Evaluate the Application of Machine Learning Algorithms to the Management of Postpartum Hemorrhage
CERSI Collaborators: Moeun Son, MD, MSCI (Yale) (PI), Jennifer Culhane, PhD, MPH (Yale) (PI), Molly Jeffery, PhD (Mayo Clinic), Lisbet Lundsberg, PhD, MPH (Yale), Joseph Ross, MD, MHS (Yale)
FDA Collaborators: Kristie Baisden, DO, Leah Berhane, MD, Susan Bersoff-Matcha, MD, Laurèn Doamekpor, PhD, MPH, Christine Lee, PharmD, PhD, Leyla Sahin, MD, Yvonne Santiago, MD, Catherine Sewell, MD, Kaveeta Vasisht, MD, PharmD, Robert Whetsel, PhD, DCS, MS
CERSI Subcontractors: Children’s Hospital of Philadelphia- Heather Burris, MD, MPH, Sara Handley, MD, MSCE, Kathryn McKenney, MD, MPH; Nemours Children’s Hospital- Kevin Dysart, MD
Project Start Date: April 26, 2022
Regulatory Science Challenge
The FDA Office of Women’s Health (OWH) has identified maternal mortality and morbidity as a priority in addressing maternal illness and death and improving care and outcomes for different populations of women. Postpartum hemorrhage (PPH), or excessive bleeding following the birth of a baby, remains a leading cause of maternal illness and mortality in the United States, accounting for 10.7% of pregnancy-related deaths from 2014-2017 according to the CDC. PPH is typically managed in such that specific actions may be taken as the condition progresses. However, there is substantial variation in how the steps during progression are carried out with regard to order, timing, and pace. This variation may contribute to differences in maternal illness and death outcomes. More advanced analytic tools are needed to inform clinical decisions in PPH management and developing these tools requires thorough examination of the course of obstetrical care.
Project Description and Goals
This project will use machine learning as a novel analytic tool to help determine the sources of variation in PPH management across several different populations of women with the goal of refining current clinical practices to improve maternal care and outcomes. Existing electronic medical record (EMR) data from two academic medical centers (Yale University and The University of Pennsylvania) will be used to create a reliable and accepted data source. Advanced machine learning techniques will be applied to the EMR data to evaluate patient characteristics associated with different patterns of PPH management. Finally, these data will be evaluated to determine if certain steps of PPH management are associated with increased risk of hysterectomy, other severe maternal diseases, and maternal death. This study aims to answer the following: 1) are specific clinical and sociodemographic attributes associated with variations in PPH management, 2) are variations/deviations from accepted patterns of PPH management associated with higher rates of severe maternal disease, including the need for hysterectomy.