Clinical implementation and validation of an AI-enhanced tool for real-time prediction of postpartum hemorrhage risk

The National Institute of Child Health and Human Development awarded Brigham & Women's Hospital Inc. $597,571 on September 3, 2026, under the Child Health and Human Development Extramural Research program (CFDA 93.865) to validate and implement an artificial intelligence-enhanced tool for real-time prediction of postpartum hemorrhage risk integrated into the electronic health record.

The project spans five years through August 31, 2031, with performance at the hospital in Boston, Massachusetts. Work includes two primary aims: refining the AI model through iterative usability testing within a simulated EHR environment guided by the Consolidated Framework for Implementation Research, and rigorously evaluating the tool's effectiveness through an individual-level randomized controlled trial at the hospital. The trial begins with a six-month silent validation phase followed by two years of testing. Primary outcomes measure improved timeliness of clinical interventions for postpartum hemorrhage; secondary outcomes include severe PPH incidence, quantitative blood loss, transfusion rates, maternal morbidity, and healthcare resource utilization. Advanced statistical methods including Cox proportional hazards models and mixed-effects regression analyses will evaluate results.

The work builds on preliminary K08-funded research demonstrating that the machine learning model identifies approximately twenty percent more patients at risk of PPH than existing tools. Current predictive approaches rely on manual data entry and static risk factors, leading to misclassification of at-risk individuals.

Abstract

Postpartum hemorrhage (PPH) remains the leading cause of maternal mortality worldwide, contributing significantly to maternal morbidity. In the united states, PPH accounts for approximately 21% of maternal deaths, with a steadily rising incidence. Current predictive tools rely heavily on manual data entry and static risk factors, leading to misclassification of at-risk individuals and inefficient use of healthcare resources. Building upon robust preliminary data from our K08-funded work demonstrating that our machine learning model identifies approximately 20% more patients at risk of PPH than existing tools, we propose to validate and implement an artificial intelligence (AI)-enhanced tool for automated, real-time PPH risk prediction integrated directly into the electronic health record (EHR). In aim 1, we will refine our AI model through iterative, stakeholder-informed usability testing within a simulated EHR environment, achieving predefined benchmarks for usability and workflow integration guided by the consolidated framework for implementation research (CFIR). In aim 2, we will rigorously evaluate the effectiveness of the refined AI tool through an individual-level randomized controlled trial conducted at a high-volume tertiary care hospital. Following an initial 6-month silent validation to confirm real-time model performance, the tool's impact will be tested over two years. The primary outcome is improved timeliness of clinical interventions for PPH; secondary outcomes include severe PPH incidence, quantitative blood loss, transfusion rates, maternal morbidity, and healthcare resource utilization. Advanced statistical methods, including COX proportional hazards models for the primary outcome and mixed-effects regression analyses for secondary outcomes, will be employed. This project establishes a scalable and generalizable framework for implementing predictive analytics in obstetric care, with potential to significantly improve maternal safety, reduce preventable morbidity and mortality, and optimize resource utilization nationally.