Several studies in the past have evaluated the impact of air pollution on coronavirus disease 2019 (COVID-19) mortality in a single country. A recent Journal of Spatial and Spatio-temporal Epidemiology study used a spatial ecological design including Canada, Italy, England, and the United States to investigate the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection risk through the residual spatial autocorrelation after accounting for covariates.
Study: Long-term exposure to air pollution and COVID-19 incidence: A multi-country study. Image Credit: BLACKDAY / Shutterstock.com
About the study
In the current study, it was hypothesized that people exposed to high air pollution for a long time have reduced natural defenses to respiratory infections, which makes them more susceptible to COVID-19. For each country, the researchers investigated the effects of long-term average exposure to particulate matter that has a diameter of fewer than 2.5 micrometers (μm) (PM2.5) and nitrogen dioxide (NO2) to determine whether these factors increase the incidence of COVID-19.
To this end, the researchers drew a comparison of geographical contrasts in air pollution and incidence rates across small areas as well. The study takes into account potential confounding factors such as lung cancer incidence as a proxy for smoking and the unemployment rate.
The current study employed three different methods. First, the air pollution interpolation, which is a geostatistical model for NO2 monitoring from 517 ground stations in the U.S. Second, the COVID-19 incidence model was used to estimate the impact of air pollution on COVID-19 cases in small areas. Third, a population attributable fraction (PAF) method for quantifying the combined impact of both PM2.5 and NO2 on COVID-19 incidences.
Results of spatial distribution and air pollution effects
The spatial distribution results estimated that the standardized incidence rate for every 100,000 people in the U.S. is much higher than in other countries. In fact, this nation is as high as a quarter of its population in some countries.
The spatial distribution results of Canada indicate that this nation has been extremely successful in controlling COVID-19, with the highest standardized incidence rate being about 1,500. The standardized incidence rate in northern Italian provinces is fairly high and was estimated to be about 3,000 with a high probability. Similarly, the northwestern provinces of England have a high exceedance probability, whereas the standardized incidence rate in the upper-tier local authorities is close to 5,000.
Any relationship between COVID-19 incidence and long-term exposure to PM2.5 and NO2 was found to be inconsistent across all four countries examined in this paper.
The analysis shows that a 1 μgm-3 increase in long-term exposure to PM2.5 increased the COVID-19 incidence rate in the U.S. to 12.6%, while it increased only moderately in Italy and England at 0.5% and 2.9%, respectively. In Canada, there is evidence of a protective effect of PM2.5, which is likely due to spatial confounding issues between the explanatory variables and the spatial random effect included in the model. All the findings on the impact of PM2.5 are also consistent with other recent studies.
Inconsistency in results across countries could also be due to some unmeasured confounders that are correlated with air quality. The U.S., for instance, is more car-dependent than the other three countries; hence, PM2.5 might be more strongly correlated with mobility in the U.S. If mobility is considered as a strong predictor of COVID-19 prevalence, failure to adjust for it as a model covariate could impact inferred COVID-19/ PM2.5 relationship in the U.S. more strongly than the other countries.
All ecological studies are inherently limited, as covariates refer to area-level and not individual-level characteristics. Moreover, area-level measurements are not reliable proxies for individual-level exposures, as individuals in a region are heterogeneous and ecological error can occur when there are dependencies and interactions at the individual level.
Therefore, in all four countries, the relationship between individual-level and area-level exposures, as well as the measured values of the area-level exposures, were different. The U.S. data obtained from its 3,108 counties provide higher-resolution spatial information as compared to the data from 93–149 regions in the remaining three countries.
Taken together, the findings from the current study demonstrate and emphasize the importance of replicating spatial analyses across multiple countries. Also, as more and more tools become available, managing spatial data on health outcomes and exposures will become simpler.
The researchers are still in the model building phase for analyses as this; therefore, before a commonly accepted methodology is agreed upon, results from several closely resembling model formulations with different confounders will have to be used and analyzed. This will help evade any possible inherent bias towards positive results.