Health effects of air pollution : innovative approaches for spatio-temporal evaluations

Abstract

Air pollution is one of the major risk factors to human health, causing both short- and long-term effects and the global burden on mortality is estimated in more than 4 million deaths every year. Most of the evidence on the short-term effects is based on studies conducted in major cities, because data or estimates of air pollutants exposures in non-urban settings have been historically lacking. This is a limitation, because a large fraction of the population lives outside the cities, where the vulnerability profile is different from that of urban populations. In the last decade, several attempts were made to estimate daily concentrations of particulate matter (PM) with high spatial resolution over large geographical domains. However, applications in Italy and Sweden, and on other pollutants as nitrogen dioxide (NO₂) and ozone (O₃), are almost lacking, leaving a gap in the knowledge of their health effects outside cities.

This thesis has been designed to fill this gap, by providing daily estimates of multiple air pollutants at the national level, and exploring the spatial heterogeneity in their health effects. Italy represented a testing ground for the development of innovative mixed-effects regression models which combined PM measurements with satellite data, land-use parameters and meteorological fields, and produced daily estimates of PM₁₀ (PM with diameter smaller than 10 μm) for each squared kilometer of the country, and each day in 2006-2012 (Study I). More recently, machine learning methodologies have been tested in the U.S., therefore, we have updated estimates of PM₁₀ till 2015 and produced new estimates of PM₂.₅ (PM < 2.5 μm), using a random forest (RF) algorithm (Study II). We replicated the same approach in Sweden, to which we added models for NO₂ and O₃, and a few spatiotemporal predictors aimed at capturing sources of air pollutants’ variations missed in the previous studies (Study III).

We collected national data on hospital discharges for all Italian public and private hospitals during 2013-2015. We created municipality-specific time-series of daily counts of acute admissions for multiple cardiovascular (CVD) endpoints, which we related to daily mean PM10 and PM2.5 concentrations. We found evidence of adverse effects of PM on total CVD admissions and on specific outcomes such as heart failure and atrial fibrillation. Also, we estimated highest effects at the lowest PM concentrations, also in non-urban municipalities (Study IV). Similarly, we collected daily mortality counts at small area level in the Stockholm county, that we analyzed in relation to daily mean exposure to PM₁₀, PM₂.₅, NO₂ and O₃. We found evidence of an association between daily O₃ and non-accidental mortality in the year-round analysis, and significant associations with PM and O₃ in the warm (April-September) period only. Effects were slightly higher in more densely inhabited areas, but we found associations also in non-urban areas outside the Stockholm city (Study V).

In conclusion, we developed novel spatiotemporal models to estimate air pollutant concentrations at fine spatial and temporal resolution in Italy and Sweden. These allowed us to document adverse short-term effects on mortality and morbidity at very low concentrations and in areas (and among populations) previously neglected by epidemiological investigations.