Content uploaded by Renaud Vincent
Author content
All content in this area was uploaded by Renaud Vincent on May 27, 2016
Content may be subject to copyright.
Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=iiht20
Download by: [Health Canada / Public Health Agency of Canada] Date: 27 May 2016, At: 10:04
Inhalation Toxicology
International Forum for Respiratory Research
ISSN: 0895-8378 (Print) 1091-7691 (Online) Journal homepage: http://www.tandfonline.com/loi/iiht20
ASSOCIATION BETWEEN PARTICULATE- AND GAS-
PHASE COMPONENTS OF URBAN AIR POLLUTION
AND DAILY MORTALITY IN EIGHT CANADIAN CITIES
R. T. Burnett, J. Brook, T. Dann, C. Delocla, O. Philips, S. Cakmak, R. Vincent,
M. S. Goldberg & D. Krewski
To cite this article: R. T. Burnett, J. Brook, T. Dann, C. Delocla, O. Philips, S. Cakmak, R. Vincent,
M. S. Goldberg & D. Krewski (2000) ASSOCIATION BETWEEN PARTICULATE- AND GAS-PHASE
COMPONENTS OF URBAN AIR POLLUTION AND DAILY MORTALITY IN EIGHT CANADIAN CITIES,
Inhalation Toxicology, 12:sup4, 15-39, DOI: 10.1080/08958370050164851
To link to this article: http://dx.doi.org/10.1080/08958370050164851
Published online: 01 Sep 2001.
Submit your article to this journal
Article views: 261
View related articles
Inhalation Toxicology,12(Supplement 4):15–39, 2000
Copyright © 2000 Taylor & Francis
0895-8378/00 $12.00 + .00
ASSOCIATION BETWEEN PARTICULATE- AND GAS-PHASE
COMPONENTS OF URBAN AIR POLLUTION AND DAILY
MORTALITY IN EIGHT CANADIA N CITIES
R. T. Burnett
Environmental Health Directorate, Health Cana da, Ottawa, and
Department of Epidemiology and Community Medicine, University of
Ottawa, Ottawa, Canada
J. Brook
Atmospheric Environment, Environment Canada, Downsview, Canada
T. Dann
Environmental Protection, Environment Canada, Ottawa, Canada
C. Delocl a, O. Philips
Statistics Canada, Ottawa, Canada
S. Cakmak, R. Vincent
Environmental Health Directorate, Health Cana da, Ottawa, Canada
M. S. Goldberg
Epidemiology and Biostatistics Unit, Re search Centre on Human Health,
INRS-Institut Armand-Frappier, University of Quebec, Laval, Quebec, and
Department of Epidemiology and Biostatistics, McGill Univers ity,
Montreal, Q uebec, Canada
D. Krewski
Department of Epidemiology and Community Medicine, University of
Ottawa, Ottawa, Canada
Although some consensus has emerged among the scientic and r egulatory communities
that the urban ambient atmospheric m ix of combustion related pollutants is a determinan t
of population health, the relative toxicity of the chemical and physical components of
this complex mixture remains unclear. Daily mortality rates and concurrent data on size-
fractionated particulate mass a nd g aseous pollut ants were obtained in eight of Can ada’s
Received 4 February 2000; se nt for revision 10 March 2000; accepted 6 April 2000.
Address correspondence to Dr. Richard Thomas Burnett, 200 EnvironmentalHealth Center, Tunney’s
Pasture, Ottawa, Canada K1A 0L2. E-mail: rick burnett@hc-sc.gc.ca
15
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
16 R. T. BURNETT ET AL.
largest cities from 1986 to 1996 inclusive in o rder to examine the relative toxicity of the
components of the mixture of ambient air pollutants to which Canadians are exposed.
Positive and statistically sig nicant associations were observed between daily variations
in bot h gas- and particulate-phase pollution and daily uctuations in mortality rates. The
association between air pollution and mortality could not be explained by temporalvariation
in either mortality rates or weather factors. Fine particulate mass (less than 2.5 µm in average
aerometric diameter) wa s a stronger predictor of mortality than coarse mass ( between 2.5
and 10 µm). Size-fractionated particulate mass explained 28% of the total health effect of
the mixture, with the remaining effects accounted for by the gases. Forty-seven elemental
concentrations were obtained for the ne and coarse frac tion using nondestructive x-ray
uorescence techniques. Sulfate concentrations were obtained by ion chromatography.
Sulfate ion, iron, nickel, and zinc from the ne frac tion were most strongly associated
with mortality. The total effect of the se four components was g reater than that for ne
mass alone, suggesting that the characteristics of the complex chemical mixture in the ne
fraction may be a b etter predictor of mortality than mass alone. However, the variation in the
effects of the constituents of the ne fraction between cities was greater t han the variation
in the mass effect, implying that there are additional toxic components of ne particulat e
matter not examined in this study whose concentrationsand effects vary between locations.
One of these components, c arbon, represents half the mass of ne particulate matter. We
recommend that measurementsof elementaland organic carbon be undertaken in Canadian
urban environments to examine their potential effects on human health.
Histo rically, extreme air pollution events, such as those experienced in Lon-
don in the 1950s and 1960s, clearly demonstrated the potential of ambient
air pollution to exacerbate cardiorespiratory disease, as reected in premature
mortality and increased hospital admissions. In the inte rvening years, consider-
able effort has been made to reduce atmospheric pollution from the combus-
tion of fossil fuels. Several countries, including Canada and t he United States,
have established stringent new guidelines and standards for air pollutants such
as sulfur dioxide, nitrogen dioxide, carbon monoxide, ozone, and particulate
matter. At present, the Canadian National Ambient Air Quality Objectives for
these pollutants are rarely violated.
Over the last decade, a series of studies has been published linking daily
variations in either deaths or admissions to hospital for cardiorespiratory dis-
eases and daily uc tuations in a number of ambient air pollutants (U.S . EPA,
1996). Although the majority of studies have focused on U.S. and European
locations with much higher air po llution levels than normally experienced in
Canada, some of the mo st convincing evidence linking air pollution to health
has been obtained from data collected in Canada.
Following Bates and Sitzo’s (1983, 1987) initial work, summertime concen-
trations of both ozone and particulate matter have been linked to respiratory
hospitalizations in southern Ontario (Burnet t et al., 1994 ), Toronto, Ontario
(Burnett et al., 1997a; Thurston et al., 1994), and in 16 of Canada’s largest cities
(Burnett et al., 1997b) . Summertime o zone levels have also been associated
wit h visits to the emergency department in t he Saint John Regiona l Hospital
for patients presenting with asthm a (Stieb et al., 1996) and to emergency-
department visits for respirato ry diseases in Montreal (Delno et al., 1996).
Elevated ambient levels of carbon monoxide have been linked to hospit aliza-
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
URBAN POLLUTION MIXTURE AND MORTALITY 17
tions for respiratory (Burne tt et al., 1997b) and cardiac (Burnett et al., 1997c)
diseases in several Canadian cities. Goldberg and collaborators (2000), in a
study to identify susceptible populations, found that increased levels of daily
particles in Montreal were associated with daily increases in nonaccidental
mortality among persons who had cancer, acute lower respiratory diseases,
any form of cardiovascular disease, chronic coronary artery diseases, or con-
gest ive heart fa ilure . Particulate sulfates were related to respiratory ho spitaliza-
tions in southern Ontario (Burnett et al., 1994, 1995), and the coefcient of
haze was linked to increases in respiratory hospitalizations in 16 cities span-
ning the breadth of the country (Burnett et al., 1997b). Particulate matter
and carbon monoxide were also asso ciated with daily mortality in Toronto,
Canada, over the 15-year period 1980 to 1994 (Burnett et al., 1998a).
Gas-phase ambient air pollution was linked to increases in daily mortal-
ity rates in 11 Canadian cities (Burnett et al., 1998b). However, particulate
mass measurements were not considered in the analysis. Daily variations in
particulate mass have been linked to daily variations in nonaccidental mortal-
ity rates in a number of locations worldwide (U.S. EPA, 1996). The purpose
of this investiga tion is, for the rst time, to examine the association between
constituents (mass, elements, and ions) of size-fractionated particulate matter
and daily mortality rates, controlling for weather factors, temporal trends, and
gaseous copollutants in Canadian urban environments.
METHODS
Environmental Data
Fine (PM
2.5
), coarse (PM
10-2.5
), and thoracic (PM
10
) measurements were
obtained from dichotomous samplers with Teon lters operating on a 6-day
schedule in 8 Canadian cities (Montreal, Ottawa, Toronto, Windsor, Winnipeg,
Calgary, Edmonton, and Vancouver) from 1986 to 1996 inclusive. Locations of
the cities are displayed in Figure 1. This d atabase had been described in detail
elsewhere (Brook et al., 1997). Forty-seven elemental concentrations were ob-
tained for the ne and coarse fraction using no ndestructive x-ray uorescence
techniques. Sulfate concentrations were obtained by ion chromatography.
For each sample, blanks were analyzed in order to identify a limit of
chemical detection. If a reading was below the limit of detection for that
sample but was positive, we included that data po int in the analysis. If a
zero reading was recorded, then a value equal to one-half the sample-specic
detection limit was used in the analysis.
Each city had one dichotomous sampler operating at any given po int in
time, except for Montre al and Vancouve r, which had two samplers. For these
la tter two cities, we averaged the available daily data between the two stations
to form a single time series for analysis.
Daily average concentrations of nitrog en dioxide (NO
2
), carbon monoxide
(CO), sulfur dioxide (SO
2
), and the coefcient of haze (COH), as well as
the daily 1-h m aximum concentration o f ozone (O
3
), were obtained for each
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
18 R. T. BURNETT ET AL.
FIGURE 1. Location of stud y communities.
monitoring station located in a specied geographic area for each city. Data
from a monitoring station were included in the study if at least two-thirds
of the days of observation had measured concentrations. We selected this
inclusion criterion in order to minimize any potential co nfounding effects on
the association between air pollution concentrations and mortality due to
differential temporal trend s in pollution levels between monitoring stations
wit hin a city. Details on the number and location of monitoring stations in
addition to the availability of data are given in Table 1. The areas were dened
in terms o f census subdivision boundaries (Table 2). We made one exce ption
to this decision rule in Winnipeg, which maintained two stations m onitoring
SO
2
from 1986 to 1991. In order to have some SO
2
data for Winnipeg, we
included the time series for this city in our analysis. Edmonton did not maintain
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
URBAN POLLUTION MIXTURE AND MORTALITY 19
TABLE 1. Monitoring Station by Census Subdivision (CSD), Pollutants Monitored, and Indication of Missing Data (M)
or Years With Low Numbers (L) of Daily Monitored Observatio ns (< 100/yr)
Station
City number CSD Pollutants monitored Comments
Montreal 50102 66025 O
3
, NO
2
, CO, SO
2
CO (L ’88), NO
2
(L ’89)
50103 66025 O
3
, NO
2
, CO, SO
2
NO
2
(M ’89–’91, L ’92)
50104 66025 O
3
, NO
2
, CO, SO
2
, PM CO (L ’86), NO
2
(L ’87), SO
2
(L ’96),
PM (M ’91)
50109 66070 O
3
, NO
2
, CO, SO
2
, COH, PM O
3
, NO
2
, CO, SO
2
, COH (L ’86),
SO
2
(L ’91), PM (M ’90, ’91, ’96)
50110 66020 O
3
, NO
2
, CO, SO
2
, COH SO
2
(L ’89, M ’96), NO
2
(L ’91)
50115 66025 O
3
, NO
2
, CO, SO
2
, COH
50116 66025 O
3
, NO
2
, CO, SO
2
, COH
50120 66140 O
3
, NO
2
, CO, SO
2
, COH No data ’86, ’87, CO (L ’89, M ’96),
NO
2
(L ’91, ’92), S O
2
(L ’89, M ’96)
Ottawa-Hull 50203 81020 NO
2
, CO, SO
2
No data ’95, ’96
60101 06014 O
3
, NO
2
, CO, SO
2
No data ’91, SO
2
(M ’95, ’96)
60104 06014 O
3
, NO
2
, CO, SO
2
, COH, PM
Toronto 60403 20019 O
3
, NO
2
, CO, SO
2
, COH, PM PM (M ’86– ’94)
60410 20001 O
3
, NO
2
, CO, SO
2
, COH
60413 20019 O
3
, NO
2
, CO, SO
2
,
60415 21005 O
3
, NO
2
, CO, SO
2
NO
2
, CO, SO
2
(M ’95, ’96)
60417/ 20004 O
3
, NO
2
, CO, SO
2
, COH, PM
60424
L ’90, PM (M ’90, ’91)– station moved
in ’90 from 60417 to 60424
60418 20004 O
3
, NO
2
No data ’95, ’96
60419 20004 O
3
, NO
2
No data ’94, ’96
60421 20008 O
3
, NO
2
No data ’86, ’87
60423 20008 O
3
, NO
2
No data ’86, ’87
Windsor 60204 37039 O
3
, NO
2
, CO, SO
2
, COH, PM PM (M ’86, ’95, ’96 )
60211 37039 SO
2
, COH, PM PM (M ’86–’89)
60212 37039 SO
2
Winnipeg 70118 11040 O
3
, NO
2
, CO, SO
2
, COH SO
2
(M ’92–’96)
70119 11040 O
3
, NO
2
, CO, SO
2
, COH, PM SO
2
(M ’92–’96)
Edmonton 90122 11061 O
3
, NO
2
, CO, COH
90130 11061 O
3
, NO
2
, CO, COH, PM
Calgary 90218 06016 SO
2
, COH
90222 06016 O
3
, NO
2
, CO, COH
90227 06016 O
3
, NO
2
, CO, COH, PM PM (M ’86)
90204 06016 PM PM (M ’88–’96)
Vancouver 100110 15025 O
3
, NO
2
, CO, SO
2
, COH
100111 15043 O
3
, NO
2
, CO, SO
2
, COH, PM
100112 15022 O
3
, NO
2
, CO, SO
2
, COH
100118 15022 O
3
, NO
2
, CO, SO
2
, PM All pollutants (L ’86)
100120 15025 O
3
, NO
2
100121 15056 O
3
, NO
2
NO
2
(L ’87)
100122 15038 O
3
, NO
2
100124 15043 O
3
, NO
2
100125 15011 O
3
, NO
2
NO
2
(M ’86)
100126 15025 O
3
, NO
2
100127 15004 O
3
, NO
2
100128 15015 O
3
, NO
2
100129 15039 O
3
No data ’86
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
20 R. T. BURNETT ET AL.
TABLE 2. Selected Characteristics of the Eight Canadia n Cities Including Census Subdivision Code, 1991
Population, Percentage of Census Metropolitan Area (CMA) Selected, and Daily Average Nonaccidental
Mortality Rate by City
1991 % CMA
City Census subdivision population selected Deat hs/day
Montreal All of Census Division 66 1,775,871 55 38.8
Ottawa-Hull 81015, 81020, 81025 , 06006, 06009,
06011, 06012, 06014
728,789 78 11.7
Toronto 20001, 20004, 20006, 20008, 20014,
20019, 21005
2,739,159 70 46.5
Windsor 37039 191,435 73 4.5
Winnipeg 11040 615,215 93 12.6
Calgary 06016 710,795 94 8.5
Edmonton 11061 616,741 73 8.7
Vancouver 15004, 15011, 15015, 15018, 15022,
15025, 15029, 15034, 15038, 15039,
15043, 15051, 15803
1,318,693 82 22.1
any SO
2
stations within the selected ce nsus subdivision boundaries. Daily
average or daily 1-h maximum values were averaged among available data
from all stations in each city to form a single time series for analysis.
Data on daily average temperature, daily average relative humidity, and
maximum change in barometric pressure within a day (a measure of frontal
activity) were obtained from airports within or near each city.
Mortality Data
We abstracted the number of persons who died from a nonaccidental
underlying cause (International Classication of Disease 9th Revision ICD9:
<
800) on each day from 1 January 1986 to 31 December 1996 and who
lived and died in any one of the specied census subdivisions. Deaths were
then aggregated at the city level.
Statistical Methods
Adjusting f or Temporal Variation
The objective of our analysis was to
relate daily variations in air pollution levels to daily variations in mortality rates
on a temporal basis. There are, however, a number of other factors inuencing
temporal variation in both deaths and air pollution. More people die in winter
than summer. This is like ly, in part, due to increased infection in those months
in which people spend a greater amount of time indoors (Leech et al., 1996).
The average daily number of deaths per city varies fro m 19.1 to 19.4 from
Monday to Saturday, with a lower rate observed on Sunday (18.8 deaths/day).
Air pollution levels also vary by season and day of the week. Finally, weather
factors, such as temperature and humidity, are predictors of both deaths and
air pollution.
We did not wish to attribute variation in mortality to variation in air pol-
lut ion if the variation in mortality was due, in whole or in par t, to common,
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
URBAN POLLUTION MIXTURE AND MORTALITY 21
but possibly unrelated, temporal cycles and weather factors. We therefore ad-
justed the time series of the daily number of deaths for temporal trends using a
LOESS nonparametric smoothed representation of day of study (Cleve land &
Devlin, 1988) and fo r day-of-the-week effects for each city separately. Bartlett’s
test for “white noise” and examination of the autocorrelation functions were
used to assess serial correlation in the data (Priestly, 1981). The serial correla-
tion structure in the residuals after tting models with LOESS smoothers was
examined. We considered LOESS smooth functio ns of day of study of length
30, 90, and 150 days.
Determining the Weather Model
We also adjusted the mortality series
for the inuences of weather. The weather variables were adjusted for temporal
trends using the LOESS smoother of day of study and dummy variables for day-
of-the-week effects. The temporally adjusted effects of weather on mortality
were modeled using spline-smoothed functions of daily average temperature,
daily average relative humidity, and maximum change in barometric pressure
wit hin a day, recorded on the date of death and one day p rior to death. In
order to identify the smallest number of weather variables req uired to predict
deaths, we used a forward inclusion stepwise regression procedure to select
a minimally sufcient set of weather variables ne eded to statistically predict
daily variations in mortality rates. Akaike’s Information Criterion (AIC) is a
linea r function of the residual deviance and the model degrees of freedom
and was used to identify important predictors of mortality. We found that daily
average temperature and maximum change in barometric pressure recorded
on the day of death were sufcient to predict mortality am ong the varia bles
considered.
The Regression Mode l
We assumed a common form of the relationship
betwe en air pollution or weather and mortality among cities and a common
effect of air pollution a nd weather. The city-specic temporally adjusted time
series were pooled with parameter estimates obtained from a single m odel
based on data from all cities combined. This approach was adopted due to the
limited number of particulate observations in each city. Temporally adjusting
all time series prior to relating them was done in order to remove unwanted
cycles in the data at a yearly and seasonal tim e scale that might confound the
association among the variables at the daily time scale that we are interesting
in detecting.
The air pollution, weather, and mort ality time series were adjusted for
temporal trends using a LOESS smoother of day of study and dummy variables
for day of week. The ltered time series were then re lated using the model
E
(
y
tj
)
=
f
(y)
tj
exp
{b
T
(
x
tj
f
(x)
tj
)
+
g
T
s
(
w
tj
f
(w)
tj
)
}
where
E
(
y
tj
) is the expected number of deaths on day
t
in the
j
th city; exp
is the exponential function;
b
T
is the transpose of the vector of unknown
parameters relating the vector of temporally ltered air pollution values on
day
t
of the
j
th city to temporally ltered deaths;
x
tj
is the vector of air pollution
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
22 R. T. BURNETT ET AL.
values;
g
T
is the transpose of the vector of unknown parameters relating the
spline smoothed function,
s
(
·
), of the temporally ltered weather variables to
temporally ltered deaths; and
w
tj
is a vector of weather factors. Here,
f
(y)
tj
(
f
(x)
tj
;
f
(w)
tj
), are the predicted values of daily deaths (air pollution; weather) as
obtained from the generalized additive model with temporal trends and day-
of-the-week effects. We assumed that the residual variance was proportional
to the expected response, thus accommodating over- (or under-)dispersion
relative to Poisso n variation.
We used generalize additive mo dels, as implemented in S-PLUS (Statistical
Sciences, Inc., 1993), to model the data, the reby obtaining relative excess daily
mortality rates expressed for a unit change in air pollution,
b
, as well as their
corresponding standard errors corrected for non-Poisson variation. We report
the percentag e increase in daily numbers of deaths corresponding to the mean
air pollution concentration as well as the ratio of the parameter e stimate to its
standard error (
T
value).
Using Principal Component Analysis for Multipollution Regression
Models
Unstable parameter estimates may be obtained in multipollutant
models due to the positive correlation between pollutants in the urban atmo-
sphere. Consequently, we estimate air pollution effects within the urban air
pollution mixture using principal components (PC) regression analysis meth-
ods with the varim ax rotation ( SAS Institute, Inc., 1989). We rst obtained
orthogonal linear combinations (factors) of the preltered air pollution vari-
ables. With the varimax rotation option, each factor explains a similar amount
of the variation in pollutant concentrations. All factors were re gressed aga inst
temporally preltered daily number of deaths simultaneously, adjusting for
preltered weather variab le s using the generalized additive models described
earlier. Parameter estimates associated with these factors are nearly uncorre-
la ted.
We used the following procedure to translate the results from the prin-
cipal component regression analysis to estimates of the percent increase in
mortality associated with an increase in each pollutant, as evaluated at their
mean concentrations. First, the facto r loadings (coefcients of the linear com-
binations of variables for each factor) were normalized such that the loadings
for each factor summed to unity. Second, the regression coefcients for each
factor were normalized so that they also summed to unity. Then the matrix
of normalized factor regression co efcients was multiplied by the vector of
normalized regression coefcients of the factors to obtain the percenta ge con-
tribution for each pollutant to the total effect on mortality. The tot al effect of
the mixture on mortality was determined by multiplying the regression coef-
cients, obtaine d from a model including all pollutants, times the corresponding
mean value, and summing among pollutants. The effect of each pollutant on
mortality was then determined by multiplying the proportion of the effect of
the mixture attributable to each pollutant by the total effect of the mixture.
Standard errors of these estimates were derived using the standard errors of
the factor regression coefcients.
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
URBAN POLLUTION MIXTURE AND MORTALITY 23
RESULTS
Summary Information
City-specic daily mor tality rates are given in Table 2, along with the 199 1
population and the percentage of t he Census Metropolitan Area (CMA) se-
lected for study in each city. For most cities, 70–80% of the population in the
CMA was selected. The selection criterion was based on proximity of the pop-
ula tion to a monitoring station. Only 55% of the population in the Montreal
CMA (the island of Montreal) was selected, due to limited monitoring data
outside the island.
The percentage of missing values for the gaseous pollutants, their distri-
bution, and the number of monitoring stations used in the analysis are given
in Table 3 by city, based on the combined air pollution data among stations
wit hin a city. In general, there were only a few missing days among the total
of 4018 possible days of observation per city. Exceptions were for measure-
ment s of SO
2
in Winnipeg (45.6% missing) and Edmonton (100% missing).
There were far fewer particulate measurements due to the 6-day sampling
schedule.
Windsor had the highest ave rage concentrations for ozone, nitrogen diox-
ide, sulfur dioxide, and ne particulate mass, with Winnipeg having generally
the lowest concentrations for the se pollutants.
Temporal Filtering
To select the length of a temporal ltered, we stipulated that the residua l
time series should be consistent with a white noise process. The 90-day span
for day of study did ind eed produce such a residual in all cities except Montreal
and Windsor. Examination of the time series in Montreal revealed 4 days with
very high numbers of deaths (12 July 1987 with 80 deaths; 13 July 1987
wit h 94 deaths; 17 June 1994 with 95 deaths; and 18 June 1994 with 82
deaths). Removal of these 4 days was sufcient to produce a residual time
series consistent with white noise (
p
> .05, Bartlett’s test). When the 30-day
span was used, we found negative serial correla tions for time lags up to 7 days,
ranging from 0.0 5 to 0.10, and the resultant time series was signicantly
different (
p <
.001) from the expected white noise series. For the 150-day
span, the white noise assumption was rejected for Montreal and Toronto
(
p <
.05), but not for the other citie s. There was some evidence to reject
the hypothesis of white noise in Windsor for the 90-day span (
p
=
.01). Here,
a negative serial correlation of 0.05 at a lag of 4 days was observed. We
therefore selected a span of 90 days for the main analysis, as there was little
evidence to reject the hypothesis of white noise in 7 out o f 8 c ities.
Correlation Among Pollutants
Ozone was weakly correlated with the other air pollutants ( Table 4), and
the other pollutants were moderately correlated with each other. The highest
correlatio n was observed for NO
2
and CO (
r
=
.65). The gases were more
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
TABLE 3.
Summary Statistics for Gas- and Particulate-Phase Pollution by City
City
Gas/PM (units) Statistic Montreal Ottawa Toronto Windsor Winnipeg Edmonton Calgary Vancouver All cities
O
3
(ppb) % Miss 0.0 0.4 0.0 2.2 1.5 0.1 0.0 0.0 0.5
Mean 28 28 36 36 31 32 34 27 31
CV 50 44 51 60 39 42 34 43 48
95th Perc 54 49 72 78 53 55 53 46 5 9
Max 116 93 163 159 99 89 89 114 163
# Mon 8 2 9 1 2 2 2 12 38
NO
2
(ppb) % Miss 0.4 0.5 0.0 4.6 1.1 0.1 0.0 0.0 0.8
Mean 22 20 25 26 15 25 26 20 22
CV 38 42 32 38 46 38 34 30 40
95th Perc 37 35 38 43 29 43 41 31 3 8
Max 80 69 72 134 56 72 77 51 134
# Mon 8 3 9 1 2 2 2 12 39
CO (ppm) % Miss 0.3 0.0 0.0 2.6 0.8 0.1 0.0 0.0 0.5
Mean 0.6 0.8 1.0 0.9 0.6 1.2 1.0 1.1 0.9
CV 65 52 37 51 42 60 62 51 59
95th Perc 1.4 1.6 1.6 1.6 1.1 2.7 2.1 2.2 1.8
Max 4.6 3.3 4.2 4.0 3.0 6.3 7.2 4.5 7.2
# Mon 8 3 5 1 2 2 2 4 27
24
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
SO
2
(ppb) % Miss 0.3 0 .8 0.0 0.1 45.6 100 2.4 0.0 18.6
Mean 5.6 3.8 5.0 7.3 1.2 NA 3.7 5.1 4.7
CV 80 88 81 58 107 NA 66 58 81
95th Perc 15 10 13 15 4 NA 8 11 12
Max 35 34 30 27 27 NA 23 24 35
# Mon 8 3 5 3 2 0 1 4 26
PM
2.5
(
µ
g/ m
3
) No. obs . 863 526 900 851 554 571 599 683 5547
Mean 15.0 11.6 15.4 17.7 9.5 9.9 10.3 12.7 13.3
CV 69 74 67 64 71 72 69 57 71
95th Perc 34 29 36 40 21 24 25 29 32
Max 72 54 71 86 71 56 52 43 86
PM
10-2.5
(
µ
g/ m
3
) No. obs. 854 434 889 851 541 508 598 565 5240
Mean 12.4 9.2 10.9 13.4 16.8 14.0 15.2 8.9 12.6
CV 66 70 60 65 76 73 68 58 72
95th Perc 29 21 23 29 38 35 36 18 30
Max 82 52 68 99 78 73 84 35 99
PM
10
(
µ
g/m
3
) No. obs. 853 433 889 850 538 508 598 565 5234
Mean 27.3 20.4 26.4 31.0 26.2 23.4 25.5 21.5 25.9
CV 59 61 56 54 62 57 58 52 59
95th Perc 60 42 53 63 56 50 55 44 54
Max 121 76 102 110 112 86 114 63 121
Note
. % Miss, percentage of mis sing observations of a possible 4018; CV, coefcient o f variation
´
100; 95th Perc, 95th percentile; Max, maximum
concentration; # Mon, number of monitoring sites; No. obs., number of ob servations.
25
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
26 R. T. BURNETT ET AL.
TABLE 4. Correlations (
´
100) of Temporally Filtered Gas- and Par ticulate-Phase Pollution
Pollutant O
3
NO
2
CO SO
2
PM
2.5
PM
10-2.5
PM
10
O
3
1 12 5 8 29 23 32
NO
2
1 65 49 53 34 53
CO 1 42 44 29 45
SO
2
1 40 25 40
PM
2.5
1 37 84
PM
10-2.5
1 81
PM
10
1
strongly associated with the ne fra ction than coarse particulate mass (Ta-
ble 4). Correlations among pollutants were relatively stable among cities. For
example, the correlation between concentrations of CO and NO
2
ranged from
.61 to .73, while the correlation between PM
2.5
and SO
4
ranged from .55 to
.80. The city-specic correlations between O
3
and the other gases were also
lo w (
r <
.20). The correlation between PM
2.5
and PM
10-2.5
was also highly
stable among cities (.3 4
< r <
.43).
Modifying Effects of Gaseous Po llutants on
Particulate–Mortality Association
We next considered the associatio n between particulate pollutio n and
mortality after contro lling for the gaseous pollutants. For this analysis, only
days on which there were measurements of PM
10
were considered. Here,
the gas and particulate air pollution data were ltered with a 90-day LOESS
smoother for day of study and adjusted for day-of-the-week effect s.
The percent increases in mortality associated with increases in air po llu-
tants evaluated at their mean concentrations are given in Ta ble 5 for time lags
of 0 and 1 days. Time lags of 2 to 5 days were also examined. The strongest
association with mortality for all pollutants considered were for lags of 0 or
1 days. The effects for P M
2.5
, PM
10-2.5
, and PM
10
were adjusted for each of the
four gases separately. At lag 0, the effect of O
3
was insensitive to adjustment
for the ma ss measurements. However, the effects at lag 0 of NO
2
, SO
2
, and
CO were explained by the mass measurements. In turn, much of the ne mass
effect could be explained by O
3
. For lag 1, the ozone effect was reduced by
half following adjustment for ne mass. In addition, we found effects reduced
by half for ne mass after adjusting for NO
2
. The effects of CO and SO
2
were
also reduced by half after adjustment for ne mass. Adjusting for the coarse
fraction had less o f a modifying effect on the gases than ne mass. Adjusting
for PM
10
had a modifying effect on the gases similar to adjusting for ne mass.
The simultaneous effect of PM
2.5
, PM
10-2.5
, and the four gases on mortality
was examined using principal component analysis (see Table 8, Model I). The
effects of each of the pollutants were reduced as compared to the effects
of the pollutants examined separately. The combined effect of PM
2.5
and
PM
10-2.5
(1.6%) was less than the combined effect of the four gases (4.1%),
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
URBAN POLLUTION MIXTURE AND MORTALITY 27
TABLE 5. Percent Increase in Temporally Filtered Daily Nonaccidental Deaths Associ-
ated With Incre ases in Particulate- and Gas-Phase Pollutants Evaluated at Their Study
Mean Adjusted for Particulate Phase Pollutants,Temperature,and Change in Barometric
Pressure for Those Days in Which There Were PM
10
Measurements Available
Lag 0 Lag 1
Gas PM Gas PM Gas PM
None PM
2.5
NA 1.2 (2.3) NA 1.6 (3.1)
PM
10-2.5
NA 0.6 (1.0) NA 0.9 (1.4)
PM
10
NA 1.4 (2.1) NA 1.9 (2.8)
O
3
None 4.0 (4.2) NA 1.6 (1.8) NA
PM
2.5
3.6 (3.6) 0.6 (1.2) 0.8 (0.9) 1.5 (2.6)
PM
10-2.5
3.9 (4.0) 0.1 (0.2) 1.4 (1.5) 0.7 (1.1)
PM
10
3.7 (3.7) 0.6 (0.9) 0.9 (0.9) 1.7 (2.4)
NO
2
None 1.3 (1.2) NA 4.0 (3.6) NA
PM
2.5
0.2 ( 0.2) 1.3 (1.9) 3.1 (2.4) 0.7 (1.1)
PM
10-2.5
1.0 (0.9) 0.4 (0.7) 3.9 (3.3) 0.1 (0.2)
PM
10
0.0 ( 0.0) 1.4 (1.7) 3.3 (2.5) 0.7 (0.9)
CO None 0.4 (0.4) NA 2.0 (2.3) NA
PM
2.5
0.7 ( 0.7) 1.4 (2.4) 1.1 (1.1) 1.5 (2.2)
PM
10-2.5
0.1 (0.2) 0.6 (1.0) 1.8 (2.1) 0.5 (0.8)
PM
10
0.5 ( 0.6) 1.6 (2.1) 1.2 (1.3) 1.5 (2.0)
SO
2
None 0.3 (0.5) NA 1.1 (2.0) NA
PM
2.5
0.3 ( 0.5) 1.2 (2.3) 0.6 (0.9) 1.3 (2.1)
PM
10-2.5
0.2 (0.3) 1.3 (2.2) 1.0 (1.7) 0.7 (1.0)
PM
10
0.2 ( 0.4) 0.6 (0.9) 0.6 (1.0) 1.5 (2.0)
Note. Concentrations based on day of death (lag 0) and day prior to death (lag 1)
(ratio of log-relative risk to standard erro r given in parentheses). NA, not applicable.
indicating part iculate mass does not fully characterize the effect of the mixture
of urban atmospheric pollution on mortality in Canadian cities.
Effects of Elemental and Ion Concentrations From PM
2.5
on Mortality
The preceding analyses show a positive and statistically signicant associa-
tion between ne mass and mortality, although the association fo r the coarse
fraction was weaker. In o rder to identify components of the ne fraction that
may be related to mortality, we examined the association between elemental
and ion concentratio ns from the ne fraction and daily nonaccidental deaths.
Summary statistics for the elemental and ion data from the ne fraction are
given in Table 6. Out of a possible 5547 days in which ne mass was col-
lected, concentrations were available on 4255 days for most elements, except
for Ce, Pr, Nd, W, Hg, Mg, and Na, as monitoring for these compounds only
commenced in May of 1992.
The correlations b etween ltered mortality and ltered elemental concen-
trations are given in Table 7. Sulfur (S) showed the highest correlation with
ne mass (
r
=
.74), with Pb, Si, Fe, K, Zn, Mn, P, and Se modestly correlated
wit h PM
2.5
(
r
=
.3–.5). Several elements were weakly correlated (
r
=
.2–.3)
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
28 R. T. BURNETT ET AL.
TABLE 6. Summary Statistics for Ele mental Data (
µ
g/ m
3
) from PM
2.5
Percentiles
Average Percent
detection above Number of
Element limit (DL) DL Mean 50th 75th 95th observations
Pb 0.0008 97 0.0405 0.0115 0.0447 0.1710 4437
Al 0.0031 91 0.2019 0.0571 0.1570 0.8264 4254
Si 0.0017 93 0.1251 0.0900 0.1560 0.3336 42 55
S 0.0007 97 1.0053 0.6840 1.2200 2.9420 4255
Fe 0.0021 99 0.0812 0.0565 0.0971 0.2332 4255
K 0.0023 99 0.0786 0.0603 0.0963 0.1871 4255
Ca 0.0028 98 0.0773 0.0524 0.0902 0.2170 4255
Zn 0.0009 99 0.0258 0.0134 0.0277 0.0903 4255
Br 0.0004 91 0.0110 0.0029 0.0102 0.0464 4442
Mn 0.0020 96 0.0134 0.0103 0.0165 0.0339 4255
P 0.0009 79 0.0266 0.0142 0.0330 0.0945 4255
Cu 0.0018 63 0.0077 0.0046 0.0106 0.0234 4255
As 0.0005 56 0.0009 0.0006 0.0012 0.0027 4255
Ni 0.0010 47 0.0016 0.0006 0.0020 0.0048 4255
Sr 0.0003 52 0.0005 0.0003 0.0006 0.0014 4255
Cl 0.0023 54 0.0492 0.0079 0.0335 0.1982 4255
Se 0.0005 49 0.0010 0.0004 0.0011 0.0036 4255
Ti 0.0047 44 0.0062 0.0029 0.0077 0.0160 4255
Ga 0.0011 36 0.0023 0.0006 0.0022 0.0096 4255
V 0.0037 3 1 0.0037 0.0020 0.0040 0.0106 4255
Sc 0.0037 29 0.0056 0.0026 0.0050 0.0232 4255
Cr 0.0029 22 0.0022 0.0016 0.0017 0.0060 4255
Co 0.0014 12 0.0009 0.0007 0.0008 0.0021 4255
Ge 0.0007 16 0.0005 0.0004 0.0004 0.0013 4255
Rb 0.0003 12 0.0002 0.0002 0.0002 0.0005 4255
Y 0.0003 7 0.0002 0.0002 0.0002 0.0004 4255
Zr 0.0004 29 0 .0003 0.0002 0.0004 0.0010 4255
Nb 0.0004 8 0.0002 0.0003 0.0003 0.0007 4255
Mo 0.0005 32 0.0005 0.0003 0.0005 0.0018 4255
Pd 0.0017 2 0.0009 0.0 012 0.0012 0.0014 4255
Ag 0.0018 9 0.0013 0.0013 0.0013 0.0028 3648
Cd 0.0038 22 0.0017 0.0017 0.0017 0.0044 4255
In 0.0023 21 0.0021 0.0018 0.0023 0.0060 4255
Sn 0.0029 47 0.0051 0.0022 0.0068 0.0135 4255
Sb 0.0030 42 0.0048 0.0024 0.0069 0.0164 4255
Te 0.0031 26 0.0036 0.0024 0.0045 0.0136 4255
I 0.0033 9 0.0020 0.0025 0.0025 0.0042 4255
Cs 0.0056 11 0.0032 0.0045 0.0045 0.0055 4255
Ba 0.0074 27 0.0054 0.0060 0.0060 0.0133 4255
La 0.0094 19 0.0059 0.0078 0.0078 0.0129 4255
Ce 0.0023 29 0.0024 0.0014 0.0028 0.0080 2038
Pr 0.0028 31 0.0028 0.0018 0.0033 0.0082 2038
Nd 0.0033 41 0.0039 0.0028 0.0050 0.0109 2038
W 0.0030 49 0.0039 0.0023 0.0058 0.0103 2038
Hg 0.0008 28 0.0006
0.0005 0.0006 0.0012 2038
Mg 0.0040 80 0.0233 0.0141 0.0279 0.0756 2038
Na 0.0463 76 0.2428 0.1510 0.3046 0.7801 2038
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
TABLE 7. Percent Increase in Filtered Daily NonaccidentalDeaths Associated
With an Increase in Filtered Elemental Concentration (
µ
g/ m
3
) Equivalent to
the Study Mean Based on PM
2.5
Percent increase
Correlation (
´
100)
Element with PM
2.5
Lag 0 Lag 1
Pb 33 0.0 ( 0.1) 0.3 (0.8)
Al 16 0.1 (0.6) 0.1 ( 0.7)
Si 34 0.3 (0.7) 0.4 (1.0)
S 74 0.8 (1.9) 1.4 (3.2)
Fe 48 0.3 (0.5) 1.2 (2.3)
K 41 0.4 (0.9) 0.6 (1.3)
Ca 21 0.8 (2.2) 0.5 (1.2)
Zn 37 0.6 (1.8) 0.8 (2.4)
Br 29 0.1 ( 0.4) 0.1 (0.3)
Mn 47 0.1 (0.3) 0.5 (1.3)
P 46 0.2 (0.9) 0.5 (1.7)
Cu 6 0.4 (1.7) 0.2 ( 0.8)
As 21 0.3 (1.4) 0.1 (0.3)
Ni 26 0.5 (1.5) 0.7 (1.8)
Sr 12 0.1 (1.1) 0.2 ( 1.8)
Cl 16 0.1 (0.8) 0.1 ( 0.7)
Se 35 0.0 (0.1) 0.2 (0.8)
Ti 28 0.3 ( 1.1) 0.2 (0.9)
Ga 7 0.2 (0.4) 0.9 ( 1.7)
V 26 0.1 ( 0.4) 0.1 (0.2)
Sc 0 0.6 (1.7) 0.1 (0.4)
Cr 13 0.7 (1.4) 0.3 (0.5)
Co 25 1.7 (2.0) 1.4 (1.6)
Ge 7 0.4 ( 0.5) 1.0 (1.2)
Rb 21 0.6 (1.1) 0.2 (0.3)
Y 4 0.4 ( 0.4) 0.0 ( 0.0)
Zr 11 0.9 (2.5) 0.2 ( 0.6)
Nb 3 0.0 ( 0.0) 0.8 ( 0.7)
Mo 8 0.6 ( 1.1) 0.5 (0.8)
Pd 3 0.5 ( 0.3) 4.3 ( 2.4)
Ag 1 0.0 ( 0.7) 0.0 ( 0.6)
Cd 15 0.2 ( 0.8) 0.2 (0.7)
In 8 0.7 (1.6) 0.2 (0.4)
Sn 5 0.1 (0.5) 0.0 (0.1)
Sb 15 0.1 ( 0.1) 0.1 (0.2)
Te 10 0.4 (1.2) 0.3 ( 0.6)
I 8 0.2 ( 0.2) 0.1 (0.1)
Cs 4 0.9 ( 0.9) 0.2 (0.2)
Ba 11 0.7 ( 1.0) 0.4 ( 0.5)
La 0 0.1 (0.1) 1.4 (1.8)
Ce 2 0.4 ( 0.6) 0.5 (0.7)
Pr 7 0.6 ( 0.7) 0.1 (0.2)
Nd 7 0.4 ( 0.4) 0.2 (0.2)
W 5 1.8 ( 1.3) 0.1 (0.1)
Hg 1 1.7 (1.8) 1.4 ( 1.2)
Mg 21 2.1 (3.0) 0.7 ( 0.9)
Na 17 0.9 (1.6) 0.1 (0.2)
Note. Concentrations recorded on the day of death (lag 0) or the day prior
to deat h (lag 1) (T values given in parentheses).
29
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
30 R. T. BURNETT ET AL.
wit h ne mass (Ca, As, Ni, Ti, V, Co, Rb, Mg). The remaining elements (Al, Cu,
Sr, Cl, Ga, Sc, Cr, Ge, Y, Zr, Nb, Mo, Pd, Ag, Cd, In, Sn, S b, Te, I, Cs, Ba, La,
Ce, Pr, Nd, W, Hg, Na) displayed little association with PM
2.5
.
Sulfur, Ca, Zn, Cu, Sc, Co, Zr, and Mg displayed some evidence of a
positive association with mortality for concentrations measured on the day
of death (
T
ratio > 1.645, corresponding to a one-sided
p
value of .0 5). For
those concentrations recorded on the day prior to death, S, Fe, Zn, P, Ni, and
La displayed some evidence of an effect on mortality. Note that only S and
Zn displayed a positive and statistically signicant association for both lagging
times.
In further analyses, we examined concentrations of the sulfate ion (SO
4
)
instead of the element sulfur due to chemical analytical considerat ions (Brook
et al., 1997). There were 4438 days in which sulfate concentrations were
obtained from the ne fraction, with a lag 0-day effect of 0.9% (
T
=
1.9) and
a lag 1-day effect of 1.2% (
T
=
3.5) evaluated at the study mean of 2.6
µ
g/ m
3
.
Sulfates were highly correlated wit h ne mass (
r
=
.73).
We further consid ered the four components of the ne fraction that were
most stat istically signicantly positively associated with mortality from lag 1
day data: sulfates, zinc, nickel, and iron. We observed some variation in the
effect of PM
2.5
on mortality among cities, ranging from a 1.1% increase in daily
mortality to a 6.9% increase. The range in the PM
10-.2.5
effect was 1.0% to
3.8%, while the range in effect for PM
10
was from 0.4% to 7.0%. The variability
in the effect estimates among cities for the elements and sulfate based on
PM
2.5
was much greater than the corresponding effect for mass. The range
in the percent increase in mortality for sulfate was from 4.3 to 3.9, for iron
was 5.0 to 3.1, for nickel was 10.7 to 2.2, and for zinc was 3.3 to 3.7.
The range in increase in mortality based on the sum of all four compounds
was 8.4% to 7.7%, with 2 of 8 cities displaying a negative asso ciation with
mortality.
Sulfate was weakly correlated with Fe (
r
=
.2 8), Ni (
r
=
.15), and Zn (
r
=
.26). Ni was also weakly correlated with the other factors (
r <
.26). Fe and
Zn were clearly associated (
r
=
.59 ) however. Sulfates, nickel, and zinc were
also weakly corre lated with the four gases (
r <
.32 ). However, the correlation
betwe en iron and three of the gases (CO, NO
2
, and SO
2
) was slightly higher
(.35
< r <
.39 ).
An index of elemental carbon (Groblicki et al., 1981), the coefcient of
haze (COH), was monitored daily in each city. The effect for COH at lag
0 days, evaluated at the study mean of 2.6 hundred linear feet, was 0.2%
(
T
=
.3 ), while the effect at lag 1 day was 0.7% (
T
=
1.1) based on the
days in which PM
10
were available. COH was p ositively correlated with ne
mass (
r
=
.5 7), iron (
r
=
.4 3), sulfates (
r
=
.35), zinc (
r
=
.3 3), and nickel
(
r
=
.22 ). However, COH was highly associated with both CO and NO
2
(
r
=
.71), mo destly correlated with SO
2
(
r
=
.49 ), and weakly related to O
3
(
r
=
.05). Note that COH displayed the second largest corre lation with PM
2.5
next to SO
4
.
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
URBAN POLLUTION MIXTURE AND MORTALITY 31
We further considered the combined effects of the four most strongly sta-
tistically associated constituents of part iculate pollution to mortality based on
la gged 1-day measurement data (SO
4
, Fe, Ni, and Zn). Four linear combina-
tions or components of these four variables were determined using principal
component analysis techniques with varimax rotation, and these four compo-
nent variables were regressed against ltere d mortality counts simultaneo usly.
The component loa dings were positive for all pollutants and components ex-
amined. The effects of these components on mo rtality were adjusted sepa-
rately for the four gases and COH. The effects of sulfur dioxide and COH in
this analysis were negative. As the effects of O
3
, CO, and NO
2
were positive,
we considered a further principal compone nt analysis with seven variables
(SO
4
, Ni, Fe, Zn, O
3
, CO, and NO
2
). The factor loa ding for each component
and pollutant were po sitive, reecting the correla tion structure in the original
data. [Ozone concentrations recorded on the day of death were used in this
analysis since they displayed a much stronger assoc iation with mo rtality than
concentrations recorded on the day prior to death (se e Table 5).] These seven
components were then related to ltered morta lity with the resulting estimates
of daily mortality given in Table 8 ( Model II). Each of these seve n pollutants
was positively related to nonaccidental mortality, with the largest effect ob-
served for ozone (2.0%) and the least effect for CO (0.7%). The effect of each
individual pollutant on mort ality is also given in Table 8. The estimates for the
TABLE 8. Percentage Increase in Daily Filtere d Nonaccidental Deat hs Associated With In-
creases in Daily Filtered Air Pollution Concentrations in Single and Multiple Pollutant Model
Specications, Controlling for Weather Variables
Multiple-pollutant models
Pollutant
a
Mean Single-pollutant
(units) concentration model Model I Model II
PM
10
(
µ
g/ m
3
) 25.9 1.9 (2.8)
b
NA
d
NA
PM
2.5
(
µ
g/ m
3
) 13.3 1.6 (3.1) 1.0 (2.9) NA
PM
10-2.5
(
µ
g/ m
3
) 12.6 0.9 (1.4) 0.6 (1.6 ) NA
O
3
(ppb) 31 3.4 (2.6) 1.6 (3.4) 2.0 (3.2)
NO
2
(ppb) 22 3.9 (3.0) 1.1 (3.2) 1.2 (2.7)
SO
2
(ppb) 4.7 1.1 (1.6) 0.7 (2.1) NA
c
CO (ppm) 0.9 2.1 (2.1) 0.7 (1.9) 0.7 (1.7)
SO
4
(
µ
g/ m
3
) 2.6 1.2 (3.5) NA 1.3 (3.5)
Zn (
µ
g/m
3
) 0.0258 0.8 (2.4) NA 0.8 (2.1)
Ni (
µ
g/ m
3
) 0.0016 0.7 (1.8) NA 0.8 (1.9)
Fe (
µ
g/ m
3
) 0.0812 1.2 (2.3) NA 0.8 (1.8)
Note. Effect estimates in multiple pollutant models obtained using principal component
analysis.
a
Pollutants recorded the day prior to death except for o zone which was recorded on day
of deat h.
b
T statistic.
c
Negative effect, removed from model.
d
NA, not included in model.
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
32 R. T. BURNETT ET AL.
gases were based on days in which elemental data were available. The effects
for all four gases were reduced after adjusting for constituents of particulate
matter. The effects for SO
4
, Zn, and Ni were not sensitive to adjustment for
the gases. However, the effect of Fe on mortality was slightly reduced (1.2%
to 0.8%).
Effects of Elemental and Ion Concentrations from PM
10-2.5
and
PM
10
on Mortality
Of the 47 components of PM
10-2.5
examined for lag 0-day data, 3 dis-
played positive and statistically signicant (
T
> 1.6 45) associations (indium,
sulfate, and ammonium) and 3 displayed negative and statistically signicant
associations (gallium, tungsten, and lead). For lag 1-day data, 4 compounds
displayed positive and statistically signicant associations with mortality (scan-
dium, manganese, nickel, and zinc), while a negative, stat istically signicant
association was observed for 6 elements (germanium, iodine, cesium, cerium,
praseodymium, and t ungsten). These results should be interpreted with cau-
tion since we found more compone nts negatively associated with mortality
than that displayed positive effects.
Only nickel and zinc displayed a positive and statistically signicant asso-
ciation with mortality at the same lagging time for both PM
2.5
and PM
10-2.5
data. We added the concentrations for nickel and zinc obtained from PM
2.5
and PM
10-2.5
to reect concentrations based on PM
10
data. A 2.9-
µ
g/m
3
(study
mean concentration) increase in nickel recorded on the day prior to death was
associated with a 1.1% increase in nonaccidental mortality (
T
ratio of 2.5). A
40.8-ng/m
3
(study mean concentration) increase in zinc was associated with
a 1.2% increase in mortality (
T
ratio of 3.1) .
We note that the sulfate concentrations recorded on the day of death were
associated with mortality (0.7%,
T
=
2.2), but not for concentrations recorded
on the day prior to death ( 0.1%,
T
=
0.2 ). Sulfate concentrations were much
lo wer for the PM
10-2.5
data (0.3
µ
g/m
3
) than for PM
2.5
data (2.6
µ
g/m
3
). The
la g 0-day increase in mortality for sulfates was 0.9% (
T
=
2.3), while the lag
1-day increase was 1.2% (
T
=
3.1 ) based on a change of 2.9
µ
g/m
3
. These
values were identical to those obtained from the PM
2.5
data, suggesting that
all the information in the association between sulfate concentrations and
mortality was in the ne particulate fraction and not the coarse fraction.
DISCUSSION AND CONCLUSIONS
Positive and statistically signicant associations were observed in this study
betwe en daily variations in both gas and particulate phase pollution and daily
uctuations in mortality rates in eight Canadian cities from 1986 to 1996. The
air pollution association with mortality could not be explained by temporal
variation in either mortality rates or weather factors.
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
URBAN POLLUTION MIXTURE AND MORTALITY 33
Health Effects of Components of the Urban Mixture
Fine particulate mass was a stronger predictor of mortality than coarse
mass, a result also reported by Schwartz et al. (1996) in six U.S. cities. The re-
sults of this study conrmed the association between nonaccidental mortality
and co ncentrations of gas-phase pollutio n that was observed previously in a
study of 11 Canadian cities from 1981 to 1991 ( Burnett et al., 1998b). Bates
(1998) suggested that the association with the gaseous pollutants, in particular
NO
2
, and mortality was due to particulate matter. The current study addresses
a limitation in our earlier work by including size-fractionated particulate mass.
Size-fractionated particulate mass accounted fo r 28% of the total effe ct of the
urban pollution mix on mortality, as characterized by the four gases and ne
and coarse mass. Howe ver, if additional constituents of particulate matter are
examined, such as elemental concentrations and sulfates, the risk attributable
to par ticulate phase pollution increased to 49%. This suggests that mass may
be a relatively crude indicator of the toxicity of particulate phase pollution in
urban environments.
We also note that the risk o f death associated with the gaseous pollutants
was more sensitive to adjustment for other pollutants in the urban mix than
the particle-phase pollutants. This suggests that e ither the gaseous pollutants
are indices for other air pollutants and/or theyare subject to greater exposure
error than the particulate measures.
Statistical Model
We have chosen to related tempora lly ltered weather and air pollution
values to temporally ltered mortality counts, with ltering conducted on a
city-specic basis. In this ma nner we have removed the potentially confound-
ing temporal cycles in all time series prior to linking them together. Residuals
from these temporally ltered series are then compare d across cities in a single
analysis.
An alternate statistical approa ch is to simultaneously regress unltered air
pollution and weather data on daily mortality counts adjusting for temporal
trends in the mortality time series only, using a smooth function of day of study.
This analysis would be conducted for each city separately and a summar y
estimate of the air pollutio n effect given by a weighted average of the city-
specic effects, with weights given by the inverse of the estimation error. We
conducted this “coadjustment” analysis for PM
2.5
, PM
10-2.5
, and PM
10
and
compared the results to our “preadjustm ent” approach.
The p ercent increase in daily mortality associated with an increase in
PM
2.5
evaluated at the study mean concentration of 13.3
µ
g/ m
3
was 1.9% un-
der the coadjustment method. This is slightly higher than that obtained by our
model in which both mortality and air pollution are temporally ltered prior to
linking them together (1.6%). Similar results were observed for PM
10-2.5
(1.2%
for coadjusted and 0.9% for preadjusted) and PM
10
(2.1% for c oadjusted and
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
34 R. T. BURNETT ET AL.
1.9% for preadjusted). These results suggest that there is additional information
in the time series data relating air pollution and mortality that is not captured
in the high-frequency or day-to-day cycles in the re spective series. The pread-
justed m ethod removes all non-high-frequency information in the data. The
coadjusted approach can leave some residual confounding in the cycles be-
tween mortality and air pollutio n. Particulate levels are highest in the winter,
as are mortality counts. In the coadjusted approach, the cycles in these two
time series compete to predict mortality. Thus some of the midfrequency or
seasonal cycle in mortality may be captured by a corresponding cycle in air
pollution, thus inating the air pollution effect. If the cycles are in conict,
such as with ozone or temperature and mortality, the effect estimates will be
biased downward.
The preadjustment approach controls for such confounding. Estimates of
the air pollution effect based on the preadjustment approach reect the true
association between short-term (day-to-day) variation in air pollution and mor-
tality, while estimates based on the coadjustment method reect t he sum of
short-term and residual me dium-term concentrations. This residual median-
term co ncentrations effect on mortality is subject to potentia l confounding
due to other temporally varying risk factors to a greater extent than the short-
term concentration signal. We thus reco mmend the use of the prea djustment
method, which, to a greater extent, controls for po tentially confounding tem-
poral cycles than the coadjustment technique.
Variation in Effect Among Cities
The variation in the percent increase in mortality associated with con-
centrations of ne particulate mass was less than the variation in effect of
any of the four constituents examined, and of the sum of the effects of t he
constituents. This sugg ests that there were unmeasured components of ne
particulate matter t hat affect health. Elemental and organic carbon accounts
for approximately half the mass of ne particulate matter in Canadian cities
(Brook et al., 1997). Concentrations of these co mpounds were also more
stable between cities than the elemental data (Brook et al., 1997). COH is
highly correlated with ele mental carbon. We found a positive but statisti-
cally weak association between COH and mo rtality, with a 0.7% increase
in daily mortality associated with a 0.26 change in COH per thousand linear
feet. Adjustment for the constituents of the atmospheric mix eliminated the
COH association with mortality based on days in which elem ental data were
available.
COH was measured daily in each city. The percent increase in daily deaths
associated with COH was 1.1% (
T
=
5.1) for lag 0-day concentrations and
1.2% (
T
=
5.7) for lag 1-day concentrations, with a 1.7% (
T
=
6.6) effect
observed for 2-day average values. The COH effect persisted after adjustment
for the four gases (results not shown). We thus urge caution in interpreting the
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
URBAN POLLUTION MIXTURE AND MORTALITY 35
negative results for COH in this study and recommend that daily observations
of elemental and organic carbon be taken in Canadian urban environments
to address one of the major components of ne particulate matter and its
association with human health.
Sources of Particulate Pollution
The four components of the ne fraction that were most statistically sig-
nicantly positively associated with mortality based on lag 1-day data were
sulfate, zinc, nickel, and iron. Fine-fraction Ca, Cu, Sc, Co, Zr, P, La, and Mg
were also fo und to have some association with mortality. The chemical con-
stituents on ne particles are kno wn to be associated with certain sources
(Rahn & Lowenthal, 1985), and so me elements are more source specic than
others are. Sulfur or sulfate is widely known to be from SO
2
emissions, and the
main Canadian sources are coal and oil combustion at power plant s, smelters,
and the oil and gas extraction and rening industry. Many of these sources are
located away from cities, but their S O
2
emissions contribute to sulfate parti-
cles on a regional sca le, which then impacts upon cities. Within cities the main
SO
2
sources, which are generally small relative to the sources just listed, are
power plants, reneries, industry (e.g ., steel manufacturing, pulp and paper),
and transportation. However, not all of these sources, with the exception of
transportation, are found in ea ch of the cities in included in this study.
The sources for the trace elements are not as widely known. Elemental
analysis of ne particles from a wide range of sources, some of which has be en
carried out to supp ort chemical mass balance modeling (Watson et al., 1994),
indicates that zinc emissions are associated with steel product ion, road dust,
possibly from tire wear, and incinerators. In addition, but to a lesser degree ,
zinc is found in the emissions from oil-red power plants and even motor
vehicle emissions. Iro n is also signicantly enriched in the emissions associated
wit h steel production and is prevalent in road dust and soil. Iron can also be
emitted from power plants. One o f the main sources of nickel is oil-red power
plants. To a lesser degree nickel is emit ted from oil re ning, steel production,
and certainly from nickel smelters. Traces of nickel are also emitted from motor
vehicles, and it can be found in road dust and so il. The o ther e lements listed
earlier also have some predominant sources. For example, magnesium and
zirconium are found in coal-red power plant emissions. Copper is found
in ne-particle samples collected near trafc (i.e., roadside sampling ) as well
as from copper smelters, steel production, incineration, and even vegetation
burning. Magnesium is also associated with vegetation burning and even
cooking (e.g., meat), while calcium has a wide variety of sources, although
soil is one of the main ones. Calcium can be also found in emissions from
coal-red power plants and steel production. There is clearly some overlap
in the so urces contributing to the ambient trace elements, which makes it
difcult to point toward specic sources that may be more responsible for
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
36 R. T. BURNETT ET AL.
the ne particles associated with mortality. However, some possible source
distinctions can be hypothesized. Zn may b e indicat ive of a road dust and
possibly a tire wear source and secondarily of incineration. Fe and Zn are
linked to steel production and Ni linked to oil combustion.
Toxicologica l Evidence of Elemental Effects
Our analyses reveal a stronger association between mortality and a subset
of the particle-associated metals, namely, iron, nickel, and zinc, all of which are
known pneumotoxicants (Nemeri, 1990), than to mass alone. Iron, a prototyp-
ical redox cycling metal of toxicological relevance, mediates the toxicity and
carcinogenicity of several types of particles and compounds (Fubini, 1997;
Weinberg, 1999; Ghio et al., 1999). Sulfate plays a role in the conversion
of iron into a bioavailable, catalytically active form, a phenomenon directly
relevant to particulate matter toxicity (Ghio et al., 1999).
The lung toxicity of ambient particulate matter (Adamson et al., 1999;
Vincent et al., 1997a) has also been shown to be related to the presence
of soluble zinc (Adamson et al., 2000). High do ses of zinc chloride aerosols
induce a progressive clinical co urse resembling adult respiratory distress syn-
drome (Evans, 1945; Hjortso et al., 1988). Zinc, iron, and nickel, at the internal
tissue doses pro duced from inhalation of low ambient concentrations of par-
ticles, would not act directly through mechanism s of acute injury. However,
even small amounts of these metals are biologically active, and it is plausible
that the se metals could aggravate an already existing pathology.
The role that ne particulate mass plays in the induction of stress g enes
is related to chemistry of the particles, and in particular to bioavailable tran-
sition metals (Vincent et al., 1997b). Recent studies indicate that variation in
the chemical composition of ambient particles– for example, a decrease in
the bioavailability of transition metals– results in lower potency of particles
in cell cultures, and also correlates with lower rates of morbidity in human
populations (Frampton et al., 1999). Iron from inhaled particles could amplify
inammatory processes through Fe nton-type reactions in the lungs.
Several nickel compounds are toxic and carcinogenic to the lungs (Oller
et al., 1997). The genetic and epigenetic m echanisms of action of nickel at
lo w doses, which may not be acutely toxic to the respiratory tract, could
interfere with gene expression in target cells, resulting in adve rse impacts on
cell regulation pathways critical to lung defense (Tim blin et al., 1998). Zinc
plays a central role in gene regulation and in the activation of metalloproteases
and endopept idases. Deposition of free, ionized zinc (e.g., as zinc sulfate) in
the alveoli could create local cellular loads of the meta l, interfering with cell
regulation pathways (e.g., Ellerbroek & Stack, 1999; Klug, 1999; Palacek et al.,
1999; Samet et al., 1998). Our observations begin to provide some insight
into potential chemica l determinants of the potency of ambient particles.
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
URBAN POLLUTION MIXTURE AND MORTALITY 37
CONCLUSIONS
We conclude that daily variations in urban pollution mixtures in Canadian
cities are statistically associated wit h daily variations in mortality rates. There
are many components of the atmospheric urban mixture that contribute to
its toxic ity, including pollutants in both the p articulate and gas p hase. When
evaluating the p otential population health benets of air po llut ion mitigation
strategies, we urge consideration of the health benets of the predicted reduc-
tion in all pollutants attributed to the propose d strategy and not just a single
pollutant.
REFERENCES
Adamson, I. Y., Prieditis, H., and Vincent, R. 1999. Pulmonary toxicity of an atmosphericparticulate sample
is due to the soluble fraction. Toxicol. Appl. Pharmacol. 157:43–50.
Adamson, I. Y., Prieditis, H., Hedgecock, C., and Vincent, R. 2000. Soluble zinc is the toxic factor in the
lung response to an atmospheric particulate sample. Toxicol. Appl. Pharmacol., in press.
Bates, D. V. 1998. Air pollution: The environmental Phoenix. Can. J. Public Health 89:149–150.
Bates, D. V., and Sizto,R. 1983.Relationshipbetween air pollution levels and hospitaladmissions in southern
Ontario. Can. J. Public Health 74:117–122.
Bates, D. V., and Sizto, R. 1987. Air pollution and hospital admissions in southern Ontario: The acid summer
haze effect. Environ. Res. 43:317–331.
Brook, J. R., Dann, T. F., and Burnett, R. T. 1997. The relationship among TSP, PM
10
, PM
2.5
, and inorganic
constituents of atmospheric particulate matter at multiple Canadian locations. J. Air Waste Manage.
Assoc. 47:2–19.
Burnett R. T., Dales, R. E., Raizenne, M. E., Krewski, D., Summers, P. W., Roberts, G. R., Raad-Yo ung, M.,
Dann, T., and Brook, J. 1994. Effects of low ambient levels of ozone and sulfates on the frequency o f
respiratory admissions to Ontario hospitals. Environ. Res. 65:172–194.
Burnett,R., Dales, R., Krewski, D., Vincent, R., and Brook,J. 1995. Associations between ambient particulate
sulfate and admissions to O ntario hospitals for cardiac and respiratory diseases. Am. J. Epidemiol.
142:15–22.
Burnett, R. T., Cakmak, S., and Krewski, D. 1997a. The role of particulate size and chemistry inthe associ-
ation between summertime ambient air pollution and hospitalization for cardio-respiratory diseases.
Environ. Health Perspect. 105:614–620.
Burnett, R. T., Brook, J. R., Yung, W. T., Dales, R. E., and Krewski, D. 1997b. Association between ozone
and hospitalization for respiratory diseases in 16 Canadian cites. Environ. Res. 72:24–31.
Burnett, R., Dales, R., Brook, J., Raizenne , M., and Krewski, D. 1997c. Association b etween ambient carbon
monoxide levels and hospitalizations for cardiac diseases in ten Canadian cites. Epidemiology 8:162–
167.
Burnett, R. T., Brook, J. E., Philps, O., Cakmak, S., Raizenne, M. E., Stieb, D., Vincent, R., Ozkaynak,H., and
Krewski, D. 1998a. The association between amb ient concent rations of carbon monoxide and daily
mortality in Toronto, Canada. J. Air Waste Manage. Assoc. 48:689–700.
Burnett, R. T., Cakmak, S., and Brook, J. 1998b. The effect of the urban ambient air pollution mix on daily
mortality rates in 11 Canadian cities. Ca n. J. Public Health 89:152–156.
Cleveland, W. S., and Devlin, S. J. 1988. Robust locally-weighted regression and smoothing scat terplots.
J. Am. Stat. Assoc. 74:829–836.
Delno, R. J., Murp hy-Moulton, A. M., Burnett, R. T., Brook, J. R., and Becklake,M. R. 19 96. Effects of ozone
and particulate air pollution on emergency room visits for respirator y illnesses in Montreal. J. Respir.
Crit. Care Med. 155:5 68–576.
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
38 R. T. BURNETT ET AL.
Ellerbroek, S. M., and Stack, M. S. 1999. Membrane associated matrix metalloproteinases in metastasis.
Bioessays 21:940–949.
Evans, E. H. 1945. Casualties following exposure to zinc chloride smoke. Lancet 2:368–370.
Frampton, M. W., Ghio, A. J., Samet, A. J., Carson, J. L., Carter, J . D., and Devlin, R. B. 1999. Effects of
aqueous extracts of PM10 lters from the Utah valley on human airway epithelialcells. Am. J. Physiol.
277:L960–967.
Fubini, B. 1997. Surface reactivity in the pathogenic response to particulates. Environ. Health Perspect.
105(suppl. 5):1013–1020.
Ghio, A. J., Stonehe urner, J., McGee, J. K., and Kinsey, J. S. 1999. Sulfate content correlates with iron
concentrations in ambient air pollution particles. Inhal. Toxicol. 11:293–307.
Goldberg, M. S., Bailar, J. C. III, Burnett, R., Brook, J., Tamblyn, R., Bonvalot,Y., Ernst, P., Flegel , K. M., Singh,
R., and Valois, M.-F. 2000. Identifying subgroups of the general population that may be susceptible
to short-term increases in particulate air pollutio n: A time series study in Mont real, Quebec. Health
Effects Institute, in press.
Groblicki,P. J., Wolff,G. T.,and Countess,R. T. 1981.Visibility-reducing species in the Denver “brown cloud”
I. Relationships between extinction and chemicalcomposition. Atmos. Environ. 15:2473–2484.
Hastie, T., and Tibshirani, R. 1990. Generalized additive models. London: Chapman and Hall.
Hjortso, E., Qvist, J., Bud, M. I., Thomsen, J. L., Andersen, J. B., Wiberg-Jorgensen, F., Jensen, N. K., Jones, R.,
Reid, L. M., and Zapol, W. M. 1988. ARDS after accidental inhalation of zinc chloride smoke. Intens.
Care Med. 14:17–24.
Klug, A. 1999. Zinc nger peptides for the regulation of gene expression. J. Mol. Biol. 293:215–218.
Leech, J. A., Wilby, K., McMullen, E., and Laporte, K. 1996. The Canadian human activity pattern survey:
report of methods and population surveyed. Chron. Dis. Can. 17:118–123.
Nemeri, B. 1990. Metal toxicity and the respiratory tract. Eur. Respir. J. 3:202–219.
Oller, A. R., Costa, M., and Oberdorster, G. 1997. Carcinogenicity assessment of selected nickel com-
pounds. Toxicol. Appl. Pharmacol. 143:152–166.
Palecek, E., Brazdova, M., Cernocka, H., Vlk, D., Brazda, V., a nd Vojtesek, B. 1999. Effect of transition
metals on b inding of p53 protein t o supercoiled DNA and to consensus sequence in DNA fragments.
Oncogene 18:3617–3625.
Priestly, M. B. 1981. Spectral analysis and time series. London: Academic Press.
Rahn, R. A., and Lowenthal, D. H. 1985. Pollution aerosol in the northeast: Northeastern–midwestern
contributions. Science, 228:275–284.
Samet, J. M., Graves, L. M., Quay, J., Dailey, L. A., Devlin, R. B., Ghio, A. J., Wu, W., Bromberg, P. A., and
Reed, W. 1998. Activation of MAPKs in human bronchial epithelial c ells exposed to metals. Am. J.
Physiol. 275:L551–558.
SAS Institute, Inc. 1989. SAS/STAT use r’s guid e, Version 6, 4 th ed., vol. 2. Cary, NC: SAS Institute, Inc.
846 pp.
Schwartz, J., Dockery, D. W., and Neas, L. M. 1996. Is daily mortality asso ciated specically with ne
particles? J. Air Waste Manage. Assoc. 46 :927–939.
Statistical Sciences, Inc. 1993. S-Plus users manual. Seattle WA: S tatistical Sciences, Inc.
Stieb, D. M., Burnet t, R. T., and Beverage, R. C. 1996. Association between ozone and asthma emergency
department visits in Saint John, New Brunswick, Canada. Env iron. Health Perspect. 104:1354–1360.
Thurston, G., Ito, K., and Lippmann, M. 1994. Respiratory hospital admissions and summertime haze air
pollution in Toronto, Ontario: Consideration of the role of acid aerosols. Environ. Res. 65:271–290.
Timblin, C., Berube, K., Churg, A., Driscoll, K., Gordon, T., Hemenway, D., Walsh, E., Cummins, A. B., Vacek,
P., and Mossman, B. 1998. Ambient p articulate matter causes activation of the c-jun kinase/stress-
activated protein kinase cascade and DNA synthesis in lung epithelial cells. Cancer Res. 58:4543–
4547.
U.S. Environmental Protection Agency. 1996. Air Quality Criteria for Particulate Matter. EPA/600/P-95/
001aF, April. Washington, DC: U.S. EPA.
Vincent, R., Bjarnason, S. G., Adamson, I. Y., Hedgecock, C., Kum arathasan, P., Gu´enette, J., Potvin, M.,
Goegan, P., and Bouthillier,L. 1997a. Acute pulmonary toxicity of urban particulate matter and o zone.
Am. J. Pathol. 1 51:1563–1570.
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
URBAN POLLUTION MIXTURE AND MORTALITY 39
Vincent, R., Goegan, P., Johnson, G., Brook, J. R., Kumarathasan,P., Bouthillier, L., and Burnett, R. T. 1997b.
Regulation of pro moter-C AT stress genes in HepG2 cells by suspensions of p articles from ambient air.
Fundam. Appl. Toxicol. 39:1 8–32.
Watson, J. G., Chow, J. C., Lu, Z., Fujita, E. M., Lowenthal, D. H., and Lawson, D. R., 1994. Chemical mass
balance source apportionment of PM10 during the southern California air quality study. Aerosol S ci.
Technol. 21:1–36.
Weinberg, E. D. 1999. The development of awareness of the carcinogenic hazard of inhaled iron. Oncol.
Res. 11:109–113.
Downloaded by [Health Canada / Public Health Agency of Canada] at 10:04 27 May 2016
