Correlated confounders and adjustment approaches in mixtures analysis contexts

Presentation of ongoing work examining modeling choices related to confounding control in the context of an exposure mixture, using epidemiology mixtures methods. 

Impact/Purpose

Assessments of hazards posed by exposures currently focus on single and co-pollutant adjusted models, and it is critical to more fully understand the performance of mixtures methods as they become more common to allow for their future incorporation into health assessments which inform policy guidance. Given this gap in knowledge concerning important aspects of two popular mixtures methods, in the present analyses we investigate potential for bias while using WQS and QGComp in multiple situations using simulated data, illustrating potential challenges in analysis with directed acyclic graphs (DAG). We first compare performance with DAG-guided models of varying propriety - a minimally sufficient set, then two sufficient sets. We then follow with other models that could potentially be run, though not DAG guided – a model including both confounders as well as covariates not correlated with any other variables, and a model including all available variables in the model.

Citation

Jardel, H., A. Krajewski, K. Rappazzo, AND A. Keil. Correlated confounders and adjustment approaches in mixtures analysis contexts. NiMBioS Morsels, Virtual, TN, April 18, 2024.