SEMIPARAMETRIC METHODS FOR MODELING OF TIME-DEPENDENT ENVIRONMENTAL EXPOSURES

SEMIPARAMETRIC METHODS FOR MODELING OF TIME-DEPENDENT ENVIRONMENTAL EXPOSURES Abstract Countless environmental exposures throughout the human life course and influence human development and susceptibility to disease. Assessing how multiple exposures synergize or antagonize each other to affect human health is a powerful approach to understanding the all-around impact of environmental exposures. Such approach will be needed in the new paradigm of exposome research, which encompass all non-genetic causes for diseases, from external natural to social environments and from internal macro- to microenvironments. However, many environmental investigations are challenged by the special data structure and characteristics of time-dependent exposures. In particular, multiple environmental exposures: 1) are inter- correlated; 2) are time-dependent; 3) exhibit time-varying effects; 4) have heterogeneous effects; and 5) demonstrate complex and nonlinear exposure-response relationships. The limiting availability of statistical models and analytical tools to handle these challenges hinder our ability to make inference or draw conclusions about the effects of multiple exposures on human health. Building upon our experience developing statistical methodologies and motivated by the challenges encountered in many of our collaborations in environmental research, we propose to develop and implement novel statistical methods to address important scientific questions in environmental health research, specifically through the following aims: 1) evaluate the effects of time-dependent environmental exposures on time-invariant health outcomes and identify critical windows of vulnerability; 2) characterize the effects of time-dependent environmental exposures on time-to- event and longitudinal outcomes; 3) investigate the heterogeneous impacts by environmental exposures on subpopulations; and 4) develop, distribute, and support open-source software packages for the proposed methods. All proposed models, estimation, and testing procedures will be investigated through analytical approaches, theoretical inference, numerical simulations, and applications to multiple datasets from approved human-subject studies. Upon its completion, the proposed project will provide new statistical methods that both support ongoing collaborations to address critical scientific questions and enable the environmental health research community to better assess the impact of time-dependent exposures.