Primary Submission Category: Generalizability/Transportability
Statistical Methods for Transporting the Effect of an Environmental Mixture Across Populations
Authors: Melanie Mayer, Adele Ribeiro, Brent Coull, Ana Navas-Acien, Elias Bareinboim, Linda Valeri,
Presenting Author: Melanie Mayer*
We are constantly exposed to multiple environmental exposures which work together to cause unique outcomes. To estimate their effect, observational data from a different population from which we are interested in is often what is available. Transportability has received considerable attention, however, complexities associated with multiple, continuous exposures, as we see in environmental mixtures, require one to take extra precautions. We formalize the assumptions needed and assess methods for testable assumptions, such as overlap of exposure levels and effect modifiers across populations. We also develop a statistical approach for estimating a transported effect via a flexible, machine learning model with weighting which accounts for potential non-linear/interaction effects and skewness/correlations of exposures. We demonstrate its applicability by providing a real-world application example where we analyze the effect of exposure to multiple metals on a health outcome in a multisite cohort study. We apply the methods to explain heterogeneities across study sites and to transport effects to a target population to which we observe covariate and exposure data, but not necessarily outcome information. The developed environmental mixtures transportability framework adjusts effects based on a population’s exposure/covariate distributions, expanding the data sources available for estimating environmental mixture effects for a target population of interest.
Â