Model choice for estimating the association between exposure to chemical mixtures and health outcomes: A simulation study

Abstract

Challenges arise in researching health effects associated with chemical mixtures. Several methods have recently been proposed for estimating the association between health outcomes and exposure to chemical mixtures, but a formal simulation study comparing broad-ranging methods is lacking. We select five recently developed methods and evaluate their performance in estimating the exposure-response function, identifying active mixture components, and identifying interactions in a simulation study. Bayesian kernel machine regression (BKMR) and nonparametric Bayes shrinkage (NPB) were top-performing methods in our simulation study. BKMR and NPB outperformed other contemporary methods and traditional linear models in estimating the exposure-response function and identifying active mixture components. BKMR and NPB produced similar results in a data analysis of the effects of multipollutant exposure on lung function in children with asthma.

Using non-parametric Bayes shrinkage to assess relationships between multiple environmental and social stressors and neonatal size and body composition in the Healthy Start cohort

“…The exposures included in our data set are moderately to highly correlated, and a single-exposure model is likely not sufficient to address this correlation. Using NPB as a modeling approach allowed us to consider each of these exposures at once in a framework that allows correlated exposures to be assigned similar coefficients when effects cannot be separated [ 43 ].…”

“…Additionally, LASSO does not account for uncertainty in variable selection [ 42 ]. In contrast, NPB can cluster and assign equal regression coefficients to exposures, which reduces the effect of correlation among exposures and is a strength when it is difficult to separate the effects of the two exposures [ 43 ]. In future cases where more than two exposures are identified, methods like Bayesian Kernel Machine Regression (BKMR) may be useful.…”

“…In future cases where more than two exposures are identified, methods like Bayesian Kernel Machine Regression (BKMR) may be useful. BKMR is a flexible approach for modeling the effects of exposure mixtures that is capable of identifying nonlinear main effects and interactions among mixture components [ 98 ], though it may not be well suited for selecting among a large number of predictors [ 43 ], Thus, a two-stage approach where NPB models reduce the number of candidate predictors and methods like BKMR explore the complex interactions between exposures of interest might be most appropriate.…”

“…For this data set, we needed to identify a statistical method that would be capable of handling variable selection among potentially highly correlated exposures within the context of interactions between exposures and between exposures and covariates. NPB is designed to facilitate selection among a large number of correlated predictors in a single model, thus having an advantage over traditional multipollutant models [ 42 , 43 ]. We compared the results of our NPB model to single-exposure models (linear regression models).…”

Statistical methods for chemical mixtures: a roadmap for practitioners

“…In contrast, simulation studies are a powerful tool for comprehensively comparing various methods under a broad range of data generating mechanisms. Some sporadic works on simulation studies for mixtures analysis [27, 28] have emerged, but there still lacks a systematic evaluation of the popular methods used in mixture analyses under diverse data scenarios with varied sample sizes, changing number of pollutants for continuous and binary outcomes. To address this gap, our goal is to utilize simulation studies to perform a head-to-head comparison among mixtures methods under different data settings and provide guidance for practitioners.…”