BACKGROUND AND AIM: Previous research suggests that intake of antioxidants during pregnancy may reduce the harmful effects of air pollution exposure. However, examining effect modification in the context of multiple air pollutants and antioxidants can present a number of challenges. Prior to conducting an epidemiologic study to understand the relationship between air pollution, antioxidants and birth defects, we aim to compare methods for identifying effect measure modification within higher-dimensional exposure and dietary data. METHODS: We first perform a simulation, based on a real-word case-control study of birth defects, to generate realistic data on exposure to criteria air pollutants in the first trimester of pregnancy, maternal dietary intake of antioxidants and birth outcomes. We construct multiple realistic scenarios, varying levels of correlation between features, the magnitude of association between air pollutants, antioxidants and congenital heart defects and the amount of effect measure modification by antioxidants. Using these data, we compare two data-driven methods, the Deletion-Substitution-Addition (DSA) algorithm and boosted regression trees, to the traditional use of multiple logistic regression models for evaluating effect measure modification. RESULTS:This talk will describe the performance of each method across the various exposure-outcome scenarios. CONCLUSIONS:Results from this simulation study will subsequently be applied to data from the National Birth Defects Prevention Study to investigate whether evidence exists for antioxidant intake during pregnancy to modify the relationship between early-pregnancy air pollution exposure and congenital heart defects in offspring. KEYWORDS: pregnancy, methods, environment-diet interactions
Read full abstract