Bayesian G-Computation for Estimating Impacts of Interventions on Exposure Mixtures: Demonstration With Metals From Coal-Fired Power Plants and Birth Weight

Abstract

The importance of studying health impacts of exposure mixtures is increasingly recognized but presents many methodological and interpretation difficulties. We used Bayesian g-computation to estimate effects of a simulated public health action on exposure mixtures and birthweight in Milwaukee, Wisconsin in 2011-2013. We linked birth records data with census tract level air toxics data from the U.S. Environmental Protection Agency’s National Air Toxics Assessment model. We estimated the difference in observed versus expected birthweight that theoretically would have followed a hypothetical intervention to reduce exposure to 6 airborne metals by decommissioning 3 coal-fired power plants in Milwaukee County prior to 2010. Using Bayesian g-computation, we estimated a 68 gram (95% credible interval: 25, 135) increase in birthweight following this hypothetical intervention. This example demonstrates the utility of our approach for using observational data to evaluate and contrast possible public health actions. Additionally, Bayesian g-computation offers a flexible strategy to estimate the effects of highly correlated exposures, addressing statistical issues such as variance inflation, and conceptual issues such as the lack of interpretability of independent effects.

Metabolic Signatures of Youth Exposure to Mixtures of Per- and Polyfluoroalkyl Substances: A Multi-Cohort Study

“…To examine associations between PFAS mixtures and each metabolite, we performed a metabolome-wide association study (MWAS) using a Bayesian hierarchical regression modeling approach with g-computation (BHRM-g). We implemented a Bayesian g-computation approach 66 , 67 to obtain both individual PFAS-specific estimates conditional on all other PFAS in the model and a single mixture effect estimate for the overall PFAS mixture. The approach is similar to quantile g-computation.…”

“… 68 Specifically, BHRM-g combines: a ) a g-prior specification for the corresponding exposure effects to provide robust estimation of highly correlated exposures, 69 b ) a Bayesian stochastic selection procedure to estimate the posterior inclusion probability (PIP) of each PFAS in the PFAS mixtures, 67 and c ) Bayesian g-computation in a potential outcome framework for estimating the overall mixture effect based on two hypothetical exposure profiles (explained in further detail below). 67 In contrast to Bayesian Kernel Machine Regression, which uses Gaussian process regression and models a nonlinear dose–response relationship, BHRM-g estimates a mixture effect from the additive terms from a regression model with each exposure and forces explicit specification of nonlinear effects. 70 We modeled the dose–response with a linear function that allows for the calculation of a single monotonic effect estimate that is not dependent on the baseline exposure profile; this aids in interpretation and allows for additional downstream analysis.…”

Response to “Comment on ‘A Quantile-Based g-Computation Approach to Addressing the Effects of Exposure Mixtures’”

“…Regarding the second point, it helps to note that QGC is a special case of g-computation and was motivated by our work in settings of complex exposure settings (Keil et al 2020a(Keil et al , 2018a(Keil et al , 2018b(Keil et al , 2014Keil and Richardson 2017). These methods are grounded in theory and applied examples (e.g., Taubman et al 2009) that support their use for estimating joint effects of multiple exposures.…”

Invited Perspective: Metal Mixtures and Child Health: The Complex Interplay of Essential and Toxic Elements

“…It also requires statistical methodologies that can simultaneously investigate nonlinear associations and interactions between metals and elements whose combined effects may be in opposite directions. In recent years, statistical methods have emerged to address some of these needs, as there is general recognition that the field of environmental epidemiology must move beyond a singlepollutant modeling framework (Keil et al 2021;Tanner et al 2020). Zhang et al (2021) report that prenatal exposure to the trace elements selenium (Se) and manganese (Mn) was associated with lower systolic blood pressure in children of ages 3-15 y, whereas exposure to the metals lead (Pb), mercury (Hg), and cadmium (Cd) was not associated with either systolic or diastolic blood pressure.…”

“…It also requires statistical methodologies that can simultaneously investigate nonlinear associations and interactions between metals and elements whose combined effects may be in opposite directions. In recent years, statistical methods have emerged to address some of these needs, as there is general recognition that the field of environmental epidemiology must move beyond a single-pollutant modeling framework ( Keil et al. 2021 ; Tanner et al.…”