Conference Program

Estimation and False Discovery Control for the Analysis of Environmental Mixtures (303651)

Joseph Antonelli, University of Florida 
*Srijata Samanta, University of Florida 

Keywords: False discovery control, Environmental mixtures, High-dimensional statistics

The analysis of environmental mixtures is of growing importance in environmental epidemiology, and one of the key goals in such analyses is to identify exposures and their interactions that are associated with adverse health outcomes. Typical approaches utilize flexible regression models combined with variable selection to identify important exposures and estimate a potentially nonlinear relationship with the outcome of interest. Despite this surge in interest, no approaches to date can identify exposures and interactions while controlling any form of error rates with respect to exposure selection. Often environmental mixtures have moderate to weak signals with outcomes of interest, so it is important to control the FDR to maximize power to detect small effects while avoiding false positives. We propose two novel approaches to estimating the health effects of environmental mixtures that simultaneously 1) estimate and provide valid inference for the overall mixture effect, and 2) identify important exposures and interactions while controlling the FDR. In the first approach we extend the ideas of debiased lasso to the group setting with interactions. In the second approach we adapt the knockoffs procedure for model selection in the first stage and use data splitting procedure in the second stage for providing inference. We show that our approaches are far faster computationally than most of the state-of-the-art approaches. Moreover, the ability to control FDR leads to substantial gains in power to detect weak signals, without sacrificing in terms of estimation accuracy or inference on the resulting mixture effect. We also have improved exposure/interaction selection relative to existing approaches that do not provide explicit FDR control. To further illustrate our approaches we apply them to a study of persistent organic pollutants (data from NHANES) and find that controlling the false discovery rate leads to substantially different conclusions.