The effect of known and unknown confounders on the relationship between air pollution and Covid-19 mortality in Italy: A sensitivity analysis of an ecological study based on the E-value

Abstract

Back in December 2019, the novel coronavirus disease 2019 (Covid-19) started rapidly spreading worldwide, especially in Italy that was among the most affected countries. The geographical distribution of air pollution and Covid-19 mortality in Italy suggested atmospheric pollution as a worsening factor of severe Covid-19 health outcomes. The present nationwide ecological study focused on all 107 Italian territorial areas, aiming to assess the potential association between Particulate Matter concentration, less than 2.5 μm in diameter (exposure), and Covid-19 mortality rate (outcome) throughout 2020, by looking at 28 potential confounders. A potential positive association between exposure and outcome was observed when performing a multivariate regression analysis with a Negative Binomial model, suggesting that an increase of 1 μg/m 3 in the exposure is associated with an increase of 9.0% (95% CI: 6.5%–11.6%) in the average Covid-19 mortality rate, conditional on all 28 potential confounders. A sensitivity analysis, based on the E-value, shows that a hypothetical unmeasured confounder would have to be associated with both PM 2.5 concentration and Covid-19 mortality rate by a rate ratio of at least 1.40-fold each to explain away the exposure-outcome association, conditional on all 28 covariates included in the main analysis model. Moreover, the Observed Covariate E-value (OCE) was reported to provide a contextualization of the E-value on the observed covariates included in the study. The OCE sensitivity analysis shows that a set of unknown confounders similar in size and magnitude to the set of the considered climatic factors could potentially explain away the estimated exposure-outcome association. Consequently, the role of climatic factors in the Covid-19 pandemic is worth of further investigation.

Bayesian profile regression to study the ecologic associations of correlated environmental exposures with excess mortality risk during the first year of the Covid-19 epidemic in lombardy, Italy

“…They showed that higher levels of long-term ambient fine particulate matter (PM2.5) air pollution were associated with increased excess mortality during the early period of the COVID-19 pandemic. Other researchers have since corroborated this association in Italy ( Aloisi et al, 2022 ; De Angelis et al, 2021 ; Ye et al, 2021 ) and other parts of the world ( Bourdrel et al, 2021 ; Katoto et al, 2021 ; Kogevinas et al, 2021 ; Tchicaya et al, 2021 ). However, exposure to ambient PM2.5 does not occur in isolation from other co-occurring air pollutants.…”

“…Notably, the data showed that PM2.5 effects varied significantly by exposure profile clusters, with strong positive associations between PM2.5 and COVID-19 deaths in clusters with higher-income economies of Western Europe and North America (including Italy). A nationwide mortality study in Italy in 2020 found that omitting ambient temperature from multivariable negative binomial regression models resulted in a downward (attenuation) bias of the PM2.5 effect estimate on mortality in the first year of the pandemic ( Aloisi et al, 2022 ).…”

“…The significant spatial heterogeneity of Covid-19 case fatality rates suggests that a confluence of spatial determinants of health can make some sub-populations more vulnerable to severe Covid-19 outcomes. Extensive epidemiological data has emerged showing that different place-based factors, such as underlying distributions of age and sex of the population, access to health care or health facilities, ambient temperature, and air pollution, appear to explain some of the observed spatial heterogeneity of Covid-19-related mortality ( Aloisi et al, 2022 ; Biggeri et al, 2020 ; Blangiardo et al, 2020 ; Calderón-Larrañaga et al, 2020 ; Coker et al, 2020 ; Conti et al, 2020 ; De Angelis et al, 2021 ; Gibertoni et al, 2021 ; Kejžar and Lusa, 2020 ; Mannucci et al, 2020 ; Michelozzi et al,020; Sandrini et al, 2020 ; Scortichini et al, 2021 ; Ye et al, 2021 ).…”

A correlational analysis of COVID-19 incidence and mortality and urban determinants of vitamin D status across the London boroughs

“…Although the epidemiology of COVID-19 is still developing, there is substantial evidence linking the pathological characteristics of COVID-19 critical illness and the causes of death in COVID-19 patients with the conditions caused and/or exacerbated by long-term exposure to air pollutants like NO 2 and PMs. The hypothesis of a COVID-19 air pollution link has been supported by a fast-growing body of literature reporting evidence of positive association between outdoor air pollution and COVID-19 morbidity and mortality in different parts of the world 18 . It is reasonable to suggest that since chronic air pollution has a negative impact on the cardiovascular and respiratory systems, and immune function, it increases the risk of mortality, exacerbates COVID-19 infection symptoms and worsens COVID-19 patients' prognosis 19 – 21 .…”

Long-term air pollution exposure and COVID-19 case-severity: An analysis of individual-level data from Switzerland

“…As a result, most of the studies assessing the effects of exposure to air pollution have applied an ecological study design; i.e., the air pollution estimates were averaged over the same level of spatial aggregation as the COVID-19 data and these aggregates were compared to the COVID-19 incidence, deaths, and/or case fatality rates. Examples include descriptive analyses based on several correlation indices (such as Pearson and Spearman) between the COVID-19 outcomes and the exposures to different air pollutants in separate cities or countries around the world ( Bashir et al, 2020 ; Daes et al, 2021 ; Fatorini and Regoli, 2020 ; Telo-Leal and Macías-Hernandez, 2021 ; Zorn et al, 2020 ; Zoan et al, 2020 ), regression analyses evaluating the association between air pollution exposures and COVID-19 incidence, severity, and lethality, such as simple linear regression models ( Li et al, 2020 ), multivariate Poisson ( Jiang et al, 2020 ) and negative binomial ( De Angelis et al, 2021 ; Aloisi et al, 2022 ) regression models that account for demographic, socio-economic, and meteorological variables; generalized additive models (GAM) ( Zhu et al, 2020b ) and hierarchical multiple regression models ( Coccia, 2020 ).…”