Estimating the impact of short-term sulfur dioxide exposure on the COVID-19 incidence: A spatiotemporal study in the continental United States

Abstract

Numerous studies have investigated the association of short-term SO2 exposure with the incidence of COVID-19, but few studies have focused on the spatiotemporal heterogeneity in the associations, which are crucial for identifying high-vulnerability regions and periods of the epidemic and formulating cost-effective prevention policies. We used a recently proposed multi-stage approach to explore this issue, with the continental US serving as an example. Firstly, rough state-specific SO2-COVID-19 associations were independently estimated by time-series generalized additive model (GAMs). A Leroux-prior-based conditional autoregression (LCAR) was used to spatially smooth the estimations. A time-varying GAM-LCAR analytical process was used to obtain the temporal change of association. Finally, a spatiotemporal LCAR was used to explore whether vaccination rates and temperature were the effect modifiers. The results showed significant positive associations in the middle and northeastern regions with relative risks (RRs) varying from 1.10 to 1.25, weak positive associations in the western and middle-eastern regions with RRs less than 1.10, and insignificant associations in the southeastern regions. On average, a 1-ppb increase of SO2 corresponded to a 6.25% (95%CI: 3.71%–9.00%) increase in the cases of COVID-19. In the time dimension, the SO2-COVID-19 association exhibited no significance from April 2020 to the end of 2020, and then gradually rose from January 2021 to December 2021. The vaccination rate was recognized as a significant modifier with a stronger association at a higher rate. Such spatiotemporal pattern suggested that middle and northeastern regions and the post-pandemic era deserved more attention and targeted cost-effective policies.