imprint:
[Erscheinungsort nicht ermittelbar]: Utrecht University, 2021
Language:
English
Origination:
University thesis:
Dissertation, Utrecht University, 2021
Footnote:
Description:
Air pollution concentrations in North America and Europe have decreased substantially over the past decades. Despite that, recent studies suggested that the well-documented adverse health effects of air pollution may persist at levels below the current air quality guidelines and standards. This has raised questions about the appropriateness of the existing standards and guideline values for regulated pollutants including particulate matter with an aerodynamic diameter ≤ 2.5 and ≤ 10 μm (PM2.5 and PM10), nitrogen dioxide (NO2) and ozone (O3). The composition of PM2.5 varies in time and space, which may result in differences in toxicity and risk to health of PM2.5. Mixed findings have been reported for associations between specific PM2.5 components and health endpoints. Recent developments in air pollution epidemiology require air pollution exposure estimates covering large geographical areas at sufficiently fine spatial resolution. Land use regression (LUR) models have been widely used in air pollution exposure assessment and are often developed with linear regression algorithms. More flexible modeling algorithms, such as regularization algorithms and machine-learning algorithms, have been applied recently. Few studies have compared exposure models developed with different algorithms. No study has compared the health effects related to air pollution concentrations estimated with these different exposure models. The main objective of the research described in this thesis is to develop Europe-wide LUR models for key pollutants PM2.5, NO2, O3, black carbon (BC) and PM2.5 elemental composition, as well as to evaluate the mortality effects of PM10, PM2.5 and PM2.5 composition. We compared exposure models developed with different statistical algorithms, and the health effects assessed by applying these different exposure models. This research has benefited from the unique air pollution dataset collected by a purpose-designed monitoring campaign of the 'European Study of Cohorts for Air Pollution Effects' (ESCAPE), and the large study populations in the 'Effects of Low-level Air Pollution: A Study in Europe' (ELAPSE) pooled cohort with detailed individual-level covariates. A systematic review of the literature of effects of long-term exposure to PM2.5 and PM10 on mortality documented significantly positive associations with all-cause as well as cause-specific mortality. Associations remained at concentrations below the current WHO air quality guideline level of 10 µg/m3 for PM2.5, predominantly in North-American studies. Effect estimates were heterogenous, probably related to differences in particle composition, exposure level and methodological differences between studies. We developed Europe-wide LUR models for assessing long-term exposure to PM2.5, NO2, O3, black carbon (BC) and PM2.5 elemental composition. We found only limited differences in model performance using supervised linear regression and a range of algorithms including machine-learning methods. Models were applied in the ELAPSE project to assign exposure at the individual level. PM2.5 was associated with increased mortality also at concentrations below current guideline values. Long-term exposures to especially vanadium in PM2.5 was associated with increased mortality risk, with associations observed for both random forest- and supervised linear regression-modeled exposures. For the other elemental components studied, associations were generally weaker when exposure was assessed with random forest compared to supervised linear regression in two-pollutant models.