There have been several
studies on the short-term
effects of air pollution
on hospital admissions
(Burnett et al. 1997a,
1997b; Le Tertre et al.
2002; Pope 2000; Samet
et al. 2000), but most
have examined single
cities. Such single-city
studies have been criticized
for being applicable
only to the city under
study and for using different
modeling approaches.
These comments have led
to multicity meta-analyses
where the results are
pooled—for example,
the National Morbidity,
Mortality, and Air Pollution
Study (NMMAPS) conducted
on behalf of the Health
Effects Institute in
the United States and
the APHEA (Air Pollution
and Health: A European
Approach) studies in
Europe. NMMAPS examined
the associations between
daily hospital counts
for cardiovascular admissions
in the elderly and air
pollutants in 14 cities
in different regions
of the United States
(Dominici et al. 2002b;
Samet et al. 2000). The
APHEA studies have taken
place in two stages,
and the latest (APHEA2)
comprised eight European
cities in the investigation
of associations of air
pollution on daily cardiovascular
admissions (Le Tertre
et al. 2002). Multicity
studies have also been
conducted in Canada (Burnett
et al. 1997a, 1997b).
Despite these studies,
the strength of the association
between outdoor air pollution
and health effects is
still unclear because
of the complexity of
the time-series modeling.
In addition, when multiple
pollutants have been
examined, the independent
effects of each pollutant
are usually addressed
in multipollutant models,
but these are sensitive
to the modeling assumptions.
If the association with
one pollutant is nonlinear
or varies by season,
then a two-pollutant
model assuming a linear
relationship with each
pollutant might not give
the independent effect
of the second pollutant.
Therefore, the case–crossover
design (Maclure 1991),
which is less sensitive
to model assumptions,
is more appropriate.
This method investigates
the effects of acute
exposures and can also
examine both multiple
exposures and interactions
between exposures. It
has been applied to the
analysis of the acute
effects of environmental
exposures, especially
air pollution (Sunyer
et al. 2000). The method
matches case days to
nearby control days and
hence controls for covariates
that change slowly over
time (e.g., age, smoking
behavior, and usual diet).
Such matching also controls
for seasonal variation
and time trends in the
health event (Bateson
and Schwartz 2001).
In this study we aimed
to find associations
between outdoor air pollutant
and cardiovascular disease
(as measured by counts
of hospital admissions)
in cities in Australia
and New Zealand. The
study used two age groups, ≥ 65
years of age (elderly)
and 15–64 years
of age, although the
focus here is on the
elderly. The study also
examined differences
in the associations between
cities.
Data collection. Daily
hospital and pollution
data were collected for
the years 1998 through
2001 in five of the largest
cities in Australia (Brisbane,
Canberra, Melbourne,
Perth, Sydney) and two
cities in New Zealand
(Auckland, Christchurch).
In 2001, these cities
covered 53% of the Australian
population and 44% of
the New Zealand population.
Cardiovascular
health data and air
pollution data. Health
data were collected
for all cardiovascular
emergency hospital
admissions from state
government health
departments in Australia
and the New Zealand
Health Information
Service (Ministry
of Health). The cardiovascular
disease categories
used in the study
are shown in Table
1, and summary statistics
and demography for
each city are shown
in Table 2.
The pollutants considered
were particulate matter < 2.5 µm
in diameter (PM2.5)
and < 10 µm
in diameter (PM10)
in micrograms per cubic
meter; nitrogen dioxide
in parts per billion;
carbon monoxide in parts
per million; and ozone
in parts per billion.
Tapered element oscillating
microbalance (TEOM) air
samplers provided the
PM data. Daily pollutant
levels were calculated
by averaging over a network
of monitors in each city.
The summary statistics
for air pollutants and
weather are shown in
Table 3.
CO and NO2 were
the only pollutants monitored
in all seven cities on
a daily basis. For PM2.5,
daily measurements were
available in four of
the Australian cities:
Brisbane, Melbourne,
Perth, and Sydney. PM10 was
measured on a daily basis
in these four cities
and in Christchurch.
Statistical methods. We
used the time-stratified
case–crossover
method to find associations
between pollutants and
daily counts of hospital
admissions (Janes et
al. 2005). Controls were
chosen from strata of
length 28 days; days
either side of the case
day were excluded to
reduce the correlation
between case and control
exposure. The method
controlled for long-term
trend, seasonal changes,
and respiratory epidemics
by design. Using covariates,
there were additional
controls for temperature,
current minus previous
day’s temperature,
relative humidity, pressure,
extremes of hot and cold
(coldest and warmest
1% of days), day of the
week, public holiday
(yes/no), and day after
a public holiday(s) (yes/no).
Rainfall was also included
in some investigational
models.
The pollutant exposure
was the average of the
current and previous
day. Changes in admissions
are shown for a one interquartile
range (IQR) increase
in pollutant, using the
mean IQR across cities.
This makes the increases
from different pollutants
more comparable. An IQR
increase can be thought
of as the difference
between a moderately
good day and a moderately
bad day. The IQRs were
3.8 µg/m3 for
24-hr PM2.5,
7.5 µg/m3 for
24-hr PM10,
5.1 ppb for 24-hr NO2,
0.9 ppm for 8-hr CO,
and 8.8 ppb for 8-hr
O3.
To estimate the average
effect for all cities,
we combined the estimates
across cities using a
random effects meta-analysis
(Normand 1999) and quantified
the differences (heterogeneity)
between cities using
the I2 statistic
(Higgins and Thompson
2002). I2 values > 80%
indicate that differences
between cities are high; > 50%,
notable; > 20%, mild;
and < 20%, small.
To test whether one city
had an undue influence
on the meta-analysis,
we used a leave-one-city-out
sensitivity analysis
(Normand 1999).
We examined differences
in the increases between
cities using a hierarchical
model to incorporate
variables that differ
between cities and therefore
could modify the results
(effect modifiers) (Dominici
et al. 2002a). The increases
in admissions in each
city were regressed against
potential city-level
effect modifiers such
as average pollutant
level, temperature, and
percentage of the population ≥ 65
years of age. Differences
were examined only where
there was notable heterogeneity
(defined by I2 > 50%).
When a health outcome
showed a significant
association with more
than one pollutant, we
ran a multipollutant
model using a matched
case–crossover
approach (Schwartz 2004).
Matching is a traditional
approach to control for
potential confounding
in epidemiology. With
control days that are
both close in time to
the case day and also
matched on the level
of another pollutant,
the effect estimate cannot
be confounded by the
other pollutant. Matched
control days were defined
as 24-hr PM2.5 within
2 µg/m3,
24-hr PM10 within
3 µg/m3,
24-hr NO2 within
1 ppb, 24-hr CO within
0.5 ppm, and temperature
within 1°C.
All analyses were conducted
using SAS software (SAS
Institute Inc. 2001).
In the absence of an a
priori opinion
of which pollutants
were important to health,
we used a statistical
significance level
of 5%, with no correction
for multiple comparisons.
Although this increased
the chances of finding
spurious associations,
it reduced the chances
of missing any important
associations during
this early stage of
investigation of the
effects of air pollution
in Australia and New
Zealand.
In this study we used
monitoring data provided
by the relevant monitoring
agency in each city.
The data sets have been
used without extensive
analysis or corrections
beyond the basic quality
control needed to ensure
data validity for the
case–crossover
analysis. Some data sets
were not fully used (e.g.,
the PM10 data
from Auckland) because
they did not fully meet
the strict requirements
of the study but are
still regarded as valid
data sets for the purposes
for which they were gathered.
Table
4![Table 4](tab4sm.gif) |
Table
5![Table 5](table5sm.gif) |
![Figure 1](fig1sm.gif)
Figure 1. Estimated
increases (mean
and 95% CI) for
cardiac admissions
in the elderly
by city for four
pollutants (average
lag, 0–1;
one IQR increase).
(A) Maximum 8-hr
CO. (B) Average
24-hr NO2. (C)
Average 24-hr PM2.5.
(D) Average 24-hr
PM10. |
Table
6![Table 6](table6sm.gif) |
The associations between
pollutants and cardiovascular
hospital admissions are
shown in Table 4. In
the elderly, significant
associations were found
between the pollutants
CO, NO
2, and
PM and five categories
of cardiovascular disease
admissions. Arrhythmia
showed no associations
in the elderly but did
in the 15- to 64-year
age group. Stroke was
the only disease category
to show no associations
in either age group.
O
3 was the
only pollutant to show
no associations.
In elderly admissions,
the two largest statistically
significant increases
were for cardiac failure,
with a 6.9% increase
for a 5.1-ppb unit increase
in NO2 and
a 6.0% increase for a
0.9-ppm increase in CO.
For the elderly age
group, the relative risks
for all cardiac admissions
associated with CO, NO2,
PM2.5, and
PM10 are shown
for each city and the
meta-analysis in Figure
1, whichhighlights
some of the differences
in risk among the cities.
This heterogeneity is
quantified by the I2 statistics
in Table 4. The I2 statistics
indicate that more than
half of the results had
small heterogeneity.
Notable heterogeneity
was more often observed
in the elderly group.
Table 5 shows a much
reduced I2 when
Sydney was left out for
the association between
CO and cardiac admissions.
Figure 1A shows that
the association in Sydney
was much larger than
in the other cities.
The association was also
larger in Perth, but
the confidence intervals
(CIs) were wider. Table
5 and Figure 1B show
that when Christchurch
was left out, the association
between NO2 and
cardiac admissions was
similar for the remaining
cities.
Statistically significant
effect modifiers were
found only for associations
with PM2.5.
For cardiac admissions,
there was a greater association
with PM2.5 in
cities with less humidity.
For cardiac failure,
there was a greater association
with PM2.5 in
cities with higher pressure
and a greater percentage
of elderly.
Multipollutant results
using a matched case–crossover
analysis are shown in
Table 6. None of the
estimated increases changed
greatly when cases and
controls were matched
on temperature. The estimated
increase due to NO2 fell
greatly when cases and
controls were matched
on CO.
Cardiovascular
admissions in the
elderly. This
study found many
associations between
air pollution and
cardiovascular admissions
in cities in Australia
and New Zealand.
For every condition
but arrhythmia, the
increases in hospital
admissions were greater
in the elderly than
in the younger age
group (Table 4),
most likely because
the elderly are a
frailer population
with probable preexisting
heart problems. The
frailty of the elderly
is also the most
likely reason that
they did not show
increases in arrhythmia.
Arrhythmia and cardiac
failure are related
conditions because
atrial fibrillation
is a type of arrhythmia
and may precipitate
cardiac failure in
elderly people (Cowie
et al. 1999). Hence,
exposure to NO2 and
CO that led to arrhythmia
in the younger age
group led to the
more serious condition
of cardiac failure
in the elderly.
We found associations
at concentrations below
normal air quality health
guidelines (Table 3).
This suggests that current
air pollution guidelines
need to be revised. There
is good reason to believe
that lowering air pollution
levels would lead to
improvements in cardiovascular
health.
This results presented
here are based on statistically
significant findings.
Although this is not
ideal practice, there
is limited space in this
article; a complete set
of results will be available
in a forthcoming report
(Expansion of the Multi-City
Mortality and Morbidity
Study, National Environment
Protection Council).A
non-statistically significant
association does not,
of course, mean that
a relationship does not
exist. This is particularly
important for those admissions
with smaller numbers
of events and hence less
power (e.g., stroke in
the younger age group).
Differences among
cities. Differences
in the associations
between cities in
this study were mostly
not notable (I2 < 50%).
This suggests that
the relationship
between exposure
and disease was often
similar. There was
more notable heterogeneity
in the elderly population,
which is partly due
to the greater size
of the associations
in this age group.
In an attempt to explain
the notable heterogeneity,
we used effect-modifier
analyses. However, we
found effect modifiers
only for associations
with PM2.5.
The effect of PM2.5 in
Australian cities depended
on the percentage of
the elderly and average
weather conditions. Less
average humidity and
higher average pressure
led to a greater association.
To investigate these
modifications further,
we reran the case–crossover
models in each city including
an interaction term for
24-hr PM2.5 and
rainfall (results not
shown). Higher rainfall
led to a smaller association
between cardiovascular
admissions and PM2.5 in
all four cities. This
is not surprising, considering
that rainfall is a primary
removal mechanism for
PM10 and PM2.5,
but less so for gaseous
pollutants.
Comparison with
three other large
studies. The
aim and design of
this study were similar
to those of three
other large studies:
the APHEA2 study
of eight cities in
Europe (Le Tertre
et al. 2002), NMMAPS
with 14 U.S. cities
(Samet et al. 2000),
and a Canadian study
of 10 cities (Burnett
et al. 1997b). The
results here for
PM pollution in terms
of elderly cardiac
admissions are similar
to those found in
APHEA2 and NMMAPS,
and congestive heart
failure in the Canadian
study. For example,
we found the mean
increase for cardiac
admissions for the
15–64-year
age group to be less
than half that in
the older group,
a result similar
to that of the APHEA2
study. The APHEA2
study also found
that the heterogeneity
in total cardiac
admissions (all ages)
was related to the
percentage of elderly.
However, the results
for the confounding
effects on PM associations
by including CO are
different (but PM
was not monitored
in every city here).
A difference between
this study and the three
large multicity studies
is that emissionsources
for PM, and therefore
the PM composition, differ.
For example, Chan et
al. (1999) found significantly
higher contributions
from sea salt and nonanthropogenic
crustal sources for both
PM2.5 and
PM10 in
Brisbane than in overseas
cities.
Another important difference
from the other studies
is in the statistical
methods used here. The
NMMAPS and APHEA2 studies
used generalized additive
models, in which confounding
was estimated by adding
the copollutant into
the model. In the APHEA2
study, the PM10 associations
were significantly reduced
in the multipollutant
models by the inclusion
of NO2 (as
found here) and slightly
(but becoming statistically
insignificant) for CO.
However, using black
smoke to estimate the
PM impacts showed no
confounding by CO and
much less by NO2.
There was no significant
confounding of the PM
associations by CO or
NO2 found
in NMMAPS.
Addressing confounding
between pollutants. Instead
of using a multipollutant
model, we controlled
for confounding by
matching in the case–crossover
analysis. For elderly
hospital admissions,
the CO associations
remained of a similar
size when matched
with NO2.
Conversely, the NO2 became
smaller when matched
with CO (Table 6).
Matching was also used
to control for the important
confounder of temperature.
The results changed little
when matched on temperature,
strongly suggesting that
the association between
air pollution and cardiovascular
disease is not confounded
by temperature.
Is outdoor air
pollution a good
indicator of exposure? A
problem in interpreting
the results from
this study is that
it used outdoor air
pollution concentrations
measured at fixed-point
monitors (ambient
concentrations),
whereas people spend
most of their time
indoors. Recent studies
in Baltimore, Maryland
(Sarnat et al. 2001)
and Boston, Massachusetts
(Sarnat et al. 2005)
indicate that such
ambient concentrations
may be poor surrogates
for actual exposure
to air pollution,
especially in winter
when buildings are
more sealed. However,
winters in Australia
are mild, meaning
that people will
likely spend more
time outdoors and
that houses are designed
to lose heat rather
than trap it. Hence,
exposure to the air
may be high all year
round in Australia
(winter exposure
in New Zealand may
be more similar to
that in Baltimore).
A similar conclusion
was drawn by a study
of the effects of
cold temperatures
on cardiovascular
disease (Barnett
et al. 2005). In
that study, regions
with mild winters
showed greater increases
in cold-related cardiovascular
events than did regions
with usually cold
winters.
The study in Baltimore
also found that ambient
concentrations for CO
and NO2 were
often better surrogates
for actual exposure to
PM than to CO and NO2,
especially in summer
(Sarnat et al. 2001).
However, the more recent
Boston study did note
that there were some
correlations between
ambient concentrations
and actual exposure to
these gases in summer
(Sarnat et al. 2005).
Outdoor concentrations
for pollutants such as
NO2, CO, and
PM often arise from the
same combustion emissions
sources, such as motor
exhausts.
CO as a marker
for pollution sources. There
is evidence that
air pollutants (NO2,
CO) may trigger fibrillation
in people with a
history of serious
arrhythmia (Peters
et al. 2000). The
effect of CO on cardiovascular
disease is well known,
with CO replacing
oxygen in the blood
stream, but at the
low CO concentrations
prevailing in the
cities under study,
it cannot be simply
assumed that if CO
is the “cause” of
any effects found
here, it is due to
this mechanism. The
associations found
for CO, NO2,
and PM are not additive,
but probably refer
to the impacts of
a similar pollutant “mix.” Given
that the CO associations
show the least change
when matched with
the other pollutants
(Table 6), this indicates
that the air pollutant
mixture arising from
emission sources
dominating the CO
emissions (usually
human combustion
sources) is the primary
cause of the association,
not the effect of
CO itself.
For both Australian
and New Zealand cities,
the results show that
increases in outdoor
concentrations of CO,
NO2, and PM
have significant associations
with increases in cardiovascular
admissions for adults,
especially the elderly
(≥ 65
years of age). Associations
were found at concentrations
below normal air quality
health guidelines. There
were significant associations
between air pollution
and arrhythmia admissions
in the younger age group,
which were not apparent
for the elderly. For
the elderly, there were
significant associations
between air pollution
increases and increases
in hospital admissions
for ischemic heart disease
and myocardial infarction,
and these were not apparent
for the younger group.
Atrial fibrillation can
precipitate cardiac failure,
especially in the elderly,
and a significant relationship
has been identified here
in the adult age group
(15–64 years) between
increases in hospital
admissions for arrhythmia
and increases in air
pollution.
The associations for
NO2 appear
to be stronger in Australian
than in New Zealand cities,
whereas those of CO are
similar for cities in
both countries. In Australian
cities, PM10 and
PM2.5 had
a similar association,
apart from that for arrhythmia.
These PM2.5 associations
differed among cities
due to different climate
conditions for humidity
(the lower the humidity,
the greater the association).
It is difficult to
separate the associations
for different pollutants
because there are common
emission sources for
CO, NO2, and
PM (e.g., motor vehicle
exhausts). Also, outdoor
concentrations are often
not good surrogates for
actual exposure, with
outdoor levels for the
gases sometimes being
good surrogates for actual
exposure to PM, especially
in summer.