DISTRIBUTED LAG MODELS FOR ESTIMATING ACUTE EFFECTS OF MIXED ENVIRONMENTAL EXPOSURES IN THE CASE-CROSSOVER DESIGN

We present a Bayesian modeling framework designed to estimate the immediate effects of combined environmental exposures on suicide risk within a case-crossover design. Our method addresses a limitation observed in current distributed lag modeling approaches for multiple environmental exposures, which primarily focus on cohort or case-control data rather than casecrossover design. We utilize sparsity-enforcing spike-and-slab priors for variable selection, allowing the identification of significant exposures linked to the health outcome. To address clustered observations, we integrate random effects into the model. Additionally, we enhance computational efficiency and reduce dimensionality by implementing cubic polynomial reduction on the distributed lag surface. In a simulation study comparing our dimension reduction approach with a method estimating full model parameters without dimension reduction, we evaluated two referent schemes (unidirectional and bidirectional). The results demonstrate that our strategy, incorporating dimension reduction, outperforms full model parameter estimation in terms of false discovery rate, power, and mean squared error. We applied our framework to real-world data examining the association between a mixture of ambient environmental exposures and suicide risk in Utah.