Air pollution mixtures and pregnancy: assessing exposure, estimating risk, and predicting susceptibility with machine learning

About This Grant

PROJECT SUMMARY/ABSTRACT Air pollution is a leading health threat, and 131 million people in the US live in a county with unhealthy levels of air pollution. Pregnancy is a period of increased susceptibility to the effects of air pollution, which is a complex mixture of hazardous chemicals. Common pollutants, including fine particulate matter (PM2.5), ozone (O3), nitrogen dioxide (NO2) and excess heat, have each been associated with placental insufficiency. This condition affects half a million pregnancies annually in the US and is a major source of perinatal morbidity and mortality. PM2.5, O3, NO2, and heat may interact to potentially exacerbate risk of placental insufficiency, but most methods lack the flexibility and interpretability to characterize this risk. The goal of this study is to determine whether mixtures of PM2.5, O3, NO2, and heat synergistically increase risk of placental insufficiency during pregnancy. This will be done using an emerging machine learning method for causal inference, which adjusts for confounders and calculates confidence intervals. Residential exposure to ensemble-modeled 1 km2 estimates of these pollutants are available for all 9,447 participants in our existing prospective pregnant cohort. In AIM 1A, we use the causal random forest algorithm to estimate the effects of mixtures exposure on placental insufficiency for each gestational week, accounting for time-to-event structure. Variability in these effects will be characterized in AIM 1B with an uplift model, which will describe effect modification across body mass index, a risk factor for placental insufficiency. Although residential air pollution exposure is commonly used in health models, this exposure assignment contributes to uncertainty in the health effects of air pollution. In AIM 2A we will use a microsimulation activity space model developed at Oak Ridge National Laboratory to create simulated movement patterns for our pregnant cohort. The effect of activity space exposures on risk of placental insufficiency will be compared against the effect of residential exposures in AIM 2B. This study will provide insight into the effects of air pollution mixtures on placental insufficiency, as well as effect modifiers and uncertainty. The results could alter our conclusions about the safety of air pollution during pregnancy. Training will take place at the University of Utah and Oak Ridge National Laboratory under the mentorship of experts in maternal-fetal medicine, atmospheric science, machine learning, computation, and trustworthy data science. Through this training plan, the applicant will develop the foundational skills to prepare for an academic career dedicated to studying maternal air pollution exposure with advanced methods.