Abstract P428: A Polyexposomic Risk Score for Hypertensive Disorders of Pregnancy Using External Exposome Data

Abstract

Hypertensive disorders of pregnancy (HDP) are one of the most common pregnancy complications that lead to increased maternal and infant morbidity and mortality. While effective prevention of HDP can be attained using low-dose aspirin in high-risk pregnant women, identification of these high-risk women has been ineffective. Existing risk assessment models of HDP predominantly rely on biomarkers such as angiogenesis-related factors which are not routinely tested at the population level. The external environment has been shown to play an important role in the development of HDP, yet no predictive model of HDP has been developed considering the totality of the external environment, or the external exposome. To address this, we developed a polyexposomic risk score of HDP using external exposome-wide data among 6,989 women from the 2007-2012 linked Florida Vital Statistics Birth Records and Pregnancy Risk Assessment Monitoring System data. A total of 5,510 external exposome factors characterizing women’s surrounding natural, built, and social environment during pregnancy (i.e. climate, air pollution, noise, greenness, walkability, food access, socioeconomics, social capital, housing, and safety) were collected, harmonized, integrated, and spatiotemporally linked to pregnant women based on their geocoded residential addresses and pregnancy periods. The data were randomly divided into training (80%) and testing (20%) sets. A gradient boosting decision trees model was trained using the CatBoost library with hyper-parameters tuned by a grid search based on 5-fold cross validations. The model has a test-AUC (areas under curve) of 0.735. The most important features for predictions include neighborhood socioeconomic status, housing characteristics, meteorology factors, and air pollutants. External exposome data can provide important predictive information to identify individuals at high risk of HDP. The model is feasible for implementation at the population level to help guide precision preventions. Future efforts are warranted to integrate external exposome data with electronic health record data to further improve the model performance.