Air pollution health impacts: the knowns and unknowns for reliable global burden calculations

This editorial refers to ‘Loss of life expectancy from air pollution compared to other risk factors: a worldwide perspective’, by J. Lelieveld et al., pp. 1910–1917.

Background

Recent years have seen increasing interest focused on the contribution of air pollution to global health, with a growing emphasis on cardiovascular disease (CVD) risk. This attention from the research community, media, healthcare providers, and the public has paralleled the occurrence of severe air pollution episodes in urban centres of rapidly developing economies, and those resulting from landscape fires on multiple continents. The American Heart Association¹ and the European Society of Cardiology² have published position papers concluding that fine particulate matter (PM_2.5) air pollution exposure is a modifiable risk factor for cardiovascular morbidity and mortality and the World Heart Federation highlighted the impact of air pollution on CVD for World Heart Day in 2018.³ The World Health Organization (WHO) included air pollution among its top 10 threats to global health in 2019.⁴

Lelieveld et al.⁵ conducted a disease burden assessment to calculate excess mortality and loss of life expectancy from outdoor air pollution (both PM_2.5 and Ozone) globally and by country. In their work, they apply a chemical transport model to simulate atmospheric levels of long-term PM_2.5 and Ozone concentrations and apply previously published concentration–response functions linking air pollution levels to risks of non-accidental mortality. They estimate 8.8 million excess deaths (95% confidence interval (CI) 7.11–10.41) and 2.9 years of loss of life expectancy (95% CI 2.3–3.5) were attributable to exposure to outdoor air pollution in 2015. These estimates are more than double previous air pollution burden estimates. For example, the Global Burden of Disease (GBD) Study⁶ estimated that in 2017, 2.9 million deaths (95% CI 2.5–3.3) were attributable to outdoor air pollution, while the WHO⁷ estimated 4.3 million deaths (95% CI 3.6–5.0) from PM_2.5 air pollution in 2016.

The above estimates suggest substantial disease burden from outdoor air pollution, but there are big differences in their magnitude. A large proportion of the estimated burden from all of these analyses are in locations of high pollution (mostly in low- and middle-income countries), where observational studies are uncommon and relevant data largely missing. Burden estimates are therefore uncertain and sensitive to methodologic assumptions regarding extrapolation of concentration–response relationships from observational studies conducted mostly in high-income countries with low concentrations. Lelieveld et al.⁵ relied on the Global Exposure Mortality Model (GEMM),⁸ which derives cause-specific mortality estimates from 41 observational cohort studies from 16 countries. Relying on the GEMM assumes that the existing air pollution and mortality cohort literature (based on studies largely from North America and Europe) captures the exposure–response relationships for different causes of mortality across the full spectrum of PM_2.5 concentrations and different populations without systematic biases. In particular, the effects estimated for the GEMM at higher PM_2.5 concentrations are highly sensitive to a cohort study of Chinese males with very high smoking prevalence. The GBD approach uses additional literature from studies of household air pollution from the use of polluting cooking fuels and second-hand smoking to inform the upper portion of the exposure–response curve. Given that over 54% of global PM_2.5 exposures are above 35 µg/m³ (the WHO Level 1 Interim Target)⁹ and outside of the range of many of the included cohort studies, the application of the GEMM to project global disease burden can be misleading and should be treated with considerable caution.

Further, the GEMM is based on analyses of non-accidental mortality, and the application of this model to global disease burden estimates assumes similar distributions of different non-communicable disease rates in high-income countries (from which the GEMM is derived) as in low and middle-income countries. This is problematic since the relative frequencies of the various causes of deaths differ markedly between regions of the world.¹⁰ Lelieveld et al.⁵ also did not calculate attributable loss of life expectancy based on standard methods using survival curves as in other PM_2.5 burden estimates,¹¹ but instead multiply the attributable years of life lost per capita (to air pollution) by the WHO maximum life expectancy of 91.9 years. This formulation is not derived and in the context of a comparative assessment (across countries and other risk factors) is problematic and likely provides upwardly skewed estimates of attributable loss of life expectancy.

The knowns and unknowns

The large population health impact of air pollution results from a high prevalence of exposure combined with small relative risks. Relative risks used to estimate the attributable burden of disease from air pollution are derived from meta-regressions of multiple cohort studies of mortality or disease incidence, mainly conducted in North America and Europe. These observational studies of the impacts of long-term exposure are supported by daily time series studies of the impacts of short-term exposure variation conducted throughout the world,¹² and by an extensive literature of experimental studies (animal toxicology and controlled human exposures).^1,2 Based on experimental evidence there is now a physiologic basis for the adverse health effects of exposure to outdoor air pollution, with a dominant pathway by which pulmonary inflammation ‘spills over’ to systemic inflammation which in turn increases the risk of CVDs.

However, detecting and estimating the small relative risks linking long-term air pollution exposure to chronic disease is difficult. Observational epidemiologic methods are subject to a range of confounders whose effects can be several times larger than the potential effects of the air pollution exposure being studied. Recently, Pope et al. (2020) reviewed over 25 years of cohort studies examining PM_2.5 and mortality and concerns regarding publication bias, confounding, and exposure measurement error. Funnel plots, while not conclusive, did not identify evidence of publication or selection bias. Based on 75 studies, they calculated a linear hazard ratio (HR) for all-cause mortality of 1.09 (95% CI 1.07–1.11) per 10 µg/m³ increase in PM_2.5. Effect estimates were similar in analyses that adjusted for individual-level smoking [pooled HR 1.07 (95% CI 1.06–1.10)], climate variables [HR 1.14 (95% CI 1.05–1.24)], and contextual variables [HR 1.09 (95% CI 1.07–1.17)]. In their analysis, Pope et al. concluded that studies with individual-level confounding data did not observe smaller estimated HRs for all-cause mortality, suggesting that a lack of control for individual-level covariates was not driving the associations seen with PM_2.5 in meta-analyses. Still, most cohort studies include rather minimal individual-level covariate adjustment and some degree of residual confounding is likely. Moreover, since air pollution exposure is typically estimated at an ecological level (i.e. modelled or measured concentrations applied to residential locations of study participants) air pollution exposure measurement error and confounding by spatially varying contextual variables is likely a greater contributor to bias.

Improving air pollution burden calculations

While the translation of air pollution results to the policy and public is a top priority (and disease burden estimations are a powerful approach), a primary focus of the research community should remain on strengthening the evidence regarding air pollution impacts on human health.

Several key questions remain that are especially important to global air pollution burden calculations. Most importantly, there is a need for high quality studies in low- and middle-income countries and especially studies at high PM_2.5 concentrations. In addition, identification of susceptible populations, competing risk factors, and potential differences in particle toxicity due to different composition and source contributions are priorities. Further, additional research on specific and high prevalence diseases that may be causally linked to air pollution can help explain the gap between estimates derived from all-cause mortality and those based on specific diseases, as in the GBD. For example, a recent extensive case–control analysis of inpatient MEDICARE hospitalization claims data identified short-term air pollution impacts on a far larger set of diseases than currently considered in most cohort analyses, including high prevalence diseases such as chronic kidney disease and neurological disorders.¹³ This trend of a greater number of diseases being linked to air pollution parallels the diverse diseases associated with tobacco smoking.¹⁴ Greater emphasis in primary research should also focus on measures of disease incidence. When mortality is used as a metric, major diseases associated with air pollution (e.g. CVD) have very different fatality rates across countries,¹⁰ which is problematic for assessing the contribution of air pollution exposures to the development (and avoidance) of disease. Causal inference modelling can also strengthen the weight of evidence for the impacts of long-term air pollution on chronic disease.¹⁵

Comparative risk assessments and global burden calculations are useful for highlighting the importance of air pollution but equally useful for illuminating limitations in the underlying evidence as a means to ultimately improve understanding. While the weight of evidence suggests that air pollution increases the risk of CVD, the magnitude of this impact is uncertain, especially in the context of global burden estimation. In the end, reducing much of the uncertainty and improving the overall understanding of the health impacts of air pollution globally will require additional high quality epidemiologic studies involving diverse populations exposed to high air pollution concentrations and a better understanding of the biochemical pathways of disease causation.

Conflict of interest: The authors report no conflicts of interest.

The opinions expressed in this article are not necessarily those of the Editors of Cardiovascular Research or of the European Society of Cardiology.

Funding

M.B. holds a grant (#136893) from the Canadian Institutes for Health Research. P.H. holds a grant with the Office of the Director, National Institutes of Health (award DP5OD019850). S.Y. is an investigator on both of these grants. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Canadian Institutes for Health Research or National Institutes of Health.

References