Backgrounders
	Health Canada announces project to explore altruistic (not for payment) donation of sperm and eggs in Canada

Information

April 2005

Estimated Number of Excess Deaths in Canada Due To Air Pollution

Stan Judek, Barry Jessiman, Dave Stieb,
Air Health Effects Division, Health Canada, and
Robert Vet, Meteorological Service of Canada, Environment Canada

August 30, 2004

Summary and Results

The Air Health Effects Division (AHED) was mandated to estimate the annual number of excess deaths due to current air pollution levels in Canada. The results presented here are based on non-accidental mortality counts and National Air Pollution Surveillance (NAPS) data for the years 1998 to 2000, and pollutant-mortality concentration response functions (CRFs) from epidemiological studies. These estimates can be viewed as the number of deaths which could be prevented each year if air pollution from human sources within North America were eliminated.

The overall estimate is determined as the sum of the estimated number of excess deaths associated with both short- and long-term exposure to air pollution. The statistical methodology employed in the literature allows this addition,1 although some authors believe there may be overlap between these two outcomes, meaning they are not strictly additive.2 The geographic unit of analysis was the Census Division (CD). Assigning annual pollutant concentration levels to a CD with confidence for all four gases and fine particulate matter (PM2.5) resulted in estimating the number of excess deaths for a population of 8.9 million Canadians, distributed among 8 CDs, from the total population of 28.7 million Canadians in the 1996 Census.

The annual excess number of deaths associated with short-term exposure was estimated to be (to the nearest hundreds) 1,800 + 700. The annual excess number of deaths associated with long-term exposure was estimated to be 4,200 + 2,000, although it might take five years or more after reducing air pollution levels to fully realize these preventable deaths. These ranges reflect the corresponding uncertainties in the CRFs employed, with each range roughly indicating a 95% confidence interval. Adding these values results in a total excess deaths estimate of 5,900 + 2,100. This is a conservative estimate, as approximately only one third of the Canadian population, mainly residents of large, urban areas, is included in this base case analysis.

A document presenting the results of a sensitivity analysis for the base case will follow.

Rationale

An original estimate of 5,000 premature deaths for 11 major Canadian cities was generated in 1999 by Health Canada. This estimate included only a short-term exposure component and was scaled up to an estimate of 16,000 premature deaths for the entire Canadian population. This extrapolation is questionable because exposure to air pollution is generally lower in areas other than the original 11 cities upon which the 5,000 premature deaths estimate was based.

This previous estimate utilized concentration response functions employing Generalized Additive Models (GAMs) in their statistical methodology, the preferred approach to time-series analysis at that time. Since then, statistical and numerical flaws have been discovered with the implementation of this methodology by Health Canada, University of Ottawa and Johns Hopkins University researchers.3 These flaws have the tendency to over-estimate the magnitude of the regression parameters, while under-estimating the corresponding standard errors.4

It was therefore appropriate to update this estimate utilizing short-term exposure concentration response functions not employing the GAM methodology, and including a long-term exposure component.

INPUTS

Geographic Units

The 1996 Census identified 282 Census Divisions, each with sets of latitude/longitude coordinates defining their boundaries. All NAPS sites that were found inside the boundaries were mapped to the CD; their data contributed to the determination of annual concentration levels for the CD. Data from the 1996 census rather than the more recent 2001 census were employed because of the availability of the boundary data at the time of analysis, and presumably Census Divisions did not notably change their geographic definition in the years 1998 to 2000 until the following census.

Mortality Counts

Non-accidental counts of deaths (ICD-9 code <800 or excluding accidents and suicides) by Census Subdivision (CSD) and year were accumulated to determine the counts by CD for each of the years 1998, 1999 and 2000.

Pollutants

Daily 24-hour carbon monoxide (CO), nitrogen dioxide (NO2), sulphur dioxide (SO2), daily 1-hour ozone (O3), and every six-day 24-hour Dichotomous Sampler fine particulate matter (PM2.5) data from available mapped sites and days in 1998 to 2000 were used in estimating annual concentrations for a CD. For each day in a given year, the concentration levels available at the mapped sites were averaged; a series of day to day values over the course of a year was thus determined for a CD. For simplicity of whole number division, 360 (as compared with 365) days of data in a year were expected for the gases and 60 days of data in a year were expected for PM2.5. This translated to 90 and 15 days of data, respectively, expected per quarter. An annual concentration level was determined for a CD when there was at least 50% of the expected number of days of data in each quarter; this is similar to the criterion employed in one of the epidemiological studies.5 In this case, the average of the 4 quarterly means became the annual concentration assigned to the CD. These were the overall annual pollutant concentrations, not adjusting for background levels (see next section).

Day to day and annual population weighted distances to the closest pollutant site providing data were determined. For all 8 CDs, distances to the closest site were less than 16 km for all pollutants and all years, ranging from 14.3 km for SO2 to 15.8 km for PM2.5.

Background Concentrations

It was assumed that each of the five pollutants has a "background" concentration defined as the concentration that would occur if all North American anthropogenic sources of the pollutants and their precursors were turned off. In theory, then, the background concentrations would be caused entirely by natural emissions of the pollutants and their precursors and by the inflow of the pollutants and their precursors from outside of North America (e.g., from the oceans and other continents).

The estimated "background" concentrations of the five pollutants were difficult to establish. In the end the method involved a combination of qualitative and quantitative approaches based on concentration measurements at rural and remote monitoring sites. This resulted in a set of monthly-average background concentrations for CO, O3 and SO2 (all of which have strong seasonal cycles in their ambient concentrations) and annual-average background concentrations for NO2 and PM2.5 (for which there was too little data to establish the existence of seasonal cycles). It was noted that regional differences likely exist in the background concentrations, but, for simplicity, these were ignored. Only very low concentrations at rural and remote sites were used to establish the background concentrations based on one of the two following methods:

1) the data from rural and remote measurement sites were separated into sectors of different air mass origin and the "background" concentrations were selected as the monthly- or annual-average concentrations associated with the sectors containing no major anthropogenic sources, or

2) many years of rural and remote measurement data were plotted in a time series that allowed a qualitative selection of the lowest values that were, in turn, considered to be most representative of background air masses.

Anthropogenic Concentrations

To estimate the anthropogenic pollutant concentrations (i.e., the measured urban concentrations minus the background concentrations), we calculated the day-to-day series of urban concentrations in a CD (as in the previous section) and subtracted the associated monthly-average background concentrations of CO, O3 and SO2 and annual-average background concentrations of NO2 and PM2.5, converting negative values to zero. This resulted in a day to day series of anthropogenic concentrations of each pollutant. We then calculated the annual anthropogenic pollutant concentrations in the same way as described above using the same data completeness criteria described in the previous section.

There was one case (CD=4806 - Calgary) where the Dichotomous Sampler PM2.5 concentration levels could not be determined for one of the three years. To deal with this, daily data from an alternate continuous PM2.5 measurement instrument called the Tapered Element Oscillating Microbalance (TEOM) were used to determine the annual concentrations, using the same criterion and algorithm as described for the gases. No analysis was carried out to estimate background TEOM levels. Instead the ratio of the Dichotomous Sampler concentrations (both overall and anthropogenic) to the overall TEOM annual concentrations was calculated for the other two years and applied to the TEOM annual concentration level for the missing year. This allowed us to estimate the annual Dichotomous Sampler concentration levels (overall and anthropogenic, respectively) for the missing year.

Concentration Response Functions

A 12 Canadian city 1981-1999 time-series study6 involving daily monitoring of the four gases was the source of the concentration-response functions (CRFs) employed to estimate the excess number of deaths associated with short-term exposure to air pollution. This study was selected because it is based on all available pollutant data ever collected in Canada (from 1981-1999). The author provided results from additional multi-pollutant models to AHED; the 4-gas model was selected based on the overall t-value among the candidate models. The statistical model was a Poisson regression model. Although this multi-pollutant model excluded particulate matter, it was selected as the model which best reflected the impact of the overall air pollution mix. Because of multi-collinearity among pollutants, this model should nonetheless still reflect impacts of particulate matter. In any case, the effects of particulate matter in this study were reduced substantially when modelled together with NO2, the effect of which predominated in this analysis.

A cohort study7 involving 300,000 subjects in 50 U.S. metropolitan areas over a 7 year period provided the CRF for PM2.5, which was employed to estimate the excess number of deaths associated with long-term exposure to air pollution. The statistical model accounted for age, sex, race, cigarette smoking, exposure to passive cigarette smoke, body mass index, drinks per day of alcohol, education and occupational exposure. The statistical model was a Cox proportional-hazards model, so the relative risk was treated as one resulting from a Poisson regression. These results were employed rather than the more recent results from this study based on extended follow-up of the cohort5, because the latter may be affected by increasing exposure misclassification as members of the cohort move over time. However the results from the extended follow-up were employed in a sensitivity analysis.

Fundamental Input to the CRF

The basic input multiplier to a pollutant-specific CRF is

annual non-accidental mortality count x estimated annual anthropogenic concentration

at the CD level, averaged over 1998-2000. These averages, in turn, were summed over the 8 CDs as final input multipliers.

Summary of Inputs for Base Case

Appendix A displays the population, mortality counts, overall and estimated anthropogenic pollutant levels by Census Division, as well as background concentrations. Appendix B displays the concentration response functions.

Results

Results are shown by city in Appendix C. Expressed as a percent of deaths from all causes, results vary slightly among cities, but 95 percent confidence intervals overlap in all instances.

Sensitivity Analyses

The effect on excess death estimates of alternative analytical approaches was examined. These included alternative methods of mapping pollution data to the population, zero background concentrations and alternative sources of concentration response functions.5,8,9 These had varying impacts on the excess death estimates, which will be detailed in a later document.

Communication Issues

In communicating the results of this analysis to a non-scientific audience, there are a number of issues requiring emphasis. First, it must be clear that the analysis only examines mortality impacts in selected cities, and that impacts both in other cities, and on morbidity (eg. hospital admissions), would be substantial. Second, it should be communicated that this analysis is not simply a matter of counting death certificates which mention air pollution, but that it involves applying the results of complex statistical models. Third, to provide perspective on the relative impact of air pollution, it is recommended that results be expressed both as counts and percentages of total deaths. On a related point, results should be compared to other estimates of air pollution effects, to historical estimates of air pollution burden of illness, and to effects of other risk factors. As an example, as part of the World Health Organization Global Burden of Disease project, Ezzati et al.10, recently estimated that 1.4% of all deaths worldwide could be attributed to urban outdoor air pollution, assuming a background concentration of 7.5 ug/m3 of PM2.5, somewhat higher than in the present analysis. A similar proportional effect was observed for North America. Health Canada estimated that approximately 22% of deaths in Canada could be attributed to active and passive smoking11. From the historical perspective, it's been estimated that 3,000-4,000 deaths occurred during the London smog episode of December 1952 (the population of London at the time, 8.6 million, was similar to the size of the population considered here). This occurred in the span of only a few weeks (as opposed to 5,900 deaths over a full year in the current analysis). It's been suggested that as many as an additional 9,000 excess deaths may have occurred if the subsequent 2 month period after the episode was also considered12. Finally, 95% confidence intervals should be reported as a reflection of uncertainty, with reference to "margin of error" statements which typically accompany opinion poll results, as a means of explaining the concept.

References

1. Burnett, R.T.; Dewanji, A.; Dominici, F.; Goldberg, M.S.; Cohen, A.; Krewski, D. On the Relationship between Time-Series Studies, Dynamic Population Studies, and Estimating Loss of Life Due to Short-Term Exposure to Environmental Risks. Environ Health Perspect. 2003; 111(9):1170-4

2. Kunzli, N.; Medina, S.; Kaiser, R.; Quenel, P.; Horak, F. Jr.; Studnicka, M. Assessment of deaths attributable to air pollution: should we use risk estimates based on time series or on cohort studies? Am J Epidemiol. 2001; 153(11):1050-5

3. Dominici, F. Recent NMMAPS Updates Resulting from Discoveries about Generalized Additive Model Software. (accessed June 8, 2004)

4. Ramsay, T.O.; Burnett, R.T.; Krewski, D. The Effect of Concurvity in Generalized Additive Models Linking Mortality to Ambient Particulate Matter. Epidemiology. 2003; 14(1):18-23

5. Pope, C.A. III; Burnett, R.T.; Thun, M.J.; Calle, E.E.; Krewski, D.; Ito, K.; Thurston, G.D. Lung Cancer, Cardiopulmonary Mortality and Long-term Exposure to Fine Particulate Air Pollution. JAMA. 2002; 287(9):1132-41

6. Burnett, R.T.; Stieb, D.M.; Brook, J.R.; Cakmak, S.; Dales, R.E.; Raizenne, M.E. Associations between short-term changes in nitrogen dioxide and mortality in Canadian cities. Accepted for publication in Archives of Environmental Health.

7. Krewski, D.; Burnett, R.T.; Goldberg, M.S.; Hoover, K.; Siemiatycki, J.; Abrahamowicz, M.; White, W.H.; et al. Reanalysis of the Harvard Six Cities Study and the American Cancer Society Study of Particulate Air Pollution and Mortality. A Special Report of the Health Effects Institute's Particle Epidemiology Reanalysis Project. Final version. July 2000.

8. Stieb, D.M.; Judek, S.; Burnett, R.T. Meta-Analysis of Time-Series Studies of Air Pollution and Mortality: Update in Relation to the Use of Generalized Additive Models. J Air Waste Manag Assoc. 2003; 53(3):258-61

9. Stieb, D.M.; Burnett, R.T., Smith-Doiron, M.; Chen, Y.; Goldberg, M.S.; Cakmak, S.; Dales, R.E.; Dann, T.; Brook, J.R. Air Pollution, Mortality and Cardiorespiratory Hospital Admissions in 11 Canadian Cities, 1998-2000: Application To A Reformulated Air Quality Index. DRAFT (June 13, 2003)

10. Ezzati, M.; Lopez, A.D.; Rodgers, A.; Vander Hoorn, S.; Murray, C.J.; Comparative Risk Assessment Collaborating Group. Selected major risk factors and global and regional burden of disease. Lancet. 2002; 360(9343):1347-60.

11. Makomaski Illing, E.M.; Kaiserman, M.J.. Mortality attributable to tobacco use in Canada and its regions, 1998. Can J Public Health. 2004; 95(1):38-44.

12. Bell, M.L.; Davis, D.L.; Fletcher, T.. A retrospective assessment of mortality from the London smog episode of 1952: the role of influenza and pollution.
Environ Health Perspect. 2004; 112(1):6-8.

* pollutant concentrations are expressed as measured at ambient monitoring stations and (in parentheses) with background subtracted.

*Totals may differ from sum of component values due to rounding (to nearest 10 for city values and nearest 100 for totals).