Statistical Methods Development and Applications to Perinatal Epidemiology

In many epidemiological studies, an index might be an efficient and parsimonious way to capture the effect of a group of exposures (exposure mixture). In these cases, my group has been developing Bayesian additive regression tree–based approaches that allow heterogeneity of the effect of an index. We are also exploring machine-learning and artificial-intelligence methods for the purpose of predicting birthweight. Improved prediction of birthweight has important implications for managing delivery of fetuses that are too large and for taking preventive measures for fetuses that are too small. In diagnostic accuracy, we have continued in ROC curve (visualization designed for evaluating the performance of a machine learning classification system) modeling under various constraints.