A Real-time, Automated Maternal Hemorrhage Risk Assessment Platform Using Artificial Intelligence

Noma Ai Inc., Pittsburgh PA

Investigators

Abstract

Summary/Abstract NOMA AI, Inc. is developing a real-time, automated, AI-led risk assessment software solution for maternal hemorrhage. Maternal hemorrhage is a leading cause of maternal mortality and morbidity worldwide and in the U.S. Significant disparities exist, with black mothers facing up to four times higher risk of mortality. In addition to the risk of death, maternal hemorrhage can result in serious long-term complications; many patients require blood transfusions and intensive care. Despite evidence that the majority (54-93%) of cases are preventable, maternal hemorrhage and associated blood product needs have to-date proven exceedingly difficult to predict. Thus, there remains a significant unmet care gap despite mandates and intense broad efforts at improving preparedness and matching patients to risk-appropriate workflows. Current strategies for managing maternal hemorrhage begin with separating patients into risk groups based on specific known risk factors. This process is typically done manually using paper or electronic questionnaires, defined by criteria developed by various consortia. However, these criteria lack both sensitivity and positive predictive value: nearly half of mothers that require blood transfusion are categorized falsely as low risk. The poor performance of existing tools suggests that critical risk factors and predictive features are not being captured. NOMA AI proposes to solve this problem by leveraging state-of-the-art machine learning (ML) techniques, as well as a much more comprehensive suite of risk factors, to develop improved and dynamic risk assessment solutions that capture complex temporal patterns in patient conditions, better informing doctors and reducing risk for mothers. In this Phase I effort, we will perform the first comprehensive multi-center retrospective and prospective evaluation accounting for not only model performance, but also fairness and bias of a complementary EHR-integrated clinical decision support software. Our improved and dynamic risk stratification models will result in earlier recognition of risk and enable proactive interventions, smarter allocation of resources, automation of tedious manual processes, reduced blood transfusions and better patient outcomes. Our Aims are to 1) develop and evaluate improved maternal hemorrhage risk assessment models using a suite of dynamic patient data and ML solutions, and 2) develop a comprehensive bias mitigation/fairness pipeline for AI-led applications, which will help to address some of the prevalent racial and ethnic disparities within maternal hemorrhage as well as serve future efforts to facilitate a more equitable quality of care. This work will provide a foundation for a Phase II follow-on effort in which we obtain key clinical feedback needed to develop our user interface and integrate it into clinical workflows, as well as investigate critical milestones of real-time EHR integration and conduct prospective validation studies in real-time, real-world clinical environments.