Identifying county-level effect modifiers of the association between heat waves and preterm birth using a Bayesian spatial meta regression approach.

High temperature is associated with adverse health outcomes, particularly for vulnerable subpopulations including pregnant individuals and their unborn babies. Several recent studies have investigated the association between temperature and preterm birth at different geographic scales and across different spatial locations. However, there has been less focus on characterizing spatial heterogeneity in risks and identifying modifiable factors that contribute to the observed variation. In this work, we carry out a two-stage modeling approach to (i) estimate county-level short-term associations between heat waves and preterm birth across eight states in the United States and (ii) explore county-level factors that modify these associations using a newly developed hierarchical Bayesian spatial meta-regression approach. Specifically, we extend the traditional meta-regression framework to account for spatial correlation between counties within the same state by modeling the effect estimates using a variant of the conditional autoregressive model. We report several variables that modified the associations between heatwaves and preterm birth, including housing quality, energy affordability, and social vulnerability for minority status and language barriers. An R package, SpMeta, is developed for analyses that aims to synthesize area-level risk estimates while accounting for spatial dependence.