Reviewers: Reviewers: Sophie M. Woodward, Xiao Wu | 📒📒📒◻️◻️
Reviewers: Sophie M. Woodward, Xiao Wu |📒📒📒◻️◻️
RR:C19 Evidence Scale rating by reviewer:
Potentially informative. The main claims made are not strongly justified by the methods and data, but may yield some insight. The results and conclusions of the study may resemble those from the hypothetical ideal study, but there is substantial room for doubt. Decision-makers should consider this evidence only with a thorough understanding of its weaknesses, alongside other evidence and theory. Decision-makers should not consider this actionable, unless the weaknesses are clearly understood and there is other theory and evidence to further support it.
Review: The manuscript analyzes a large individual-level dataset to determine associations between long-term air pollution and COVID-19 mortality in London. An increased risk of dying from COVID-19 with increased exposure to air pollution was found when only adjusting for age, sex, and vaccination status in the regression model. Such a positive
association is not maintained after additionally adjusting for geographic factors (such as population density), ethnicity, and deprivation levels. Overall, the results seem potentially informative.
The strengths of this study derive from its large and representative population size (737,356 individuals who tested positive for COVID-19, including 9,315 COVID-19 related deaths), consideration of multiple pollutants, and comprehensive data linkage through the Office for National Statistics Public Health Data Asset (ONS-PHDA). The authors use Cox proportional hazard regression models to estimate the hazard ratios of dying from COVID-19 associated with increases in long-term (specifically, 2016 annual average) NO2, NOx, PM10, and PM2.5 concentrations separately. While adjusting only for age, sex, and vaccination status, the authors found a significant increased risk of COVID-19 mortality with increased air pollution exposure; but this association became non-significant when adjusting additionally for population density, rural/urban variables, ethnicity, adjusted household deprivation, and stratifying by local authority codes (the latter model is termed their ‘main model’).
Some aspects of this paper require further clarifications: First, the authors emphasize that after adjustment for ethnicity and deprivation the hazard ratios were reduced to unity, but this may be due to the sequential order of the models fit. The hazard ratios first became insignificant after geographic factors (population density, rural/urban variables, and local authority codes) were added to the model and remained insignificant when ethnicity and deprivation were subsequently added. The following hypothesis can not be ruled out: these geographic covariates capture most of the variation in air pollution, thus reducing the association of exposure and outcome to null when they are included in the model. While such a phenomenon is unlikely, it merits further discussion/exploration by the authors as to why controlling for these geographic covariates may influence the statistical results. Second, the confidence intervals of the hazard ratios become wider after accounting for geographic confounders. The discussion primarily focuses on ethnicity and deprivation as the culprits behind the null association, potentially overlooking other explanations of the findings. It can be helpful to see results from additional model fits in which ethnicity and deprivation are added to the base model without geographic confounders, as well as further description/definition of the local authority code, and analyses that add each geographic confounder to the base model one at a time.
A second limitation includes the temporal misalignment of the confounders: age, sex, ethnicity, adjusted household deprivation, and disability are sourced from the 2011 Census. If one hypothesis is that the deprivation serves as a major confounder, it would be important to obtain more recent data as the temporal changes in Census data may affect results.
A third limitation includes source and linkage of the exposure data. The authors use London Air Quality Network estimates of annual average air pollution concentration in 2016 as their exposure, suggesting that it could serve as a reasonable proxy for long-term exposure since spatial variability in air pollution across London has been relatively stable in recent years. As acknowledged in their discussion, other variables such as occupation and transport may affect an individual’s exposure.
A fourth limitation of this study is a lack of discussion on representativeness. Although the authors access a large, representative sample of medical records of London residents, only 54.1% of confirmed positive cases were linked to the ONS-PHDA, and it can be helpful if the authors could confirm that the reduced sample remains representative.
In conclusion, this work presents an interesting analysis of the association between long-term exposure to four air pollutants and risk of dying from COVID-19 using a comprehensive individual-level dataset. With some additional exploration and discussion, this manuscript has the potential to offer scientists and policy-makers valuable insights.