We develop novel statistical techniques for nonparametric Bayes analysis of high-dimensional covariate data, directly motivated by the largest population-based study ever conducted on the causes of birth defects. The methods we develop will enable borrowing of information and shrinkage across high-dimensional environmental, biomedical, pharmacological, and sociodemographic risk factors (and interactions among them) and across a multitude of birth defects, many of which are too rare to be studied in isolation. Using a hierarchical structure directly motivated by embryonic development, the borrowing of information can be informed by our knowledge of mechanistic development of the embryo. These novel methods may significantly impact the study of rare congenital malformations. The methods to be developed have broad application in public health and medicine, where exposures or characteristics of interest may be great in number and interactions are important, such as the examination high-dimensional gene by environment and gene-gene interactions.
This project addresses a critical need of finding clues to the etiology and pathogenesis of congenital mal- formations, using data from the largest population-based study ever conducted on the causes of birth defects. While birth defects are the leading cause of infant mortality, the leading cause of death among children aged 1-4, and the fifth-ranked cause of premature mortality in the United States, many individual defects are too rare to be studied comprehensively, even in studies that are very large. Our new statistical methods for sparse shrinkage incorporate current knowledge of embryonic development and allow some borrowing of information across differ- ent birth defects while keeping each defect as a separate entity of interest in the statistical model. These novel methods will allow investigators to investigate the simultaneous influence of multiple exposures and combinations of exposures on multiple outcomes.
Showing the most recent 10 out of 20 publications
