Methods/Statistics
Analysis of Complex Mixtures Using Flexible Bayesian Quantile-Based G-Computation Maria Kamenetsky* Maria Kamenetsky Kamenetsky Kamenetsky Kamenetsky Kamenetsky Kamenetsky Kamenetsky Kamenetsky National Cancer Institute
Methods for environmental mixtures (“mixtures analyses”) consider the overall effect of multiple concurrent exposures instead of solely one exposure at a time. Mixtures analyses are particularly important for environmental exposures as overall effects can differ from effects of any single exposure alone. Data on these exposures are often collected in small sample sizes and exhibit complex dependence structures, such as spatial or temporal correlation and hierarchical exposure groupings. However, commonly-used mixtures methods that target a single overall effect are limited in their ability to formally model these features, potentially results in biased inference and poor prediction. To address this limitation, we develop a novel flexible Bayesian quantile-based g-computation (BQGC) approach to mixtures. This Bayesian framework allows for the integration of more data complexities and common characteristics of environmental mixtures. We demonstrate our approach using dietary data from the New England Bladder Cancer Study (NEBCS), a large population-based case-control study. Exposure to arsenic, a known carcinogen, is primarily through contaminated drinking water and diet. Arsenic is metabolized into dimethylarsinic acid (DMA), monomethylarsonic acid (MMA), and inorganic arsenic (iAs), with DMA comprising the largest proportion in human urine. We explore the association between a mixture of dietary exposures and urinary DMA in the region among control participants (N=1,109). Using BQGC with integrated spatial dependence (BQGC+Spatial), those exposed to the dietary mixture had urinary DMA values that were 93% higher than compared to those never exposed (95% credible interval: 32%, 187%), with no evidence of residual spatial autocorrelation. Our innovations improve the accuracy of statistical evidence of the mixture’s effect on disease risk and expand the flexibility of mixtures analysis in public health.
