Environmental Research 91 (2003) 8–20

Associations between ambient air pollution and daily mortality

among persons with congestive heart failure$
Mark S. Goldberg,a,b,* Richard T. Burnett,c Marie-France Valois,a Kenneth Flegel,a
John C. Bailar III,d Jeffrey Brook,e,f Renaud Vincent,c and Katja Radong
a
Department of Medicine, McGill University, Montreal, Que., Canada
b
Joint Departments of Epidemiology and Biostatistics and Occupational Health, McGill University, Montreal, Que., Canada
c
Environmental Health Directorate, Health Canada, Ottawa, Ont., Canada
d
Department of Health Studies, University of Chicago, Chicago, IL, USA
e
Air Quality Processes Research Division, Meteorological Service of Canada, Environment Canada, Ont., Canada
f
Department of Public Health Sciences, University of Toronto, Toronto, Ont., Canada
g
Institute of Occupational and Environmental Medicine, Ziemssenstr. 1, 80336 Munich, Germany
Received 24 September 2001; received in revised form 22 March 2002; accepted 10 April 2002

Abstract

We conducted a mortality time series study to investigate the association between daily mortality for congestive heart failure
(CHF), and daily concentrations of particles and gaseous pollutants in the ambient air of Montreal, Quebec, during the period
1984–1993. In addition, using data from the universal Quebec Health Insurance Plan, we identified individuals X65 years of age
who, one year before death, had a diagnosis of CHF. Fixed-site air pollution monitors in Montreal provided daily mean levels of
pollutants. We regressed the logarithm of daily counts of mortality on the daily mean levels of each pollutant, after accounting for
seasonal and subseasonal fluctuations in the mortality time series, non-Poisson dispersion, weather variables, and other gaseous and
particle pollutants. Using cause of death information, we did not find any associations between daily mortality for CHF and any air
pollutants. The analyses of CHF defined from the medical record showed positive associations with coefficient of haze, the
extinction coefficient, SO2, and NO2. For example, the mean percent increase in daily mortality for an increase in the coefficient of
haze across the interquartile range was 4.32% (95% CI: 0.95–7.80%) and for NO2 it was 4.08% (95% CI: 0.59–7.68%). These
effects were generally higher in the warm season.
r 2002 Elsevier Science (USA). All rights reserved.

Keywords: Mortality; Time series; Congestive heart failure; Ambient air pollution; Particulates; Nitrogen dioxide

1. Introduction 1992; Schwartz, 1994). It has been suggested that only

persons in poor health are at high risk (Bates, 1992;
There is substantial and consistent evidence for an Frank and Tankersley, 2002; Goldberg, 1996). A
association between increases in levels of ambient air principal hypothesis is that exposure to air pollutants
pollution and daily mortality (Dockery and Schwartz, may cause acute pulmonary disease, such as bronchio-
1995; Goldberg, 1996; Pope III, et al., 1995; Schwartz, litis or pneumonia, thereby leading to congestive heart
failure (CHF) in persons with myocardial damage or
$ cardiac disease (Bates, 1992). Alternatively, exposure to
This study was supported financially through contracts with the
Health Effects Institute, Cambridge, MA, the Toxic Substances
ultrafine particles may invoke alveolar inflammation,
Research Initiative, Health Canada, and the Canadian Institutes for release inflammatory mediators, exacerbate lung condi-
Health Research. Ethics approval was obtained from the Institutional tions, and increase coagulability of blood thereby
Review Board of McGill University and from the Commission de leading to acute episodes of cardiovascular disease
1’acc"es a" 1’information du Qu!ebec. (Seaton et al., 1995). By extension, it can be hypothe-
*Corresponding author. Department of Epidemiology and Biosta-
tistics, McGill University, 1020 Pine Ave. W., Montreal, Que., Canada
sized that persons with CHF may be particularly
H3A 1A3. Fax: +514-398-4503. susceptible to the effects of air pollution (Bates, 1992;
E-mail address: [email protected] (M.S. Goldberg). Kodavanti et al., 1998; Seaton et al., 1995). In the few

0013-9351/03/$ - see front matter r 2002 Elsevier Science (USA). All rights reserved.
PII: S 0 0 1 3 - 9 3 5 1 ( 0 2 ) 0 0 0 2 2 - 1
 M.S. Goldberg et al. / Environmental Research 91 (2003) 8–20 9

studies in which CHF has been investigated, it was the mix of patients that consulted them. As well, the
found generally that daily hospitalizations for CHF QHIP paid the costs of prescriptions for persons aged 65
(Burnett et al., 1997; Morris et al., 1995; Morris and and over. The information on dispensed prescriptions
Naumova, 1998; Poloniecki et al., 1997; Schwartz, 1997) included the date, the name and generic type of
and daily mortality among persons with CHF (Kwon medication, quantity, duration of prescription, and
et al., 2001) increased when levels of ambient particles authorization for refills.
and gaseous pollutants increased. In the present paper, We classified a subject as having CHF before death if
we present a time series investigation to determine (Goldberg et al., 2000): (1) for ages under 65 years, there
whether persons with preexisting CHF are at higher risk were at least two recorded billings with a diagnosis of
of dying when levels of ambient air pollution increase. ICD9 428 and (2) for ages 65 years and over, there were
at least two recorded billings with a diagnosis of ICD9
428 or one billing with ICD9 428 and at least one
2. Materials and methods prescription for diuretics. We used a 1-year time window
for these events to account for broad variations in the
2.1. The study population frequency of physician visits, missing diagnostic infor-
mation on the billing record (about one-half of the
The study took place in Montreal, Quebec, a large billing records have diagnoses listed), and variations in
metropolitan area that experiences relatively low levels how pharmaceuticals are prescribed by physicians.
of air pollution (Goldberg et al., 2000; 2001a, b).
Subjects were residents of Montreal who died in the 2.3. Air pollution and weather data
metropolitan area during the period 1984–1993 and who
were registered with the universal Quebec Health Measurements of sulfur dioxide (SO2), nitrogen
Insurance Plan (QHIP). They were first identified from dioxide (NO2), and carbon monoxide (CO) were made
the computerized provincial database of death certifi- every 2 h at fixed-site monitoring stations in Montreal.
cates and each deceased subject was then linked to the SO2 was measured at 13 stations using ultraviolet
population register of the QHIP. fluorescence and electrical conductivity from changes
in chemical composition of a bromine solution (Philips
2.2. Definitions of CHF 9700 and Monitor Lab 8850); NO2 was measured at
eight stations using chemiluminescence (Thermo Elec-
We conducted separate analyses for subjects (1) who tron 14B); and CO was measured at 12 stations using
had an underlying cause of death of CHF (International infrared absorption (Thermo Electron 48). The coeffi-
Classification of Diseases, Ninth Revision (ICD9), 428 cient of haze (COH), which measures organic and
(World Health Organization, 1980)) or (2) who were inorganic carbon, was also measured every 2 h at several
attributed as having CHF before dying from a natural fixed-site monitoring stations in Montreal. Particle mass
cause (independent of the cause provided on the death (total suspended particles (TSP)), particles having
certificate). For the latter, we made use of diagnoses and aerodynamic diameters of 10 mm and under and 2.5 mm
specific health services rendered by physicians both in and under (PM10, PM2.5, respectively), and sulfate from
and out of hospital as well as filled prescriptions these metrics) was measured at a frequency of every 6
reimbursed by the QHIP. The QHIP provides universal days (Brook et al., 1997a, b). From July 1992 to
coverage for all costs of medical services dispensed in- September 1995, the measurement schedule for PM10
province and it provides complete or partial coverage and PM2.5 increased at one site to daily sampling. We
for services and hospital admissions out-of-province. also made use of measurements of daily total sulfates
Individuals are given unique health insurance plan (1986–1993) from an acid rain monitoring station at
numbers and these are recorded on each medical Sutton, Quebec, a rural community about 150 km
transaction and are used administratively to check the southeast from the city. These data represent back-
validity of claims. Physicians are paid by the QHIP on a ground levels throughout southwest Quebec, including
fee-for-service basis and out-of-province health care is Montreal (Brook et al., 1997b). The average correlation
paid by the patient and reimbursed partially or in full by between sulfates measured at this station and the two
the QHIP. This leads to almost complete reporting to Montreal stations was 0.9.
the QHIP. Laboratory tests and the results from these Visibility, barometric pressure, temperature, total
tests are not part of the database. When submitting each precipitation (distinguishing snow from rain), relative
bill for service, the physician had the option to record humidity, and dew point temperature were measured at
one ICD9 diagnostic code. Unlike the billing data, Dorval International Airport. Visibility was converted
which are examined by the QHIP for consistency and into an extinction coefficient (a measure of light
accuracy, diagnoses were not verified. In addition, most scattering and absorption, due mostly to sulfates) after
physicians used a restricted set of ICD9 codes reflecting accounting for relative humidity (Delfino et al., 1994;
10 M.S. Goldberg et al. / Environmental Research 91 (2003) 8–20

Kinney and Ozkaynak, 1991; Ozkaynak et al., 1985). measure of the deviance) from among various sets of
We used the measurement at noon when there was no models that included different weather variables across
precipitation or the hour closest to noon without lags 0–5 days.
precipitation. Filtered and weather-adjusted single-pollutant models
We also developed statistical models to estimate fine using daily mean values for the concurrent day across
particle mass and sulfates from PM2.5 (hereafter referred the fixed-site monitoring stations were considered. We
to as predicted PM2.5) when measurements during 1986– also estimated mortality with the previous day’s level of
1993 were not taken (Goldberg et al., 2000; 2001a, b). air pollution (lag 1 day) as well as with the average of
COH, the extinction coefficient, and total sulfates from lags of 0–2 days (referred to as the ‘‘3-day mean’’). These
Sutton were used as predictor variables, and the R2 for lag periods were selected a priori in order to investigate
the prediction model was 0.72 for PM2.5 and for sulfate the short-term effects of pollution on mortality. Because
from PM2.5 it was 0.80. pollutants can vary considerably by season, especially
ozone and particles, we also assessed associations
according to a ‘‘cold season’’ (October–March) and a
2.4. Statistical methods ‘‘warm season’’ (April–September). Last, we were
interested in determining whether effects of pollution
For the two definitions of CHF, separate analyses may affect women and men differently and, thus,
were conducted to estimate associations with daily levels separate analyses were conducted according to sex.
of air pollution and day of death. In both sets of Assuming linearity in the response function for the air
analyses, we excluded subjects whose deaths were from pollution variables, we estimated the relative increase in
accidents, poisonings, and injuries (ICD9 X800). We the logarithmic number of daily deaths per unit increase
ignored the underlying cause of death in the analyses in the concentration of the various pollutants. The
that made use of QHIP data to identify persons with percent change in the mean number of daily deaths for
CHF before death. Average daily concentrations of an increase of pollutant equal to its interquartile range
pollutants were used in all analyses, and these were was calculated (referred to as the mean percent change
derived by taking a simple daily average for each in daily mortality (MPC)). Associated upper and lower
monitor and then averaging these to obtain a final daily 95% confidence limits (95% CI) on the mean percent
mean value. We assumed that the daily counts of death change were obtained assuming that the estimated
were distributed approximately as a Poisson variate with regression coefficient was distributed normally, with
constant over- or under-dispersion and we used quasi- the standard error corrected for non-Poisson dispersion.
likelihood estimation to model the logarithm of daily
counts of cause-specific deaths as functions of the
predictor variables. Numerical problems (Dominici 3. Results
et al., 2002; Ramsay et al., 2002) have been found
recently with the use of regression splines or locally There were a total of 133,904 non-accidental deaths
weighted regression smoothers within the generalized during the study period 1984–1993. CHF accounted for
additive models framework (Hastie and Tibshirani, 2.1% of these (2,740 deaths; Table 1), with the majority
1990) to estimate non-parametric smoothings for con- among persons age 65 years and over. All results
tinuous covariates. Instead, we used natural splines to presented subsequently are for this age group only.
model the effects of continuous covariates. Thus, within Using our definition of CHF from the morbidity data,
the context of a generalized linear Poisson model, we we found 14,615 subjects 65 years of age and over who
regressed the natural logarithm of the daily number of had CHF before death (Table 1). Of these subjects, 66%
deaths on the natural spline for day-of-study, thus were coded as having a circulatory underlying cause of
providing an adjustment for seasonal and subseasonal death, and only 6.3% were attributed as dying from
variations (temporal filter), on a dummy variable to CHF. Among those subjects whose underlying cause of
account for annual trends in daily mortality, and on death was given as CHF, only 37% were classified from
natural splines of potential confounding effects of the QHIP data as having CHF before death.
relevant weather variables. For each analysis, we Figs. 1 and 2 and Table 1 show that there was little
selected the temporal filter having the number of knots overdispersion in either of the CHF mortality time
that produced a filtered time series that was consistent series. Table 1 also shows the distributions of pollutants
with a white noise process, using Bartlett’s statistic and weather variables; levels of pollutants in Montreal
(Priestly, 1981). This produced a residual time series that are fairly low with respect to other North American and
had the least amount of serial autocorrelation. We then European cities.
sought the combination of weather variables that For the analyses of CHF defined using the underlying
yielded a minimum value of the Aikake Information cause of death, we used a natural spline having 71
Criterion (Hastie and Tibshirani, 1990) (a penalized degrees of freedom to adjust for temporal trends and we
Table 1
Distribution of essential variables, Montreal, 1984–1993

Units Number of days of Mean Standard Minimum Percentiles Inter-

measurements deviation quartile
25th 50th 75th 100th range

Congestive heart failure

Coded as underlying
cause of death
X65 years 2528 1795 0.7 0.85 0 0 0 1 7 1.0
Total 2740 1884 0.75 0.89 0 0 1 1 7 1.0
Morbidity 1 year before
death
X65 years 14615 3552 4.0 2.19 0 2 4 5 15 3.0
Total 16794 2592 4.6 2.35 0 3 4 6 16 3.0
TSP mg/m3 603 53.1 22.6 14.6 37.0 48.7 65.6 211.1 28.57
PM10 mg/m3 624 32.2 17.6 6.5 19.7 28.5 41.1 120.5 21.32
PM2.5 mg/m3 636 17.4 11.4 2.2 9.4 14.7 21.9 72.0 12.51
Sulfate from PM10 mg/m3 437 4.7 4.4 0.3 1.9 3.6 5.7 30.7 3.84
Sulfate from PM2.5 mg/m3 446 4.3 4.2 0.2 1.6 3.1 5.1 29.2 3.51
Sulfate from TSP mg/m3 607 4.3 2.9 0.3 2.3 3.6 5.3 19.2 3.02
Sulfate from the Sutton mg/m3 2680 3.3 3.6 0 1.3 2.2 3.8 30.0 2.50
monitorino Stationa
COH 0.1 COH 3653 2.4 1.5 0.1 1.3 2.1 3.2 15.6 1.85
Units per
327.8 linear
meters
Predicted PM2.5a mg/m3 3653 17.6 8.8 4.6 11.5 15.4 21.0 71.7 9.5
Predicted Sulfate from mg/m3 3653 4.1 3.6 0.02 1.9 3.1 4.8 30.1 2.9
PM2a
Extinction km 1 3454 0.15 0.10 0.01 0.06 0.15 0.17 1.87 0.11
SO2 mg/m3 3653 17.8 11.2 3.9 10.3 14.6 21.8 105.7 11.50
NO2 mg/m3 3653 41.7 15.4 8.8 30.9 39.5 50.2 143.5 19.34
CO ppm 3653 0.8 0.5 0.1 0.5 0.7 1.0 5.1 0.50
M.S. Goldberg et al. / Environmental Research 91 (2003) 8–20

O3 mg/m3 3653 29.0 17.1 2.8 16.6 26.0 37.9 163.9 21.34
Mean temperature 1C 3653 6.4 11.8 27.3 2.6 7.5 16.5 28.8 19.10
Change in maximum 1C 3653 0.0 4.7 25.0 2.5 0.4 2.8 19.4 5.30
temperature from
previous day
Mean dew point 1C 3653 1.2 11.6 33.8 6.6 2.0 10.8 23.1 17.40
temperature
Relative humidity % 3653 69.4 11.9 33.0 61.0 70.0 78.0 99.0 17.00
Change in barometric kPa 3653 0.0 0.9 4.2 0.5 0.0 0.6 4.4 1.10
pressure from previous
day
a
For the period 1986–1993.
11
12 M.S. Goldberg et al. / Environmental Research 91 (2003) 8–20

found that a natural spline on four degrees of freedom

for mean temperature and a linear term for change in
Number of deaths per day

6 barometric pressure from the previous day, both

evaluated at lag 0 days, minimized the AIC. We did
not find any statistically significant associations for any
4
of the pollutants evaluated at any of the three lag
periods considered (Table 2).
2 Among persons classified as having CHF 1 year
before death using the QHIP data, we used a natural
spline having 61 degrees of freedom to adjust for
0 temporal trends and we found that a natural spline on 3
0 1000 2000 3000 4000 degrees of freedom for mean temperature and a natural
Days from January 1, 1984 spline on two degrees of freedom for change in
Fig. 1. Scatterplot of daily number of nonaccidental deaths among barometric pressure from the previous day, both
persons age 65 years and over who were classified as dying from CHF. evaluated at lag 0 days, minimized the AIC. Daily
The solid line is the locally weighted regression smoother (LOESS) mortality was significantly increased at all three lag
smooth representing the long-term trend in the data (span of 50% of periods for COH and for sulfur dioxide (Table 3). We
the data). The total number of days in the time series is 3653.
also found that nitrogen dioxide was associated
positively with CHF at lag 1 day and at the 3-day
15.0 mean. At the 3-day mean, the MPCs across all
Number of deaths per day

pollutants varied between 0.56% and 4.67%.

12.5
Figs. 3 and 4 show the results of the analyses for
10.0 the 3-day mean, adjusted simultaneously for other
pollutants (CO, NO2, SO2, ozone, and the COH
7.5 (a marker for particle pollution)). For mortality
5.0 from CHF (Fig. 3), the adjustments did not alter
the findings that were reported in Table 3. For pre-
2.5 existing CHF derived from the QHIP data (Fig. 4),
0.0 except for ozone, all of these adjustments attenuated
0 1000 2000 3000 the MPCs and generally increased the confidence
Days from January 1, 1984 intervals.
Fig. 5 shows separate analyses of preexisting CHF for
Fig. 2. Scatterplot of daily number of non-accidental deaths among
persons age 65 years and over who were classified as having CHF
the cold and warm seasons. The effects were higher in
before death. The solid line is the LOESS smooth representing the the warm season for all pollutants except the two
long-term trend in the data (span of 50% of the data). The total measures of sulfates, and COH and SO2 also displayed
number of days in the time series is 3653. significant positive effects in the cold season. Fig. 6

Table 2
Summary estimates of the mean percent change in daily mortality from CHF among persons 65 years of age and over across the interquartile range
of mean daily concentrations of selected particulate and gaseous pollutants, Montreal, 1984–1993a

Lag 0 days Lag 1 day Three-day mean

Pollutant Mean percent 95% CI Mean percent 95% CI Mean percent 95% CI
changeb changeb changeb

COH 0.03 5.20 to 5.43 1.06 6.37 to 4.56 1.93 9.22 to 5.94
Extinction 0.76 3.78 to 5.51 1.87 6.62 to 3.13 1.14 7.82 to 6.02
Predicted PM2.5b 1.54 3.90 to 7.29 0.76 6.84 to 5.71 2.00 5.58 to 10.18
Sulfate from the Sutton 0.29 3.18 to 3.88 0.08 4.09 to 4.10 1.73 3.48 to 7.23
monitoring stationb
Predicted Sulfate from 0.87 3.22 to 5.13 1.01 5.72 to 3.93 2.08 3.84 to 8.37
PM2.5b
SO2 0.92 4.84 to 7.04 0.05 5.62 to 5.86 0.05 7.45 to 8.16
NO2 1.04 4.55 to 6.96 0.94 4.96 to 7.20 0.10 7.58 to 8.43
CO 0.99 6.31 to 4.63 0.12 5.29 to 5.84 1.38 8.81 to 6.66
O3 0.21 6.66 to 6.69 5.27 2.49 to 13.64 4.54 5.64 to 15.81
a
Adjusted for temporal effects, calendar year, mean temperature, change in barometric pressure from the previous 24 h, and an interaction term
between temperature and barometric pressure.
b
For the period 1986–1993.
 M.S. Goldberg et al. / Environmental Research 91 (2003) 8–20 13

Table 3
Summary estimates of the mean percent change in daily mortality among persons classified as having CHF before death, 65 years of age and over,
across the interquartile range of mean daily concentrations of selected participate and ‘gaseous pollutants, Montreal, 1984–1993a

Pollutant Lag 0 days Lag 1 day Three-day mean

Mean percent 95% CI Mean percent 95% CI Mean percent 95% CI

changeb changeb changeb

COH 2.78 0.48 to 5.13 3.72 1.36 to 6.14 4.32 0.95 to 7.80
Extinction 2.36 0.42 to 4.34 1.20 0.69 to 3.13 2.08 0.67 to 4.92
Predicted PM2.5b 2.53 0.04 to 5.18 1.69 0.79 to 4.23 2.15 1.03 to 5.43
Sulfate from the Sutton 0.66 1.03 to 2.39 0.47 1.08 to 2.04 0.56 1.53 to 2.69
monitoring stationb
Predicted Sulfate from 1.61 0.39 to 3.64 0.68 1.14 to 2.54 1.15 1.23 to 3.58
PM2.5b
SO2 3.30 0.86 to 5.79 3.28 0.83 to 5.80 4.67 1.30 to 8.14
NO2 2.37 0.11 to 4.92 3.25 0.66 to 5.91 4.08 0.59 to 7.68
CO 2.10 0.24 to 4.49 2.28 0.09 to 4.72 2.86 0.46 to 6.29
O3 2.13 1.14 to 5.50 0.08 2.93 to 2.85 2.34 1.78 to 6.63
a
Adjusted for temporal effects, calendar year, mean temperature, change in barometric pressure from the previous 24 h, and an interaction term
between temperature and barometric pressure.
b
For the period 1986–1993.

20

15
Unadjusted
Mean Percent Change in Daily Mortality

Adjusted
10

-5

-10

-15

-20 COH Extinction Predicted Sutton Predicted SO2 NO2 CO O3

PM2.5 Sulfate Sulfate
from PM2.5

Fig. 3. Mean percent change in daily mortality among subjects age 65 years and over who died from CHF for increases of the interquartile range of
selected pollutants, adjusted for the variables listed in footnote (a) of Table 2 and other pollutants. All pollutants were evaluated at the 3-day mean.
The estimated mean percent change in daily cause-specific mortality across the interquartile range is shown by the horizontal bars within the vertical
lines (95% confidence intervals). The statistical adjustments for the particle metrics included terms for the other gaseous pollutants (CO, SO2, NO2,
O3); these gaseous pollutants were adjusted for each other and for the COH.

shows the analyses for CHF morbidity stratified by sex. 4. Discussion

For all pollutants but ozone, larger effects were found in
men than in women, although the differences were not Among those persons whose underlying cause of
significant statistically. death was CHF, we found no statistically significant
14 M.S. Goldberg et al. / Environmental Research 91 (2003) 8–20

Unadjusted
10
Adjusted
9
8

Mean Percent Change in Daily Mortality 7

6
5
4
3
2
1
0
-1
-2
-3
-4
-5
-6
COH Extinction Predicted Sutton Predicted SO2 NO2 CO O3
PM2.5 Sulfate Sulfate
from PM2.5

Fig. 4. Mean percent change in daily mortality among subjects 65 years and over who were classified as having CHF before death, adjusted for the
variables listed in footnote (a) of Table 2 and for other pollutants. All pollutants were evaluated at the 3-day mean. The estimated mean percent
change in daily cause-specific mortality across the interquartile range is shown by the horizontal bars within the vertical lines (95% confidence
intervals). The statistical adjustments for the particle metrics included terms for the other gaseous pollutants (CO, SO2, NO2, O3); these gaseous
pollutants were adjusted for each other and for the COH.

25

20
Mean Percent Change in Daily Mortality

Warm Cold
15

10

-5

-10 COH Extinction Predicted Sutton Predicted SO2 NO2 CO O3

PM2.5 Sulfate Sulfate
from PM2.5

Fig. 5. Mean percent change in daily mortality among subjects 65 years and over who were classified as having CHF before death, according to
season, adjusted for the variables listed in footnote (a) of Table 2. The ‘‘warm season’’ includes the months April–September. All pollutants were
evaluated at the 3-day mean. The estimated mean percent change in daily cause-specific mortality across the interquartile range is shown by the
horizontal bars within the vertical lines (95%, confidence intervals).
 M.S. Goldberg et al. / Environmental Research 91 (2003) 8–20 15

15.0

12.5 Men Women

Mean Percent Change in Daily Mortality

10.0

7.5

5.0

2.5

0.0

-2.5

-5.0

COH Extinction Predicted Sutton Predicted SO2 NO2 CO O3

-7.5 PM2.5 Sulfate Sulfate
from PM2.5

Fig. 6. Mean percent change in daily mortality among subjects 65 years and over classified as having CHF before death, according to gender,
adjusted for the variables listed in footnote (a) of Table 2. All Pollutants were evaluated at the three-day mean. The estimated mean percent change in
daily cause-specific mortality across the interquartile range is shown by the horizontal bars within the vertical lines (95% confidence intervals).

associations with any of the pollutants. Among persons (Kwon et al., 2001) (Table 4). Associations between
who were classified as having CHF within 1 year of daily hospitalizations for CHF and most pollutants
death, we found that daily mortality increased in a linear (CO, NO2, SO2, ozone, PM10) were found in three
fashion for COH, the extinction coefficient, SO2, and American studies (Morris et al., 1995; Morris and
NO2, and that these associations were more pronounced Naumova, 1998; Schwartz and Morris, 1995). Another
in the warm season. We did not find any associations study of hospital admissions (Burnett et al., 1997) in the
with specific measures of fine particles or sulfates. 10 largest cities in Canada revealed positive associations
There are no data reporting the accuracy of under- with CO, NO2, SO2, and COH. In one study, an
lying causes of death in Quebec, but in other jurisdic- interaction between temperature and CO was found
tions, it has been found that the accuracy of coding (Morris and Naumova, 1998). No associations were
varies with cause of death (Alderson and Meade, 1967; found between daily hospitalization and air pollution in
de Faire et al., 1976; Engel et al., 1980; Percy et al., one study conducted in London, England (Cendon et al.,
1981). Site of cancer is usually coded reasonably 1997). On the other hand, in a follow-up of a cohort of
accurately (above 80%), but respiratory and cardiovas- persons hospitalized for CHF, Kwon et al. (2001) found
cular diseases are often confused; in particular, when results similar to ours and, in particular, observed
persons have both conditions concurrently and both stronger associations with daily mortality in the CHF
contributed to death, there may be some uncertainty cohort than among the general population.
about which cause should be selected as the primary Our findings need to be interpreted in light of the
underlying cause. In other instances, there may be errors assumptions, strengths, and limitations of the study
in selecting one underlying cause in a complex chain of design. There tends to be a high degree of correlation
health events. Given these facts, it is therefore not among day-to-day changes in the concentrations of air
surprising that we found few individuals who were pollutants (gases and particles) in urban air. This is due
diagnosed with CHF, and our negative findings are to similarities in the sources, atmospheric chemical
likely a result of limited statistical power and misclassi- processes, and influences of weather among urban
fication. pollutants. These considerations suggest that adjusting
There are a few time series studies of hospitalization one pollutant for another may not be warranted. In
for CHF (Burnett et al., 1997; Morris et al., 1995; addition, in a recent personal monitoring study, Sarnat
Morris and Naumova, 1998; Poloniecki et al., 1997; and collaborators (2001) concluded that such adjust-
Schwartz and Morris, 1995) and mortality from CHF ments for PM2.5, in particular, are not justified because
 16

Table 4
Summary of evidence of associations between hospitalization or mortality and increases in ambient air pollution

Reference and Place, dates, Outcome measure Mean number of Pollutant Meaner median Increase in levels Relative increase 95% Confidence
publication year population events per day concentration of pollutants used in daily number of interval or t-ratio
to calculate events per metric
relative increases
in daily number of
events (assumes a
linear response
function)

Schwartz and Morris Detroit, 1986– Hospital 26.6 PM10 48.0 Per IQR: 32[mg/ 1.032 1.012–1.052
(1995) 1989, X65 years admissions for mJ mean of
heart failure (ICD pollutant at lags 0
9428) and 1 days
Ozone 41.0 Per IQR: 28 ppb, 1.022 0.997–1.048
no lag
SO2 25.4 Per IQR: 18 ppb, 1.002 0.978–1.017
no lag
CO 2.38 Per IQR: 1.022 1.011–1.033
1.28 ppm, no lag
Morris et al. (1995) Chicago Primary hospital 33.7 CO 1.8–5.6 Per 10 ppm 1.22a 1.14–1.31
admissions for
heart failure
(ICD9 428)
Detroit 16.3 NO2 0.04–0.08 Per 0.1 ppm 1.10a 1.06–1.15
Houston 9.3 SO2 0.10–0.32 Per 0.05 ppm 1.05a 0.99–1.11
Los Angeles 39.5 Ozone 0.039–t.075 Per 0.12 ppm 1.00a 0.95–1.06
Milwaukee 7.4 (range of means
across cities)
New York 33.0
Philadelphia 17.0
1986–1989, X65
M.S. Goldberg et al. / Environmental Research 91 (2003) 8–20

years

Poloniecki et al. London, 1987– Hospital 24.6 Ozone 13 Per 1 ppb lagged 1 0.99 0.96–1.03
(1997) 1994 admissions for day
heart failure
(ICD9 428)
NO2 35 Per 1 ppb lagged 1 1.00 0.98–1.02
day
SO2 6 Per 1 ppb lagged 1 1.01 0.99–1.03
day
CO 0.9 Per 1 ppm lagged 1.01 0.99–1.03
1 day
Black smoke 12 Per 1 mg/m3 1.01 0.99–11.03
lagged 1 day
Burnett et al. (1997) 10 Canadian Hospital 39 CO 1.59 Per 8-h average 1.04 5.1
citiesb, 1981–1991, admissions for ppm; no lag
X65 years congestive heart
failure (ICD9428)
NO2 23 Per 23 ppb; no lag 1.04 3.0
COH 0.3 Per 0.3  103 1.00 3.6
linear feet; no lag
SO2 4.7 Per 4.7 ppb; no lag 1.02 2.4
Ozone 16 Per 16 ppb; no lag 0.97 1.6
Morris and Naumova Chicago, 1986– Primary hospital 34 CO 2.51 Per 3.054 ppm 1.09 1.06–1.12
(1998) 1989, X65 years admissions for 75th percentile
heart failure
(ICD9428)
NO2 0.044 Per 0.053 ppm 1.04 1.01–1.06
75th percentile
SO2 0.025 Per 0.030 ppm 1.09 1.05–1.13
75th percentile
Ozone 0.039 Per 0.051 ppm 1.03 0.99–l.07
75th percentile
PM10 41.0 Per 51.0 ppm 75th 1.04 1.01–1.07
percentile
Kwon et al. (2001) Seoul, Korea, all Daily mortality 1,807 deaths total CO 0.11 ppm Per 0.59 ppm 1.054 0.991–1.121
ages, cohort of from 1994–1998 (IQR)
persons with a of cohort
primary hospital
admission for
heart failure
(ICD9 428), 1994–
996
NO2 30.6 ppb Per 14.6 ppb 1.065 0.995–1.139
(IQR)
SO2 10.8 ppb Per 9.9 ppb (IQR) 1.070 0.997–1.147
Ozone 28.2 ppb Per 20.5 ppb 1.034 0.966–1.108
(IQR)
PM10 63.7 mg/m3 Per 42.1 mg/m3 1.058 0.989–1.131
(IQR)
M.S. Goldberg et al. / Environmental Research 91 (2003) 8–20

Note: Lag x refers to evaluating the association using values of exposure x days before the event. PM10, particles having an aerodynamic diameter of 10 mm; COH, coefficient of haze (a measure of
ambient carbon); CO, carbon monoxide; SO2, sulfur dioxide; NO2, nitrogen dioxide; ppm(b), parts per million (billion); IQR, interquartile range.
a
Estimated from a weighted meta-regression analysis across the estimates in each of the seven cities.
b
Montreal, Ottawa, Toronto, Hamilton, London, Windsor, Winnipeg, Edmonton, Calgary.
17
18 M.S. Goldberg et al. / Environmental Research 91 (2003) 8–20

of the lack of correlations between ambient and 24-h affect the heart indirectly through a modification of
integrated personal exposures to air pollutants. How- pulmonary endothelin homeostasis. This is based partly
ever, this conclusion may not apply to time series studies on the findings that human subjects exposed to ambient
in which the exposure metrics are concentrations of PM2.5 exhibit dose-dependent increases of levels of
ambient air pollutants. We thus believe that the multi- circulating endothelins (Vincent et al., unpublished
pollutant analyses that accounted for the joint effects of observation). Endothelins are among the most potent
NO2, SO2, CO3 and ambient particles needs to be vasoconstrictors known. Pulmonary release of endothe-
interpreted cautiously, as these adjustments may not, in lin-1 contributes to an elevation of plasma endothelin-1
fact, represent unbiased estimates of effect if these and to vasoconstriction (Kjekshus et al., 2000) and
pollutants are not true confounding variables. Thus, plasma endothelin levels correlate with severity of
these multipollutant models should be considered as disease in CHF and also predict cardiac death (Gala-
sensitivity analyses that may represent a plausible range tius-Jensen et al., 1996). Likewise, reduction of en-
of health effects. dothelin levels in CHF is associated with improvement
We could not control for the effects of infectious of symptoms (Tsumamoto et al., 1995). Endothelins
disease epidemics (e.g., influenza, which occurs mostly in appear to be involved directly in the pathogenesis of
the fall and winter when particle levels are increased) cardiovascular diseases through their vasopressor and
because there are no databases that could be used for mitogenic mechanisms, and endothelin receptor antago-
this purpose. These epidemics occur mostly in the winter nists are now being introduced in the treatment of CHF
months and the effects of ozone is mostly in the warm to reduce peripheral vascular and pulmonary resistance
months of the year, so that these epidemics should not and to increase cardiac output (Giannessi et al., 2001;
have confounded the warm-season-specific estimates. As Spieker et al., 2001). Elevated circulating endothelins
well, the temporal filter should have adjusted for may also affect cardiac arrhythmia and dysrhythmia,
epidemics lasting weeks or months. enhance myocardial ischemia and promote infarct
The rationale for our judgments as to how CHF extension, and contribute to increased systemic and
should be defined was based on our knowledge of pulmonary vascular resistance, vascular dysfunction,
clinical practice in Quebec, the lack of standardization and renal impairment in diabetes and chronic heart
of medical practice, and the information required to failure patients (Best and Lerman, 2000; Spieker et al.,
provide an adequately accurate diagnosis. For example, 2001; Zouridakis et al., 2001).
the requirement of having two sequential diagnoses on
the billing record to define CHF was based on usual
clinical practice guidelines (e.g., Moe and Armstrong,
1988). Although these diagnoses are surrogates for Acknowledgments
actual positive tests, the universality of the drug plan in
the elderly almost guarantees that we have the potential We thank the Montreal urban community and
to identify all treated cases. During the period 1984– Environment Canada for providing the NAPS, CAP-
1993, diuretics were the drug of choice for most cases of MoN, CAAMP, and meteorological data and we are
CHF, although other pharmaceuticals were used, grateful to the Ministe" re de la sante! et des services
notably digoxin and angiotensin-converting-enzyme sociaux de Que! bec for providing the health data. The
(ACE) inhibitors. We believe that the use of a authors gratefully acknowledge the assistance of Rose
combination of diagnostic codes and filled prescriptions Dugandzic, Jacques Barry, Holly Lam, Marie-Claude
for diuretics increased accuracy, although there would Boivin, David Johnson, and Claude Gagnon. Mark S.
be some misclassification involved. Another strength of Goldberg gratefully acknowledges receipt of an Inves-
this methodology is that we made use of rendered tigator Award from the Canadian Institutes for Health
services carried out both within and without hospital, so Research and support from the Fonds de la recherche en
that they should in principal perform more accurately sante! du Que! bec.
than simple hospital discharge summaries which pro-
vide, at best, a snapshot of acute illnesses or exacerba-
tions of chronic ones and cause-of-death information, References
which is notoriously inaccurate for cardiovascular and
respiratory diseases. Alderson, M.R., Meade, T.W., 1967. Accuracy of diagnosis on death
The positive association found for COH, a marker for certificates compared with that in hospital records. Br. J. Prev.
carbon particles, is consistent with previous observa- Soc. Med. 21, 22–29.
Bates, D.V., 1992. Health indices of the adverse effects of air pollution:
tions that inhaled urban particles can cause rapid and
the question of coherence. Environ. Res. 59, 336–349.
sustained elevations of levels of circulating plasma Best, P.J., Lerman, A., 2000. Endothelin in cardiovascular disease:
endothelin-1 and -3 (Bouthillier et al., 1998; Vincent from atherosclerosis to heart failure. J. Cardiovasc. Pharmacol.
et al., 2001). We hypothesize that ambient particles can 35, S61–S63.
 M.S. Goldberg et al. / Environmental Research 91 (2003) 8–20 19

Bouthillier, L., Vincent, R., Goegan, P., Adamson, I.Y.R., Bjarnason, ambient air particle pollution in Montreal, Quebec: 2. Cause-
S.G., Stewart, M., Guenette, J., Potvin, M., Kumarathasan, P., specific mortality. Environ. Res. A 86, 26–36.
1998. Acute effects of inhaled urban particles and ozone: Lung Hastie, A.T., Tibshirani, R., 1990. Generalized Additive Models.
morphology, macrophage activity and plasma endothelin-1. Am. Chapman & Hall, London.
J. Pathol. 153, 1873–1884. Kinney, P.L., Ozkaynak, H., 1991. Associations of daily mortality and
Brook, J.R., Dann, T.F., Burnett, R.T., 1997a. The relationship air pollution in Los Angeles County. Environ. Res. 54, 99–120.
among TSP, PM10, PM2.5, and inorganic constituents of atmo- Kjekshus, H., Smiseth, O.A., Klinge, R., Oie, E., Hystad, M.E.,
spheric particulate matter at multiple Canadian locations. J. A. Attramadal, H., 2000. Regulation of ET: pulmonary release of ET
Waste Manage. Assoc. 46, 2–19. contributes to increased plasma ET levels and vasoconstriction in
Brook, J.R., Wiebe, A.W., Woodhouse, S.W., Audette, C.V., Dann, CHF. Am. J. Physiol. Heart. Circ. Physiol. 278, H1299–H1310.
T.F., Callaghan, S., Piechowski, M., Dabek-Zlotorzynska, E., Kodavanti, U.P., Costa, D.L., Bromberg, P.A., 1998. Rodent models
Dloughy, J.F., 1997b. Fine particle strong acidity, sulphate, PM10, of cardiopulmonary disease: Their potential applicability in studies
PM2.5 and related gaseous species observed at multiple locations of air pollutant susceptibility. Environ. Health Perspect. 106
in Canada: concentrations and spatial-temporal relationships. (Suppl. 1), 111–130.
Atmos. Environ. 31, 4223–4236. Kwon, H.-J., Cho, S.-H., Nyberg, P., Pershagen, G., 2001. Effects of
Burnett, R.T., Dales, R.E., Brook, J.R., Raizenne, M.E., Krewski, D., ambient air pollution on daily mortality in a cohort of patients with
1997. Association between ambient carbon monoxide levels and congestive heart failure. Epidemiology 12, 413–419.
hospitalization for congestive heart failure in the elderly in 10 Moe, G.W., Armstrong, P.W., 1988. Congestive heart failure.
Canadian cities. Epidemiology 8, 162–167. Canadian Medical Association Journal 138, 689–694.
Cendon, S.P., Battlehner, C., Lorenzi, F.G., Dohlnikoff, M., Pereira, Morris, R.D., Naumova, E.N., 1998. Carbon monoxide and
P.M., Conceicao, G.M., Beppu, O.S., Saldiva, P.H., 1997. hospital admissions for congestive heart failure: evidence of an
Pulmonary emphysema induced by passive smoking: an experi- increased effect at low temperatures. Environ. Health Perspect.
mental study in rats. Brazilian J. Med. Biol. Res. 30, 1241–1247. 106, 649–653.
de Faire, U., Friberg, L., Lorich, U., Lundman, T., 1976. A validation Morris, R.D., Naumova, E.N., Munasinghe, R.L., 1995. Ambient air
of cause-of-death certification in 1,156 deaths. Acta. Med. Scand. pollution and hospitalization for congestive heart failure among
20, 223–228. elderly people in seven large US cities. Am. J. Public Health 85,
Delfino, R.J., Becklake, M.R., Hanley, J.A., Singh, B., 1994. 1361–1365.
Estimation of unmeasured particulate air pollution data for an Ozkaynak, H., Schatz, A.D., Thurston, G.D., Isaacs, R.G., Husar,
epidemiological study of daily respiratory morbidity. Environ. R.B., 1985. Relationships between aerosol extinction coefficients
Res. 67, 20–38. derived from airport visual range observations and alternative
Dockery, D.W., Schwartz, J., 1995. Particulate air pollution measures of airborne particle mass. J. Air Pollut. Control Assoc.
and mortality: more than the Philadelphia story. Epidemiology 6, 35, 1176–1185.
629–632. Percy, C., Stanek III, E., Gloeckler, L., 1981. Accuracy of cancer death
Dominici, F., McDermott, A., Zeger, S.L., Samet, J.M., 2002. On the certificates and its effect on cancer mortality statistics. Am J.
use of generalized additive models in time-series studies of air Public Health 71, 242–250.
pollution and health. Am. J. Epidemiol. 156, 193–203. Poloniecki, J.D., Atkinson, R.W., Ponce de Leon, A., Anderson, H.R.,
Engel, L.W., Strauchen, J.A., Chiazze Jr., L., Heid, M., 1980. 1997. Daily time series for cardiovascular hospital admissions
Accuracy of death certification in an autopsied population with and previous day’s air pollution in London, UK. Occup. Environ.
specific attention to malignant neoplasms and vascular diseases. Med. 54, 535–540.
Am. J. Epidemiol. 111, 99–112. PopeIII, C.A., Dockery, D.W., Schwartz, J., 1995. Review of
Frank, R., Tankersley, C.G., 2002. Air pollution and daily mortality: a epidemiological evidence of health effects of particulate air
hypothesis concerning the role of impaired homeostasis. Environ. pollution. Inhal. Toxicol. 7, 1–18.
Health Perspect. 110, 61–65. Priestly, M.B., 1981. Spectral Analysis of Time Series. Academic
Galatius-Jensen, S., Wroblewski, H., Emmluth, C., Bie, P., Haunso, S., Press, San Diego.
Kastrup, J., 1996. Plasma endothelin in congestive heart failure: a Ramsay, T.O., Burnett, R.T., Krewski, D. (2002). The effect of
predictor of cardiac death? J. Card. Fail 2, 71–76. concurvity in generalized additive models linking mortality to
Giannessi, D., Vitale, R.L., Del Ry, S., 2001. The role of endo- ambient particulate matter, Epidemiology, in press.
thelins and their receptors in heart failure. Pharmacol. Res. 43, Sarnat, J.A., Schwartz, J., Catalano, P.J., Suh, H.H., 2001. Gaseous
111–126. pollutants in particulate matter epidemiology: confounders or
Goldberg, M.S., 1996. Particulate air pollution and daily mortality: surrogates. Environ. Health Perspect. 109, 1053–1061.
who is at risk. J. Aerosol. Med. 9, 43–53. Schwartz, J., 1992. Air pollution and daily mortality: a synthesis.
Goldberg, M.S., Bailar III, J.C., Burnett, R.T., Brook, J.R., Tamblyn, Public Health Rev. 19, 39–60.
R., Bonvalot, Y., Ernst, P., Flegel, K.M., Singh, R., Valois, M. F. Schwartz, J., 1994. Air pollution and daily mortality: a review and
(2000). Identifying Subgroups of the General Population That May meta analysis. Environ. Res. 64, 36–52.
Be Susceptible to Short-Term Increases in Particulate Air Pollu- Schwartz, J., 1997. Air pollution and hospital admissions for
tion: a Time Series Study in Montreal, Quebec. Health Effects cardiovascular disease in Tucson. Epidemiology. 8, 371–377.
Institute, Cambridge, MA. Available at http://www.healtheffects. Schwartz, J., Morris, R., 1995. Air pollution and hospital admissions
org. for cardiovascular disease in Detroit, Michigan. Am J. Epidemiol.
Goldberg, M.S., Burnett, R., Bailar III, J.C., Brook, J., Bonvalot, Y., 142, 23–35.
Tamblyn, R., Singh, R., Valois, M-F., 2001. The association Seaton, A., MacNee, W., Donaldson, K., Godden, D., 1995.
between daily mortality and short-term effects of ambient air Particulate air pollution and acute health effects. Lancet 345,
particle pollution in Montreal, Quebec: 1. Nonaccidental mortality. 176–178.
Environ. Res. 86, 12–25. Spieker, L.E., Noll, G., Ruschitzka, F.T., Luscher, T.F., 2001.
Goldberg, M.S., Burnett, R., Bailar III, J.C., Brook, J., Bonvalot, Y., Endothelin receptor antagonists in congestive heart failure: a
Tamblyn, R., Singh, R., Valois, M.-F., Vincent, R., 2001b. The new therapeutic principle for the future? J. Am. Coll. Cardiol. 37,
association between daily mortality and short-term effects of 1493–1505.
20 M.S. Goldberg et al. / Environmental Research 91 (2003) 8–20

Tsumamoto, T., Hisanaga, T., Fukai, D., Wada, A., Maeda, K., Urban Particles in Wistar Rats. Health Effects Institute, Cam-
Maeda, Y., Kinoshita, M., 1995. Prognostic value of plasma bridge, MA. Research Reports of the Health Effects Institute.
soluble intercellular adhesion molecule-1 and endothelin-1 con- World Health Organization. (1980). International Classification of
centration in patients with chronic congestive heart failure. Am. Diseases, Ninth Revision, Clinical Modification. US Department
J. Cardiol. 76, 803–808. of Health and Human Services.
Vincent, R.P., Kumarathasan, P., Goegan, S.G, Bjarnason, J., Zouridakis, E.G., Schwartzman, R., Garcia-Moll, X., Cox, I.D.,
Gu!enette, D., B!erub!e, I.Y., Adamson, S., Desjardins, K., Burnett, Fredericks, S., Holr, D.W., Kaski, J.C., 2001. Increased plasma
R.T., Miller, F.J., Battistini, B. (2001). Inhalation Toxicology of endothelin levels in angina patients with rapid coronary artery
Ambient Articulate Matter: Acute Cardiovascular Effects of disease progression. Eur. Heart J. 22, 1578–1584.