Skip booklet index and go to page content

Data Sources and Methods: Air Health Indicator – Ozone and Fine Particulate Matter


To identify the cardiopulmonary mortality-air pollution association, a Bayesian hierarchical model was applied. First, the city-specific risk for each city (census division) was estimated. Annual city-specific risks of cardiopulmonary mortality due to ozone or PM2.5 were estimated by a generalized Poisson model for each census division. A generalized additive over-dispersed Poisson regression model was applied to the daily mortality counts. Second, the national risk for each year was estimated by pooling over the city-specific risks. Third, testing was done of the resulting data for annual national attributable risks (due to ozone or PM2.5) to determine whether a time trend could be detected.

The AHI is based on two temporal functions: annual air pollutant concentrations and annual mortality risks of that air pollutant at the national level. The annual air pollutant concentrations were obtained from CESI’s national Air Quality Indicator data, and the annual mortality risks were estimated in the second stage above.

The annual mortality risk estimates showed no time trend. It was therefore possible to assert that the annual mortality risk was constant over the periods analysed. The average of 15 annual national risk estimates for ozone (0.012 per 10 ppb for 1990 to 2004) and the average of 5 annual national risk estimates for PM2.5 (0.032 per 10 ppb for 2000 to 2004) were used. It was assumed that this constancy persisted and each of these averages were projected for the years 2005 to 2008, for which yearly risk estimates were not yet available due to the unavailability of mortality data for these years. The overall annual risk was derived from the product of annual air pollutant concentrations and the average of annual mortality risk estimates of ozone and PM2.5. Finally, to detect a time trend, the Sen’s test, a non-parametric linear time-trend test, was applied. There were no time trends found in ozone and PM2.5 risks.

A time-trend analysis was also performed on the AHI using the Sen’s test. A trend was detected at the 95% confidence interval for the ozone AHI but not for PM2.5.