Spatial modelling of air pollution, deprivation and mortality in Scotland

Pannullo, Francesca Giuseppina (2017) Spatial modelling of air pollution, deprivation and mortality in Scotland. PhD thesis, University of Glasgow.

Full text available as:
[thumbnail of 2017PannulloPhD.pdf] PDF
Download (17MB)
Printed Thesis Information: https://eleanor.lib.gla.ac.uk/record=b3281513

Abstract

Air pollution is not only a major risk to the environment, but also a major environmental risk to the health of the population in developed and developing countries. The health impact of both short-term and long-term exposure to air pollution has been the focus of much research in the past few decades, which has investigated the relationship between specific air pollutants, such as carbon monoxide (CO), nitrogen dioxide (NO_2), particulate matter (PM_2.5 and PM_10), and sulphur dioxide (SO_2), to cardiovascular and respiratory diseases.

The health impact of short-term exposure is conducted through time series studies, whereas long-term exposure is investigated through cohort studies. Cohort studies are considered the gold-standard research design since inference is made at the individual level and can directly assess cause and effect. However, cohort studies are costly and require a long follow-up period meaning they take a long time to conduct.

To counteract these limitations, spatial ecological studies are used instead, which make use of routinely available disease data and air pollutant concentrations at a small areal level, such as census tracts or postcodes. This is to ensure the population under study is relatively homogeneous within the areal unit in terms of socio-demographic characteristics, and thus complements inference from a cohort study. These studies quantify the health impact of exposure to air pollution by relating geographical contrasts between air pollutant concentrations and disease risk across the chosen spatial resolution. The disease data are counts of the numbers of disease cases occurring in each areal unit, and Poisson log-linear models are used to assess the pollutant-health relationship.

Other covariate information, such as socio-economic deprivation, is also included to help explain the spatial pattern in disease risk. However, the residual disease risk after the covariate effects have been accounted for tends to contain spatial autocorrelation, which has to be modelled in order to make sound inferences. Residual spatial autocorrelation is typically modelled by a set of random effects that utilise a neighbourhood matrix in order to induce spatial autocorrelation into the model. There are a number of specifications to model this, but this thesis makes use of the Leroux specification due to its flexibility in being able to model both strong and weak spatial autocorrelation.

An important issue with using a spatial ecological study design is the estimation of spatially representative pollutant concentrations that are available in each areal unit. Studies can typically use measured data from fixed-location monitors that are spatially sparse and do not provide a pollutant concentration for each areal unit; or they make use of modelled concentrations available at a fine grid square resolution, which are known to contain biases and no measure of uncertainty. There have been numerous statistical approaches to combine both sets of information in order to estimate accurate and spatially representative concentrations. This thesis will develop previous methodology that utilises extra data sources in order to improve the prediction performance of the model for use in a Scottish context.

The overarching aim of this thesis is to investigate the cardio-respiratory health effects of long-term exposure to air pollution in West Central Scotland, UK. As the majority of air pollution in this region results from vehicle emissions, nitrogen dioxide (NO_2), a traffic-related gaseous pollutant, will be used to measure air pollution. Models investigating its health effect will incorporate predicted measures of NO_2 developed in this thesis. The sensitivity of the pollutant-health effect to the choice of NO_2 concentrations, indicator of deprivation, and choice of spatial model will be investigated. Changing these factors has been shown to modify estimated pollutant-health effects.\\

Findings in this thesis demonstrated that improvements in the accuracy of fine scale spatial prediction of NO_2 concentrations can be made by utilising extra sources of data in addition to the commonly-used monitoring stations. In addition, the estimated pollutant-health effect is not robust to the choice of the aforementioned factors and the choice of these factors can have a major impact on the resulting pollutant-health effects. This justified the combination of all statistical models into a single effect size, which estimated a small, but positive effect of NO_2 concentrations on cardio-respiratory ill health. However, the estimated NO_2-health relationship was not substantial, possibly due to the NO_2 concentrations in West Central Scotland being too low. Greater variation in the exposure would be needed to observe substantial health impacts.

Item Type: Thesis (PhD)
Qualification Level: Doctoral
Additional Information: The research presented in Chapter4 has been published in the Atmospheric Environment journal with the title `Improving spatial nitrogen dioxide prediction using diffusion tubes: A case study in West Central Scotland' (2015, volume 118, p227-235), and is co-authored with Dr Duncan Lee, Dr Eugene Waclawski, and Professor Alastair H Leyland. In addition, Chapter 5 has been published in the Spatial and Spatio-temporal Epidemiology journal with the title `How robust are the estimated effects of air pollution on health? Accounting for model uncertainty using Bayesian model averaging' (2016, volume 18, p53-62), and is also co-authored with Dr Duncan Lee, Dr Eugene Waclawski, and Professor Alastair H Leyland.
Keywords: statistics, spatial modelling, bayesian, bayesian model averaging, air pollution, health.
Subjects: Q Science > QA Mathematics
Colleges/Schools: College of Science and Engineering > School of Mathematics and Statistics > Statistics
Funder's Name: Medical Research Council
Supervisor's Name: Leyland, Professor Alastair H. and Lee, Dr. Duncan
Date of Award: 2017
Depositing User: Dr Francesca Pannulli
Unique ID: glathesis:2017-8415
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 22 Sep 2017 08:26
Last Modified: 29 Sep 2017 11:52
URI: https://theses.gla.ac.uk/id/eprint/8415

Actions (login required)

View Item View Item

Downloads

Downloads per month over past year