|
Abstract:
|
Unmeasured, spatially-structured covariates can confound associations between spatial exposures and health outcomes. Recent work has demonstrated that when the spatial scale of variability in an unmeasured confounder exceeds the spatial scale of variability in the exposure, this confounding can be adjusted for. Adding thin plate regression splines to a regression model is an attractively simple approach for making the adjustment, but it does not provide an easily interpretable or tunable scale for the adjustment. We propose using frequency-based basis functions to adjust for confounding at a tunable spatial scale. We compare this approach to other adjustment methods through simulations and demonstrate its use in an analysis of air pollution exposure and cardiovascular health outcomes in a national cohort.
|
