Methods for studying the health effects of exposure mixtures like air pollutants, water contaminants, and consumer product mixtures have been the subject of high levels of research effort and resources in recent years. Demateis et al. (Am J Epidemiol. XXXX;XXX(XX):XXXX-XXXX)) demonstrate a novel set of extensions to one of these approaches, Bayesian kernel machine regression (BKMR) to study effect measure modification (EMM). In this commentary, I discuss the necessity of flexible regression methods like BKMR in the context of modern causal inference and express optimism that the continued growth of methods like BKMR can improve the link between epidemiologic data and public health actions. I expand on traditional motivations for studying EMM and comment on how they might be sometimes ill-suited for exposure mixtures. I close by remarking on how health impacts of unequal distributions of harmful exposures are masked by standard EMM-based approaches and describe how such methods, including those of Demateis et al., can be leveraged to incorporate such ideas into analysis to improve connections with public health.
Alexander P. Keil (Fri,) studied this question.