)-based predictive models have emerged as a promising approach to enhance early detection and prevention. Objectives: To evaluate the effectiveness of AI-based predictive models for PPH through a systematic review and meta-analysis. Methods: Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, we conducted a systematic search of multiple databases (EMBASE, MEDLINE, ScienceDirect, CINAHL, Google Scholar, and Thai-specific resources) for studies published from 2015 to 2025. Conference abstracts, reviews, and studies without specific PPH outcomes were excluded. Two independent reviewers screened studies, extracted data using the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS) and PROBLAST. Statistical analysis was performed using R version 4.4.3. Results: Eleven studies were included, employing algorithms such as XGBoost, Logistic Regression, Random Forest, and Gradient Boosting. The pooled AUROC was 0.850 (95% CI: 0.789–0.912), indicating good predictive performance. However, there was substantial heterogeneity (I2=99.6%), primarily due to differences in populations, PPH definitions, and modeling approaches. Most studies relied on internal validation and did not originate from Southeast Asia, highlighting a significant regional evidence gap. The risk of bias was largely unclear due to inadequate reporting on blinded predictor assessment and validation methods. Furthermore, a funnel plot analysis suggested potential publication bias, especially among smaller studies. Conclusion: AI-based models show promise for predicting PPH but require external validation to confirm generalizability. The absence of studies from Southeast Asia underscores the need for region-specific research, including in Thailand, to develop and validate context-appropriate models for clinical use.