Project Background:

While the impact of air pollution on mortality in the short term (days) and medium term (<10 years) is now well established, there are relatively few studies assessing the long-term (>10 years) impact of air pollution with even fewer assessing the very long term (25+ years).

Study Aims:

Objective 1: Development of statistical methodology: a general Bayesian profile-based multi-pollutant modelling framework to examine patterns of air pollution exposures and relate these patterns to socio-economic characteristics and health outcomes.

Objective 2: Substantive Applications: To investigate the benefit of using multi-pollutant
profiling methods to characterize patterns of air pollution exposures and relate these exposure
patterns to indicators of health and socio-economic status. Patterns of highly correlated exposures at different geographical levels will be characterized using the profiling methodologies. Input data used will be (i) exposure estimates of NO2, NOx, PM2.5, PM10, and Ozone in Greater London from the Environmental Research Group, King’s College London, (ii) air pollutant data including particulate metals concentrations in Greater London and the Thames Valley area from the European Union funded European Study of Cohorts and Air Pollution Effects (ESCAPE) and Transphorm projects.

New statistical models (Bayesian non-parametric profile regression) will be used to fit flexible regression models between exposure profiles and health outcomes such as all-cause and cardiovascular mortality and hospital admissions and birth outcomes (birthweight and pre-term birth). Models will be adjusted for lung cancer incidence (using ONS cancer registration data) as a proxy for smoking. Comparisons will be made with results using standard statistical regression techniques.

Health Data:

ONS mortality, ONS births, ONS cancer registrations, NHS Digital HES inpatients & maternity

Benefits to Public:

Results will provide further evidence on health effects of air pollution, including on metals exposures, which have been relatively little studied to date. The methodological component of the work will investigate whether looking at combinations of pollutants in this way gives better predictions of health risks than standard epidemiological techniques. Results are relevant to air pollution regulation to protect public health and will also provide information about environmental justice. The outputs of the project will be published in peer-reviewed journals during the course of the study and after completion, which is expected by 2016.

Georgious P, Richardon S, Best N. Bayesian non-parametric models for spatially indexed data of mixed type. J R Statist Soc Series B Dec 2014.