Leveraging machine learning for cardiovascular disease risk prediction and prevention in women with a history of adverse pregnancy outcomes - PROJECT ABSTRACT Each year, more than a million US women experience adverse pregnancy outcomes (APOs) associated with a two-fold higher risk of future maternal CVD. These APOs include preeclampsia, gestational hypertension, gestational diabetes, preterm delivery (<37 weeks), and delivery of a small for gestational age infant. The American Heart Association and American College of Cardiology recognize APOs as risk factors for CVD. However, it is not yet known how to identify which women with a history of an APO are most at risk; nor are specific, modifiable CVD risk factor trajectories from each APO to CVD well described. To date, risk prediction exercises have only examined the utility of adding individual APOs to existing CVD risk scores. No risk prediction exercises have addressed the clinically relevant question of which women with APOs are most at risk for CVD, and no clinical decision support tool exists to direct care after an APO. We propose to develop a risk classification approach to identify maternal risk phenotypes at the time of delivery of a pregnancy complicated by an APO that are predictive of future CVD risk factors and events. Predictive factors that inform these phenotypes may include history of individual APOs or combinations of APOs, delivery complications, infant outcomes, and maternal demographic or behavioral factors available at or shortly after delivery (e.g., preterm preeclampsia with BMI <30 kg/m2). We will use machine learning to elucidate novel maternal CVD risk phenotypes in several populations: two longitudinal Nurses’ Health Study cohorts and a retrospective cohort using electronic health record data from the Duke University Health System and University of North Carolina Healthcare System. A national board of clinical experts will advise the development of a clinical decision support tool to be tested using mixed methods among panels of obstetricians, internists, and cardiologists. Building on the existing evidence connecting APOs to CVD and leveraging the knowledge that CVD prevention should begin early, this work will create a tailored tool to help clinicians guide the millions of women with an APO history to appropriate screening, prevention, and referral to reduce their elevated risk of future CVD.
