S ocial epidemiologists are concerned with understanding the relationship between multilevel (e.g., individual, contextual, and institutional) social, economic, and political exposures on health and health disparities. 1 These many exposure levels and determinants bring complexity: co-occurring exposures may have independent, aggregate, synergistic, or antagonistic effects on health. This creates challenges in articulating a research question, choosing a relevant estimand, and obtaining unbiased and informative measures of association, among others. Social epidemiologists often develop and/or employ indices-such as the Social Vulnerability Index, 2 the Townsend Index, or Index of Concentration at the Extremes 3 -to study the association of often correlated place-based determinants of health, some of which use dimension reduction techniques. We also use multilevel models to parse the independent impacts of multiple levels of exposure, taking theory and data-informed approaches to determine which variables to incorporate in models and how to model their interactions. 1,4 Environmental epidemiology has similar challenges with complex and dynamic exposures. Historically, the field focused on identifying the effects of single or aggregate exposures while occasionally controlling for the influence of co-occurring exposures. 5 In recent years, a plethora of research and new methods have been deployed to better study the effects of multiple, often correlated exposures. 6 This direction in environmental health research is motivated partly by the recognition that people are always exposed to multiple ("mixtures" of) pollutants simultaneously, 7 and that single exposure models may derive biased or invalid estimates of association with health outcomes that are impacted by these co-occurring exposures. 8 In this issue, Bather et al. 9 introduce social epidemiologic audiences to mixture models and encourage their wider adoption in the field. The authors then consider the application of a flexible mixture method, Bayesian Kernel Machine Regression (BKMR), in a study of multilevel (individual, social, and structural factors) exposures and associations with psychological distress, among individuals with at least one police arrest. The authors introduce and explain mixture models, and we appreciate their encouraging this approach for complex statistical challenges in social epidemiology.Bather et al. 9 focus on the opportunities for mixture modeling, and specifically BKMR, for social epidemiologic analysis. We believe there are scenarios where the benefits of mixture models, including BKMR, may provide improvements over standard regression approaches commonly applied in social epidemiology. However, we feel the need to add caution to their optimism. While promising for social epidemiology research, 10 the structure of exposures and the goals of the inquiry warrant consideration when deciding
