Investigation of potential collider bias in estimating the association between long-term exposure to air pollution and COVID-19 mortality.

Background: Patient-based cohorts were frequently used to investigate air pollution-related coronavirus disease 2019 (COVID-19) evidence, which can be subject to collider bias. However, this bias has not been explored. We aimed to quantify and adjust the collider bias by limiting study population to patients with COVID-19 when estimating the association between long-term exposure to air pollution (LTAP) and COVID-19 mortality.

Methods: In a register-based cohort study including 3,721,813 residents aged 30 or older in Denmark, we followed them from 1 March 2020 to 26 April 2021. We estimated the hazard ratios of COVID-19 mortality associated with LTAP. We calculated the relative hazard ratios (RHR) by comparing the hazard ratios of COVID-19 mortality among the general population and patients (infected or hospitalized) to quantify the impact of collider bias, and further applied inverse probability weighting (IPW) to adjust the potential collider bias.

Result: We detected 138,742 positive for SARS-CoV-2, 11,270 COVID-19 hospitalizations, and 2557 deaths from COVID-19 during the study period. Although the demographic and socioeconomic characteristics differed among the three populations (general population, infected individuals, and hospitalized patients), infected and hospitalized patients experienced higher air pollution exposure compared with the general population. We observed greater associations of exposure to air pollution with COVID-19 mortality in the general population compared with the COVID-19 infected and hospitalized patients, with RHR of 0.88 (0.82, 0.95) and 0.85 (0.74, 0.97) per 0.55 µg/m increase in fine particulate matter (PM) when we limited to infected and hospitalized patients, respectively. Similar association was also observed with nitrogen dioxide exposure. After applying IPW, we observed moderate increase in the estimated association, with RHR altered to null.

Conclusion: Our findings suggest that associations between LTAP and COVID-19 mortality are likely underestimated in patient-based cohorts due to potential collider bias, while IPW could be a useful tool to adjust such bias.