A quantile-based g-computation approach to addressing the effects of exposure mixtures
share
Exposure mixtures frequently occur in epidemiologic data, particularly in the fields of environmental and nutritional epidemiology. Various strategies have arisen to answer questions about exposure mixtures, including methods such as weighted quantile sum regression that estimate a joint effect of the mixture components. Little is known about the performance of weighted quantile sum regression for estimating such effects, however, and even less is known about the benefits or drawbacks of the underlying assumptions of weighted quantile sum regression. We demonstrate a new approach to estimating the joint effects of a mixture: quantile g-computation. This approach combines the inferential simplicity of weighted quantile sum regression and the immense flexibility of g-computation, a method of causal effect estimation. We demonstrate that weighted quantile sum regression can be considered a special case of quantile g-computation, and that quantile g-computation often provides improved inference at sample sizes typically encountered in epidemiologic studies, and when the assumptions of weighted quantile sum regression are not met. We examine, in particular, the impacts of large numbers of non-causal exposures, exposure correlation, unmeasured confounding, and non-linearity of exposure effects. We show that, counter to intuition, quantile g-computation estimates can become more precise as exposure correlation increases. Quantile g-computation appears robust to many problems routinely encountered in analyses of exposure mixtures. Methods, such as quantile g-computation, that can yield unbiased estimates of the effect of the mixture are essential for understanding the effects of potential interventions that may act on many components of the mixture, and our approach may serve as an excellent tool for quantifying such effects and bridge gaps between analysis and action.
READ FULL TEXT