O A

Transcription

O A
ORIGINAL ARTICLE
Long-Term Exposure to Outdoor Air Pollution and
Incidence of Cardiovascular Diseases
Richard W. Atkinson,a Iain M. Carey,a Andrew J. Kent,b Tjeerd P. van Staa,c H. Ross Anderson,d
and Derek G. Cooka
Background: Evidence based largely on US cohorts suggests that
long-term exposure to fine particulate matter is associated with cardiovascular mortality. There is less evidence for other pollutants and
for cardiovascular morbidity. By using a cohort of 836,557 patients
age 40 to 89 years registered with 205 English general practices in
2003, we investigated relationships between ambient outdoor air pollution and incident myocardial infarction, stroke, arrhythmia, and
heart failure over a 5-year period.
Methods: Events were identified from primary care records, hospital admissions, and death certificates. Annual average concentrations in 2002 for particulate matter with a median aerodynamic
diameter <10 (PM10) and <2.5 microns, nitrogen dioxide (NO2),
ozone, and sulfur dioxide at a 1 × 1 km resolution were derived
from emission-based models and linked to residential postcode.
Analyses were performed using Cox proportional hazards models
adjusting for relevant confounders, including social and economic
deprivation and smoking.
Results: While evidence was weak for relationships with myocardial
infarction, stroke, or arrhythmia, we found consistent associations
between pollutant concentrations and incident cases of heart failure.
An interquartile range change in PM10 and in NO2 (3.0 and 10.7 µg/
m3, respectively) both produced a hazard ratio of 1.06 (95% confidence interval = 1.01–1.11) after adjustment for confounders. There
was some evidence that these effects were greater in more affluent
areas.
Conclusions: This study of an English national cohort found evidence linking long-term exposure to particulate matter and NO2 with
Submitted 25 October 2011; accepted 27 July 2012.
From the aDivision of Population Health Sciences and Education and MRCHPA Centre for Environment and Health, St George’s, University of
London, London, United Kingdom; bAEA Technology plc, Harwell IBC,
Didcot, Oxfordshire, United Kingdom; cClinical Practice Research Datalink, Medicines and Healthcare Products Regulatory Agency, London,
United Kingdom; and dMRC-HPA Centre for Environment and Health,
King’s College London, London, United Kingdom.
Supported by funding from Policy Research Programme in the Department
of Health.
The authors report no conflict of interests.
Supplemental digital content is available through direct URL citations
in the HTML and PDF versions of this article (www.epidem.com). This
content is not peer-reviewed or copy-edited; it is the sole responsibility
of the author.
Correspondence: Richard W. Atkinson, Division of Population Health Sciences and Education and MRC-HPA Centre for Environment and Health,
St George’s, University of London, Cranmer Terrace, London SW17 0RE,
United Kingdom. E-mail: [email protected].
Copyright © 2012 by Lippincott William and Wilkins
ISSN: 1044-3983/13/2401-0044
DOI: 10.1097/EDE.0b013e318276ccb8
44 | www.epidem.com
the development of heart failure. We did not, however, replicate associations for other cardiovascular outcomes that have been reported
elsewhere.
(Epidemiology 2013;24: 44–53)
E
xposure to outdoor particulate air pollution is believed to
be associated with cardiovascular disease (CVD) mortality
and morbidity.1,2 Mechanistic studies have suggested a number
of potential pathways from exposure to disease, including
release of proinflammatory mediators, changes to the systemic
autonomic nervous system balance, and translocation
of particles or particle constituents into the circulation.2
Epidemiologic evidence for adverse health effects comes
from studies of exposure over short (hours to days) and long
(eg, a few years) periods. Evidence for the latter is provided
by cohort studies, most of which have focused on all-cause
and cardiopulmonary mortality. Exposure estimates for cohort
members have been derived from average concentrations of
fine particles derived from fixed-site community monitors3–10
or have used spatial modeling techniques to estimate exposure
on a finer spatial scale.11–17 Studies of CVD mortality
report positive associations with fine particles.10,18–21 Only
a few studies have included nonfatal outcomes (such as
hospitalization) in their identification of incident events.13–16,20
Furthermore, only a small number of studies have considered
other regulated pollutants such as nitrogen dioxide (NO2),
ozone, and sulfur dioxide (SO2) in relation to CVD with less
consistent results.22,23
We investigated the associations between long-term
exposure to a number of ambient outdoor air pollutants and
the incidence of a range of CVD events using a national
cohort of English adults registered with their general practitioner. Linkage of primary care records to national registers of hospital admissions and mortality records, together
with small-area measures of social and economic deprivation, maximized case identification and control for important confounders. Annual concentrations of a range of
pollutants at a 1 × 1 km grid spatial resolution were derived
from national models that incorporate distant, point, and
local sources.24
Epidemiology • Volume 24, Number 1, January 2013
Epidemiology • Volume 24, Number 1, January 2013
METHODS
Data Sources
Clinical Practice Research Datalink
The Clinical Practice Research Datalink (formerly the
General Practice Research Database) is a large validated
primary care database that has been collecting anonymized
patient data from participating UK general (family) practices
since 1987.25 As of 2010, it represents approximately 8% of
the UK population of which 98% is believed to be registered
with a general practitioner. It includes a full longitudinal
medical record for each patient, with information on diagnoses, prescriptions, and tests performed within the practice and
information from outside sources such as outpatient consultations and admissions to hospital.
Hospital Episode Statistics and Death
Registration Data
Details on every hospital admission in the United Kingdom were provided by the Hospital Episode Statistics database. This system records clinical, patient, administrative,
and geographical information on all National Health Service
(NHS) funded inpatient episodes. Death registration data are
processed by the Office for National Statistics in the United
Kingdom. Details recorded include age, sex, address, and both
the immediate and underlying causes of death coded according to the International Classification of Diseases (ICD). Subject to the practice’s approval, the Clinical Practice Research
Datalink patients are linked to both Hospital Episode Statistics and Office for National Statistics data by a “trusted third
party” using their NHS number, sex, date of birth, and postcode. As of 2010, this linkage has been performed for 224
Clinical Practice Research Datalink practices (approximately
40% of all practices).
Study Group
Eligibility
We selected 205 English practices that, by 1 January
2003, (1) were recording data that were deemed to be “Upto-Standard” (an internal Clinical Practice Research Datalink quality control metric indicating the date from which a
practice was continually recording high-quality data), and (2)
had linked hospital admission and mortality data available
from this date. From these practices, we identified a cohort
of patients who were between the age of 40 and 89 years in
2003, were fully registered on 1 January 2003, and had been
continually so for 12 months. This resulted in 836,557 patients
eligible for our study.
Outcomes
An incident event was defined as the first event occurring between 2003 and 2007 recorded on any of our three data
sources: Clinical Practice Research Datalink (Read codes),
Hospital Episode Statistics (ICD codes), and Office for
National Statistics mortality (ICD codes). The main conditions
© 2012 Lippincott Williams & Wilkins
Air Pollution and Incidence of Cardiovascular Disease
and their ICD-10 codes were coronary heart disease (CHD),
I20-25; myocardial infarction (MI), I21-23; cerebrovascular
disease, I60-69; stroke, I61, I63-64; arrhythmias, including
cardiac arrest, I46-49, R00.1; and heart failure, I50. These
were manually mapped to corresponding Read codes (list
available from authors). Patients with disease recorded before
1 January 2003 (or with no date attached to the code) were
excluded from the analysis for that outcome. Thus, a patient
with angina first recorded in 2000 would still be eligible for
analyses of MI and other outcomes.
Air Pollution
Annual mean concentrations of particles with a median
aerodynamic diameter of <10 μm (particulate matter, PM10)
and <2.5 μm (PM2.5), SO2, NO2, and ozone (O3) for 1 × 1 km
grids covering England were estimated using air dispersion
models. Details of the modeling methodology are given in
the eAppendix (http://links.lww.com/EDE/A628). In brief,
we constructed the models for PM10, PM2.5, NO2, and SO2
by identifying all known emission sources by emission sector (eg, power generation, domestic combustion, road traffic, industry, waste) and estimating quantities of emissions.
Pollution concentrations were calculated by summing the
estimated concentrations for pollutant-specific components
from distant sources (characterized by rural background concentrations, interpolated from rural measurements), point
sources (calculated using an air dispersion model), and local
area sources (calculated using a kernel-based air dispersion
model) including the effect of weather conditions. O3 maps
were constructed by interpolating data from rural monitoring
stations and adjusting for effects of altitude and NOx emissions in urban areas by using the concentrations calculated
with the air dispersion models.
Model validation was assessed using data from the
national air quality monitoring networks (also used in the
global calibration of the model) and from a separate network
of monitors also yielding high-quality data (verification sites).
Further details are given in the eAppendix (http://links.lww.
com/EDE/A628). In brief, model validation was good for NO2
(eg, in 2002, the R2 statistics were 0.80 for the national network
sites and 0.57 for the verification sites). For PM10, there was
moderate agreement (R2 = 0.29 for national network sites and
R2 = 0.46 for verification sites in 2002). Model performance
statistics for PM2.5 were not available for 2002 (because of
the paucity of monitoring data available at that time), but in
later years, the performance was good (eg, in 2008, R2 statistic
of 0.5 at six verification sites). Concentrations of PM10 and
PM2.5 were highly correlated (r > 0.9). Therefore, to simplify
the presentation of our findings, we report results for PM10 in
the article and present results for PM2.5 in the eTables (http://
links.lww.com/EDE/A628). For SO2, however, the validation
was moderate to poor and varied substantially from year to
year (from 2002 to 2006, the R2 statistics varied from 0.23 to
0.45 at national network sites and 0 to 0.6 at the verification
www.epidem.com | 45
Atkinson et al
sites). Modeled ozone concentrations were verified using the
number of days exceeding 120 μg/m3. R2 statistics at national
network sites and verification sites for the period 2002 to 2004
were 0.71 and 0.48, demonstrating good to reasonable model
performance.
Data Linkage
Annual pollution concentrations for each 1 × 1 km grid
were linked to the patients’ residential postcodes in the Clinical Practice Research Datalink by a “trusted third party” to
ensure anonymity. In England, the 1.4 million active postcodes vary geographically in size but identify on average
about 15 residential addresses. The centroid of each postcode
was mapped to the nearest centroid of each 1 × 1 km grid.
Across the 205 practices, the number of 1 × 1 km grids with a
registered patient ranged from 27 to 252.
Covariates
We extracted the following patient information from the
Clinical Practice Research Datalink to be used as covariates
in our models: age, sex, smoking, body mass index (BMI),
diabetes, and hypertension. All covariates had to be recorded
by the start of follow-up (1 January 2003). Practice location
was designated as being in the North or South or in Greater
London.
Socioeconomic status was assessed by using the
Index of Multiple Deprivation 2007, a composite small-area
(approximately 1500 people) measure of deprivation used in
England for allocation of resources.26 The published index
was constructed by a weighted score of seven domains:
income, employment, health deprivation and disability, education skills and training, barriers to housing and services,
crime, and living environment. Because the living environment domain contains a subdomain relating to air quality, we
recalculated the overall index excluding this domain. We then
reranked areas nationally and summarized them as deciles.
Statistical Analyses
We used Cox proportional hazards models (PROC
PHREG in SAS version 9.1.3; SAS Institute, Cary, NC) to
investigate the association between outcome and air pollution.
Time to event was modeled in months, and patients were censored if they deregistered from the practice (or died) without
experiencing an event before 31 December 2007. We considered two approaches to modeling pollution: a fixed level
(2002) and a varying time-dependent covariate (based on previous years’ exposure). In the models, we adjusted cumulatively for the following: (1) age and sex; (2) smoking, BMI,
diabetes, and hypertension; and (3) Index of Multiple Deprivation. To allow comparison across pollutants, hazard ratios
(HRs) were calculated for an interquartile range (IQR) change
in each pollutant (given in Table 1). Air pollutants were modeled individually and in pairs in two-pollutant models.
To investigate the clustering impact of air pollution
levels within practice, we calculated intraclass correlation
46 | www.epidem.com
Epidemiology • Volume 24, Number 1, January 2013
coefficients. In the Cox models, we accounted for practice
in two ways. First, we used a robust sandwich estimate for
the covariance matrix, which results in robust standard errors
for the effect estimates. Second, we added to the full model a
term for “practice mean exposure,” which allows us to derive
the contribution of between- and within-exposure to the overall effect.27
RESULTS
Air Pollution Linkage
Successful postcode linkage was made for approximately
99% of patients in our cohort for all pollutants of interest
(Table 1). Annual mean concentrations of PM10 and NO2 were
higher in London than the rest of the country, whereas SO2
was highest in the North. PM10, SO2, and NO2 were all associated with greater deprivation (Fig. 1), whereas O3 was higher in
more affluent areas. The pattern of higher concentrations with
increasing deprivation was seen in the North and London, but
there was no such trend in the South (eTable 3, http://links.lww.
com/EDE/A628). Descriptive statistics for PM2.5 are shown in
eTable 4 (http://links.lww.com/EDE/A628).
Among patients with assigned exposure, the two pollutants most closely related were PM10 and NO2 (positive correlation r = 0.86). O3 was negatively correlated with PM10,
SO2, and NO2. Practice was a strong determinant in explaining the amount of variation in pollutant concentrations seen
between patients in the cohort (intraclass correlation coefficients in Table 1). For example, 94% of variation for O3 and
77% of variation for SO2 were caused by between-practice
differences.
Incidence of Cardiovascular Events
A summary of the incident recording of MI, stroke,
heart failure, and arrhythmia events is presented in Table 2.
For example, for incident MIs, 25,871 patients (3.1%) already
had an MI recorded by 1 January 2003 and therefore were
excluded from the analyses of incident MI. Among the 810,686
eligible patients, 13,956 (1.7%) had a first MI recorded in
2003 to 2007.
Men were more likely than women to experience an
MI, arrhythmia, or heart failure; for stroke, rates were similar.
There were strong effects of geography and deprivation, with
those in the North and more deprived areas experiencing a
greater rate of disease (eg, for MIs, those in the North had
an incidence of 2.1% vs. 1.6% in the South; for the most
deprived, it was 2.4% vs. 1.4% for the least deprived).
Effects of Air Pollution
The associations between assigned air pollution exposure (for IQR changes in concentrations during 2002) and
5-year incidence of disease (2003–2007) are quantified in a
series of HRs in Table 3 (results scaled to 10 µg/m3 are given
in eTable 5, http://links.lww.com/EDE/A628). The strongest associations with pollution were seen for heart failure.
© 2012 Lippincott Williams & Wilkins
Epidemiology • Volume 24, Number 1, January 2013
Air Pollution and Incidence of Cardiovascular Disease
TABLE 1. Summary of Assigned Pollutant Levels in 2002 for Study Cohort (n = 836,557)
Assigned Pollutant Exposure in 2002
Patients with linkage, no. (%)
Pollution level (µg/m3)
Mean (SD)
Range
Interquartile range
Region (no. of practices); mean (SD)
North (n = 81)
South (excluding London, n = 96)
London (n = 28)
Practice in urban area (no. of practices); mean (SD)
Yes (n = 173)
No (n = 32)
Deprivation (modified IMD) quintile; mean (SD)
1 (most deprived)
2
3
4
5 (least deprived)
Correlationa with future years exposure
2003
2004
2005
2006
Correlationa with other pollutants
NO2
SO2
O3
Intraclass correlation (ICC) by practice
PM10
NO2
SO2
O3
831,788 (99)
831,375 (99)
824,388 (99)
825,598 (99)
19.7 (2.3)
10.6–29.8
3.0
22.5 (7.4)
1.7–60.8
10.7
3.9 (2.1)
0.03–24.2
2.2
51.7 (2.4)
44.0–66.2
3.0
19.8 (2.3)
19.1 (2.0)
22.5 (1.2)
23.4 (6.3)
19.4 (6.1)
33.3 (4.5)
4.8 (2.1)
3.3 (1.9)
3.8 (1.2)
50.9 (2.4)
52.6 (2.2)
50.2 (0.8)
20.1 (2.2)
17.7 (1.9)
23.6 (7.0)
16.1 (6.2)
4.0 (2.0)
3.5 (2.6)
51.5 (2.2)
52.3 (3.2)
21.0 (2.4)
20.2 (2.5)
19.4 (2.5)
19.3 (2.2)
19.5 (1.7)
26.6 (7.1)
23.7 (7.8)
21.4 (7.9)
21.0 (7.0)
21.9 (6.1)
4.4 (2.3)
4.4 (2.5)
3.8 (2.2)
3.7 (1.9)
3.6 (1.5)
51.1 (2.2)
51.2 (2.2)
51.8 (2.6)
51.7 (2.5)
52.0 (2.1)
0.92
0.91
0.84
0.85
0.98
0.97
0.97
0.97
0.81
0.67
0.68
0.61
0.63
0.55
0.56
n/a
0.86
0.52
–0.44
0.87
—
0.50
–0.48
0.90
—
—
–0.45
0.77
—
—
—
0.94
a
Spearman’s rank correlation coefficient.
FIGURE 1. A–D, Assigned exposure in all patients (n = 836,557) by quintiles of age-sex adjusted (modified) Index of Multiple
Deprivation (IMD).
© 2012 Lippincott Williams & Wilkins
www.epidem.com | 47
Epidemiology • Volume 24, Number 1, January 2013
Atkinson et al
TABLE 2. Incidence of MI, Stroke, Arrhythmias, and Heart Failure 2003–2007
MI
(n = 810,686)
Baseline Variables
Stroke
(n = 819,370)
Arrhythmias
(n = 790,751)
Heart failure
(n = 810,188)
All Patientsa
No.
No.
Incidenceb
%
No.
Incidenceb
%
No.
Incidenceb
%
No.
Incidenceb
%
836,557
13,956
1.72
13,012
1.59
21,720
2.75
12,851
1.59
405,169
431,388
8,471
5,485
2.27
1.33
6,267
6,745
1.63
1.64
11,209
10,511
3.14
2.73
6,712
6,139
1.83
1.58
242,576
228,234
166,015
128,305
71,427
927
2,178
2,924
4,289
3,638
0.38
0.98
1.86
3.64
5.59
478
1,295
2,409
4,428
4,402
0.20
0.57
1.48
3.63
6.70
1,427
2,905
4,725
7,417
5,246
0.59
1.30
3.02
6.61
9.22
293
878
2,085
4,615
4,980
0.12
0.39
1.30
3.90
8.22
396,647
151,974
161,513
126,423
303,327
5,293
3,307
3,518
1,838
4,181
1.47
2.67
1.93
1.48
1.63
5,681
3,049
2,565
1,717
4,238
1.49
2.14
1.70
1.44
1.58
10,294
5,624
3,457
2,345
6,869
2.95
2.93
3.44
2.09
2.74
5,329
3,625
2,310
1,587
3,551
1.48
2.10
2.10
1.34
1.40
243,556
109,104
180,570
4,702
2,185
2,888
1.88
2.28
1.60
4,236
1,899
2,639
1.68
1.91
1.52
7,330
3,924
3,597
3.12
4.22
2.22
4,194
2,570
2,536
1.74
2.90
1.47
808,737
27,820
12,620
1,336
1.69
4.02
11,897
1,115
1.58
3.06
20,458
1,262
2.90
3.71
11,366
1,485
1.59
4.13
125,125
711,362
9,756
4,200
1.62
2.60
8,638
4,374
1.46
2.32
14,819
6,901
2.59
4.25
8,060
4,791
1.46
2.56
103,191
146,811
171,437
208,859
205,309
2,343
2,823
2,937
3,234
2,605
2.39
2.01
1.79
1.65
1.43
2,097
2,548
2,748
3,040
2,564
2.09
1.76
1.65
1.54
1.43
2,847
3,967
4,488
5,392
4,997
3.04
2.96
2.89
2.91
2.90
2,109
2,646
2,778
2,983
2,327
2.20
1.92
1.74
1.58
1.37
320,154
424,411
91,992
6,082
6,652
1,222
2.05
1.63
1.52
5,483
6,371
1,158
1.83
1.53
1.45
8,647
11,088
1,985
3.07
2.89
2.64
5,404
6,132
1,315
1.91
1.54
1.73
All patients
Sex
Men
Women
Age in 2003 (in years)
40–49
50–59
60–69
70–79
80–89
Smoking
Nonsmoker
Ex-smoker
Current smoker
Unknown
BMI (kg/m2)
<25
25–30
>30
Unknown
Diabetes
No
Yes
Hypertension
No
Yes
Modified IMD Quintile
1 (most)
2
3
4
5 (least)
Practice Region
North
South (excluding London)
London
a
All patients refers to all patients in cohort before exclusions are made for existing disease at baseline (MI n=25,871, 3.1%; Stroke n=17,187, 2.1%; Arrhythmias n=45,806, 5.5%;
Heart Failure n=26,369, 3.2%). Each of the columns for the diseases is based on patients without the existing disease only.
b
Adjusted for age and sex.
For example, IQR changes in PM10 (3.0 µg/m3) and NO2
(10.7 µg/m3) were both associated with 11% increases in heart
failure. These associations were reduced when adjusted for
smoking and BMI and further reduced with adjustment for the
Index of Multiple Deprivation (HR = 1.06 [95% confidence
interval {CI} = 1.01–1.11] for both PM10 and NO2). HRs and
95% CIs for PM2.5 were similar to those for PM10 (eTable 6,
http://links.lww.com/EDE/A628).
48 | www.epidem.com
The evidence for associations of PM10 and NO2 with
incident MI, stroke, or arrhythmias was weak both before and
after adjustment for potential confounders. By contrast, SO2
maintained an association with MIs even after adjustment for
the Index of Multiple Deprivation (1.05 [1.02–1.07] for a 2.2µg/m3 increment in SO2). O3 tended to show negative associations with the outcomes, which lessened with adjustment,
although heart failure persisted (6% lower risk with a 3-µg/m3
© 2012 Lippincott Williams & Wilkins
Epidemiology • Volume 24, Number 1, January 2013
Air Pollution and Incidence of Cardiovascular Disease
TABLE 3. Hazard Ratiosa Summarizing the Change in Risk of Incident MI, Stroke, Arrhythmias, and Heart Failure in 2003–2007
Associated with an Interquartile Change in Each Pollutant
Myocardial Inf.
(n = 810,686)
Stroke
(n = 819,370)
Arrhythmias
(n = 790,751)
Heart Failure
(n = 810,188)
Pollutant Baseline Variables
HRa
(95% CI)
HRa
(95 % CI)
HRa
(95% CI)
HRa
(95% CI)
PM10
Adjusted for age and sex
Further adjusted for smoking, BMI, and comorbidityb
Further adjusted for (modified) IMD
1.03
1.01
0.98
(0.99–1.07)
(0.98–1.05)
(0.94–1.01)
1.02
1.01
0.98
(0.99–1.05)
(0.98–1.04)
(0.95–1.01)
1.00
1.00
0.99
(0.97–1.03)
(0.97–1.03)
(0.96–1.02)
1.11
1.09
1.06
(1.06–1.17)
(1.05–1.14)
(1.01–1.11)
NO2
Adjusted for age and sex
Further adjusted for smoking, BMI, and comorbidityb
Further adjusted for (modified) IMD
1.04
1.02
0.98
(0.99–1.09)
(0.98–1.07)
(0.93–1.03)
1.04
1.03
0.99
(1.00–1.08)
(0.99–1.07)
(0.95–1.03)
1.00
1.00
0.99
(0.96–1.03)
(0.97–1.03)
(0.96–1.02)
1.11
1.10
1.06
(1.06–1.17)
(1.05–1.15)
(1.01–1.11)
SO2
Adjusted for age and sex
Further adjusted for smoking, BMI, and comorbidityb
Further adjusted for (modified) IMD
1.09
1.07
1.05
(1.07–1.10)
(1.04–1.10)
(1.02–1.07)
1.05
1.04
1.02
(1.04–1.07)
(1.02–1.07)
(1.00–1.05)
1.03
1.02
1.02
(1.01–1.04)
(1.00-1-04)
(1.00–1.04)
1.08
1.07
1.04
(1.07–1.10)
(1.03–1.10)
(1.01–1.08)
O3
Adjusted for age and sex
Further adjusted for smoking, BMI, and comorbidityb
Further adjusted for (modified) IMD
0.93
0.94
0.96
(0.90–0.96)
(0.91–0.97)
(0.93–1.00)
0.97
0.98
1.00
(0.94–1.01)
(0.95–1.02)
(0.97–1.04)
1.01
1.01
1.02
(0.98–1.05)
(0.98–1.05)
(0.98–1.05)
0.91
0.92
0.94
(0.87–0.95)
(0.88–0.96)
(0.90–0.98)
a
Hazard ratios refer to an IQR change in each pollutant (PM10 = 3.0 µg/m3, SO2 = 2.2 µg/m3, NO2 = 10.7 µg/m3, O3 = 3.0 µg/m3).
Diabetes and hypertension.
b
FIGURE 2. Stratified hazard ratios from fully adjusted model for incidence of heart failure and PM10 (A), NO2 (B), SO2 (C), and O3
(D) exposure.
© 2012 Lippincott Williams & Wilkins
www.epidem.com | 49
Epidemiology • Volume 24, Number 1, January 2013
Atkinson et al
TABLE 4. Within- and Between-Practice Hazard Ratiosa Summarizing the Change in Risk of Incident MI, Stroke, Arrhythmias,
and Heart Failure in 2003–2007 From an Interquartile Change in PM10
MI
(n = 810,686)
Stroke
(n = 819,370)
Arrhythmias
(n = 790,751)
Heart Failure
(n = 810,188)
Adjustment for
Modified IMD
Practice Effect
HRa
(95 % CI)
HRa
(95 % CI)
HRa
(95 % CI)
HRa
(95 % CI)
No
No
Yes
Yes
Within
Between
Within
Between
1.08
0.93
1.01
0.96
(1.02–1.15)
(0.86–1.00)
(0.96–1.07)
(0.90–1.03)
1.05
0.96
1.00
0.98
(0.97–1.13)
(0.89–1.04)
(0.93–1.06)
(0.91–1.06)
1.01
0.98
1.01
0.98
(0.97–1.06)
(0.93–1.04)
(0.96–1.06)
(0.93–1.04)
1.09
1.01
1.02
1.04
(1.02–1.16)
(0.93–1.09)
(0.96–1.09)
(0.97–1.12)
a
All models adjusted for age, sex, smoking, BMI, diabetes and hypertension. Hazard ratios refer to an IQR change in PM10 of 3.0 µg/m3.
increase). Results for the broader groups of CHD and cerebrovascular disease were similar to the respective subgroups of MI
and stroke (eTable 7, http://links.lww.com/EDE/A628).
Fitting a time-varying covariate for pollution in the model
based on previous years’ exposure through the study made no
appreciable difference to our results (data not shown). We also
investigated the robustness of effect sizes to adjustment for
other pollutants (eTable 8, http://links.lww.com/EDE/A628).
The full model effects of a 6% increase in heart failure for an
IQR change in PM10 and NO2 were reduced to 5% when SO2
was added to the model and to 3 to 4% when O3 was added.
The negative associations with O3 moved toward unity with
adjustment for all pollutants.
Effect Modification
Figure 2 shows the results from fitting the fully
adjusted model for heart failure for all pollutants stratified
by sex, age, BMI, smoking, Index of Multiple Deprivation,
and region. For PM10, NO2, and SO2, associations tended
to increase as the level of deprivation decreased. The pattern among other stratified variables was less clear, although
patients with BMI > 30 and ex-smokers had smaller associations for all pollutants. There was some variation in the
effect size of the associations by geography; larger associations were seen in the South than in the North. Within London practices, associations were greater (except for NO2),
although CIs were wide.
Clustering Effect of Practice
We investigated the influence of practice by partitioning
the HR into between- and within-practice estimates. Table 4
shows the results of doing this for PM10 in models with and
without adjustment for Index of Multiple Deprivation. In the
model not adjusting for the Index of Multiple Deprivation,
positive associations are seen within practice (ie, 8% increase
in MIs per IQR increase in PM10, which contrasts with negative
associations between practice [7% decrease in MIs per IQR
increase in PM10]). The adjustment for the Index of Multiple
Deprivation brings these divergent associations much closer
together, removing most of the within-practice effects. A similar pattern was noted with the other pollutants (data not shown).
50 | www.epidem.com
DISCUSSION
In this large, population-based cohort study, we found
inconsistent evidence to support the hypothesis that longterm exposure to particulate air pollution was associated with
increased incidence of CVD. Associations between PM and
incident CHD, MI, stroke, and arrhythmias were close to unity
but increased for heart failure. Increased incidence of heart
failure was also positively associated with NO2 and SO2.
Our study is the first to investigate the long-term effects
of air pollution on the development of CVD in a large UK
population incorporating individual risk factors. Previously,
Elliot et al28 reported positive associations between black
smoke and SO2 and subsequent mortality in Great Britain
using small-area population and socioeconomic status statistics, but lacking data on individual risk factors. More recently,
a large representative study of the UK population reported on
long-term effects in relation to the prevalence of CHD and
found small associations with PM10.29
Air Pollution Exposure Assignment
In our study, pollution estimates for 1 × 1 km grids across
England (mapped to residents’ postcodes) were estimated
using pollution emission inventories and meteorological
information combined with air dispersion models. Dispersion
models, together with land-use-regression models that predict
pollution concentrations at a given site based on surrounding
land use and traffic characteristics, provide improvements
on methods that assign exposure using measurements from
the nearest monitor.21 It has been suggested that dispersion
models are more sophisticated and reliable than land-useregression models in intraurban settings.30 The key advantage
of dispersion models over other approaches is that they
provide a better representation of the process under study.30
However, they are more expensive to implement because of
their substantial data requirements.30 A comparison of the
performance of land-use-regression and dispersion models
in predicting annual NO2 concentrations in the Netherlands
concluded there was moderate agreement between the
methods.31 Furthermore, the performance of our models was
comparable to the land-use-regression data available for the
United Kingdom.32,33
© 2012 Lippincott Williams & Wilkins
Epidemiology • Volume 24, Number 1, January 2013
In our study, the validity of the modeled exposure
data varied among the pollutants and from year to year (R2
statistics for each pollutant for years 2002–2007 are given
in the eAppendix, http://links.lww.com/EDE/A628), both in
relation to the national network sites and the smaller number of
verification sites. Although the R2 values based on verification
sites provide the strongest indication of the performance of
the model, the manner in which the national network data
are used in the calibration process does not preclude the
national network sites from providing some indication of
model performance. The difficulty in modeling PM is well
established33 because of the complexity of the PM mixture for
example. The removal of sulfur from petrol and diesel as a
result of European Union legislation and the move away from
coal to gas and electricity for domestic and industrial use have
caused SO2 levels to fall substantially in the last 50 years,
making them harder to model.32 Our validation statistics for
PM10 and SO2 were broadly similar to UK estimates derived
using land-use-regression models for PM10 concentrations
in 200133 and SO2 concentrations in 1991.32 Nonetheless,
we believe the poorer model performance for PM10 and
SO2 relative to NO2 should be a consideration in the overall
assessment of our results.
Despite the year-to-year variability in the validation
R2 statistics, our sensitivity analysis demonstrated that the
estimated HRs were robust to the choice of exposure year.
Furthermore, a time-varying-exposure survival model gave
results comparable to those derived from 2002 data alone
(data not shown). We are therefore confident in the choice of
2002 as the exposure period used for our study, despite the
lower R2 statistics compared with other years.
Misclassification of grid square pollutant concentration by the model is only one potential source of error in our
exposure assignment. Another is misclassification of personal
exposure arising from the use of exposures based solely on
residential address, rather than monitored personal exposure (from other sources such as indoor or workplace). We
acknowledge these weaknesses in our study, but note that it
is likely that this type of measurement error would bias effect
estimates toward the null.34 However, we were able to demonstrate robust associations between all pollutants and incidence
of heart failure.
Studies of PM and Cardiovascular Events
Evidence linking PM to the incidence of both fatal and
nonfatal CVD in large population cohorts has been limited
to three studies of women in the United States—the Women’s Health Initiative,20 the Nurses’ Health Study,15 and the
California Teachers Study.17 All three studies identified events
from a combination of questionnaires, medical records, and
death registration. Miller et al20 reported an increased HR for
PM10 and first cardiovascular event (MI, coronary revascularization, stroke, CHD, and cerebrovascular death) of 1.04 (95%
CI = 0.95–1.12) per 10 µg/m3. For MI, the HR for PM2.5 was
© 2012 Lippincott Williams & Wilkins
Air Pollution and Incidence of Cardiovascular Disease
1.06 (0.85–1.34), and for stroke, 1.28 (1.02–1.61) per 10 µg/
m3 increment in PM2.5. Puett et al15 reported increased HRs
for PM2.5 and PM2.5–10 for incident CHD events but with lower
CIs below 1. Lipsett et al17 reported little evidence for an association between either PM10 or PM2.5 and MI incidence (HR
0.98 [0.91–1.06] and 0.98 [0.83–1.16], respectively) but a
stronger association for both PM10 and PM2.5 and stroke (1.06
[1.00–1.13] and 1.14 [0.99–1.32], respectively) each per 10
µg/m3 increment. In our study of more than 800,000 English
adults, we found HRs for PM10 and PM2.5 with incident CHD,
MI, and stroke that were close to unity, consistent with the
rather weak evidence for MI, but inconsistent with the evidence for the stronger associations for stroke reported in the
studies referenced above.15,17,20 Of the many methodological
differences with our work, we note that the modeled levels of
particulate exposure were lower in the United Kingdom, and
that these US studies were based on women alone. Although
it has been suggested that the associations between PM and
CHD mortality are stronger in women than in men,35 we found
no evidence of this in our study.
Our study found associations between the incidence of
heart failure events and residential levels of PM10 and PM2.5
exposure. This finding extends previous evidence linking
long-term exposure to PM and mortality from heart failure.19,36 Pope et al19 analyzed data from the American Cancer
Society cohort and reported an increased HR for mortality from heart failure (including dysrhythmias and cardiac
arrest) of 1.13 (95% CI = 1.05–1.21) per 10 µg/m3 increment
in PM2.5. Beelen et al36 assessed the association between
PM2.5 and heart failure mortality in the Netherlands Cohort
Study on Diet and Cancer and reported an HR larger than
ours (2.69 [95% CI = 1.37–5.27] per 10 µg/m3 increment).
There is a more substantial body of evidence linking shortterm exposure to particulate air pollution and heart failure
mortality and morbidity.2
Results for Gases
Few studies have reported results for gaseous pollutants
and cardiovascular outcomes. Miller et al20 observed positive
associations with NO2, O3, and SO2, with CIs that included
1.0. Rosenlund et al37 reported positive associations between
NO2 and SO2 and incident MI, again with CIs that included
1.0, in a case-control study of first-time MI cases and population controls age 45 to 70 years in Stockholm county. Gan
et al16 and Rosenlund et al13 focused on traffic sources in their
studies, observing positive associations between NO2 and
CHD events with lower confidence limits close to 1.0. These
findings are broadly consistent with the general lack of evidence for an association with NO2 and O3 in our study. Our
finding of consistent associations between SO2 and increased
incidence of a range of CVDs was in line with the results from
the only other studies from the United Kingdom that assessed
long-term exposure to air pollution and disease.28,29 SO2 has
also been associated with increased numbers of deaths from
www.epidem.com | 51
Atkinson et al
all disease-related causes, cardiopulmonary causes, and lung
cancer.3,6,21 However, given the lack of biological plausibility
for health effects of low concentrations of SO2, the pollutant
may simply be an indicator for combustion sources.23
In our study, we observed either negative or null associations between O3 and CVD events. Because O3 is a regional
pollutant, variability in O3 concentrations is mainly between
regions, with little within-practice variation. Ozone is highest
in the South where disease incidence is lowest, and lowest in
the North where disease incidence is highest, and therefore,
regional differences are likely to explain the observed relationships. In addition, O3 has a pronounced seasonal pattern,
which our annual estimates do not reflect. Stronger associations have been observed in studies using summer-only concentrations.17 However, these data were unavailable to us,
which is a limitation of our study.
Adjustment for Socioeconomic Status
We presented estimates adjusted for our marker of
socioeconomic status (modified Index of Multiple Deprivation), which had the effect of moving all associations toward
the null. We think the adjustment is potentially important
because it controls for many unmeasured confounders, such
as diet and nutrition. However, because increasing deprivation
was associated with higher levels of pollution (except ozone)
in our study, there is a chance that this adjustment is removing effects of air pollution. The finer spatial resolution of the
Index of Multiple Deprivation variable in some urban areas
may represent fine-scale variability in the pollution levels not
captured in the 1 × 1 km resolution of the pollutant estimates.
Other studies have adjusted for socioeconomic status using
a variety of indicators. Some used household income and
reported little impact on pollutant associations.14,15,20 Others
have used small-area Census-based measures of deprivation
and observed increased estimates of the effects of air pollution on coronary events and mortality.13,21 Rosenlund et al13
concluded that this was likely representing an adjustment
for smoking (for which they had no data) but also noted that
in Rome the most affluent were living in the most polluted
areas, a finding also observed in the American Cancer Society
study.21 This was not the case in our study, in which PM10 concentrations generally increased with deprivation.
Effect Modification
We did not replicate other authors’ findings of effect
modification with factors such as socioeconomic status and
BMI; larger associations have been reported in the less educated15,18,38 and more obese.15,20 Instead, we tended to show
the reverse, with greater overall effects seen in those who were
least deprived and of normal weight. We attribute this to the
fact that the region that consistently showed associations (the
South) had a larger proportion of patients who were the least
deprived and of normal weight (eg, Index of Multiple Deprivation least deprived quintile: 34% of patients in the South
vs. 14% in the North). However, we also observed that the
52 | www.epidem.com
Epidemiology • Volume 24, Number 1, January 2013
overall relationship between deprivation and pollution (the
most deprived patients tended to reside in areas with higher
levels of pollution) was not apparent in the South. A Swedish
study of pollution and MI37 found larger effects in the better educated; the authors speculated that this may be due to
the fact that in Stockholm more socioeconomically privileged
people tended to live nearer the city centre. We acknowledge
that the differing nature of the relationship between socioeconomic status and pollution across England may account for
the geographical variations we observed.
Data Sources
Our study was strengthened by using three independent sources of data to identify incident events. To ensure
data quality, the Clinical Practice Research Datalink protocol
specifies internal checks on data continuity and completeness
and provides recording guidelines to all practices that contribute data to it.25 The introduction in 2004 of the Quality
and Outcomes Framework (a “pay-for-performance” contract
for primary care) further improved the recording of CVD on
primary care databases.39 In the United Kingdom, there is
complete coverage for death registrations, admissions to NHS
hospitals, and NHS-funded admissions to private hospitals.
Patients from practices consenting to data linkage (approximately 40% of current practices) are representative of all
Clinical Practice Research Datalink patients.40 Our analysis
of these data sources suggests that studies using only hospitalization or mortality records may miss some events because
of coding differences and inconsistencies among the systems
(data not shown). The misclassification of events (eg, when an
initial diagnosis by the general practitioner is not confirmed at
admission to hospital) cannot be ruled out, however.
Conclusion
Our study found evidence of an association between
long-term exposure to particulate pollution and NO2 with the
incidence of heart failure events but not with MI, arrhythmias,
or stroke. These results, from one of the few studies to be based
on a whole population sample, differ from those reported from
the United States and mainland Europe. Further research is
required to understand the reasons for these differences. Additional verification of SO2 and PM2.5 data is required to further
understand the role of these pollutants in the onset of CVD.
ACKNOWLEDGMENTS
The Clinical Practice Research Datalink (CPRD) is
owned by the Secretary of State of the UK Department of
Health and operates within the Medicine and Healthcare
Products Regulatory Agency (MHRA). CPRD has received
funding from the MHRA, Wellcome Trust, Medical Research
Council, NIHR Health Technology Assessment Programme,
Innovative Medicine Initiative, UK Department of Health,
Technology Strategy Board, Seventh Framework Programme
EU, various universities, contract research organizations, and
pharmaceutical companies.
© 2012 Lippincott Williams & Wilkins
Epidemiology • Volume 24, Number 1, January 2013
REFERENCES
1. Department of Health. Cardiovascular disease and air pollution (Committee on the Medical Effects of Air Pollutants). 2006. Available at: http://
comeap.org.uk/documents/reports.html. Accessed 15 July 2011.
2. Brook RD, Rajagopalan S, Pope CA 3rd, et al. American Heart Association Council on Epidemiology and Prevention, Council on the Kidney in
Cardiovascular Disease, and Council on Nutrition, Physical Activity and
Metabolism. Particulate matter air pollution and cardiovascular disease:
An update to the scientific statement from the American Heart Association. Circulation. 2010;121:2331–2378.
3.Dockery DW, Pope CA 3rd, Xu X, et al. An association between air
pollution and mortality in six U.S. cities. N Engl J Med. 1993;329:
1753–1759.
4. Abbey DE, Nishino N, McDonnell WF, et al. Long-term inhalable particles and other air pollutants related to mortality in nonsmokers. Am J
Respir Crit Care Med. 1999;159:373–382.
5. Lipfert FW, Perry HM Jr, Miller JP, Baty JD, Wyzga RE, Carmody SE.
The Washington University-EPRI Veterans’ Cohort Mortality Study: preliminary results. Inhal Toxicol. 2000;12(suppl 4):41–73.
6. Pope CA 3rd, Burnett RT, Thun MJ, et al. Lung cancer, cardiopulmonary
mortality, and long-term exposure to fine particulate air pollution. JAMA.
2002;287:1132–1141.
7. Hoek G, Brunekreef B, Goldbohm S, Fischer P, van den Brandt PA. Association between mortality and indicators of traffic-related air pollution
in the Netherlands: a cohort study. Lancet. 2002;360:1203–1209.
8. Enstrom JE. Fine particulate air pollution and total mortality among elderly Californians, 1973-2002. Inhal Toxicol. 2005;17:803–816.
9. Filleul L, Rondeau V, Vandentorren S, et al. Twenty five year mortality
and air pollution: results from the French PAARC survey. Occup Environ
Med. 2005;62:453–460.
10.Jerrett M, Burnett RT, Ma R, et al. Spatial analysis of air pollution and
mortality in Los Angeles. Epidemiology. 2005;16:727–736.
11.Naess Ø, Nafstad P, Aamodt G, Claussen B, Rosland P. Relation between concentration of air pollution and cause-specific mortality: fouryear exposures to nitrogen dioxide and particulate matter pollutants
in 470 neighborhoods in Oslo, Norway. Am J Epidemiol. 2007;165:
435–443.
12. Beelen R, Hoek G, van den Brandt PA, et al. Long-term effects of trafficrelated air pollution on mortality in a Dutch cohort (NLCS-AIR study).
Environ Health Perspect. 2008;116:196–202.
13. Rosenlund M, Picciotto S, Forastiere F, Stafoggia M, Perucci CA. Trafficrelated air pollution in relation to incidence and prognosis of coronary
heart disease. Epidemiology. 2008;19:121–128.
14.Puett RC, Schwartz J, Hart JE, et al. Chronic particulate exposure, mortality, and coronary heart disease in the nurses’ health study. Am J Epidemiol. 2008;168:1161–1168.
15.Puett RC, Hart JE, Yanosky JD, et al. Chronic fine and coarse particulate exposure, mortality, and coronary heart disease in the Nurses’ Health
Study. Environ Health Perspect. 2009;117:1697–1701.
16. Gan WQ, Koehoorn M, Davies HW, Demers PA, Tamburic L, Brauer M.
Long-term exposure to traffic-related air pollution and the risk of coronary heart disease hospitalization and mortality. Environ Health Perspect.
2011;119:501–507.
17. Lipsett MJ, Ostro BD, Reynolds P, et al. Long-term exposure to air pollution and cardiorespiratory disease in the California teachers study cohort.
Am J Respir Crit Care Med. 2011;184:828–835.
18.Health Effects Institute. Reanalysis of the Harvard Six Cities Study and
the American Cancer Society Study of Particulate Air Pollution and
Mortality: A Special Report of the Institute’s Particle Epidemiology
­
­Reanalysis Project. Cambridge, MA: Health Effects Institute; 2000.
19.Pope CA 3rd, Burnett RT, Thurston GD, et al. Cardiovascular mortality and long-term exposure to particulate air pollution: epidemiological
evidence of general pathophysiological pathways of disease. Circulation.
2004;109:71–77.
20.Miller KA, Siscovick DS, Sheppard L, et al. Long-term exposure to air
pollution and incidence of cardiovascular events in women. N Engl J
Med. 2007;356:447–458.
© 2012 Lippincott Williams & Wilkins
Air Pollution and Incidence of Cardiovascular Disease
21.Krewski D, Jerrett M, Burnett RT, et al. Extended follow-up and spatial
analysis of the American Cancer Society study linking particulate air pollution and mortality. Res Rep Health Eff Inst. 2009;140:5–114.
22. World Health Organisation. Health Aspects of Air Pollution with Particulate
Matter, Ozone and Nitrogen Dioxide (Report on a WHO Working Group
Meeting. 2003). Bonn, Germany. Available at: http://www.euro.who.int/__
data/assets/pdf_file/0005/112199/E79097.pdf. Accessed 19 September 2011.
23.Department of Health. Long-term exposure to air pollution: effect on
mortality (Committee on the Medical Effects of Air Pollutants). 2009.
Available at: http://www.dh.gov.uk/ab/COMEAP/DH_108151. Accessed
15 July 2011.
24.Stedman JR, Bush TJ, Vincent KJ, Baggott S. UK air quality modelling for annual reporting 2002 on ambient air quality assessment under Council Directives 96/62/EC, 1999/30/EC and 2000/69/EC (Report
AEAT/ENV/R/1564). Didcot, Oxfordshire: AEA Energy & Environment;
2003. Available at: http://uk-air.defra.gov.uk/reports/cat05/0402061100_
dd12002mapsrep1-2.pdf. Accessed 26 March 2012.
25. Medicines and Healthcare Regulatory Agency. The Clinical Practice Research Datalink. 2012. Available at: http://www.cprd.com. Accessed 7
November 2012.
26.Noble M, McLennan D, Wilkinson K, et al. The English Indices of Deprivation 2007. London, United Kingdom: Department for Communities
and Local Government; 2008.
27.Begg MD, Parides MK. Separation of individual-level and cluster-level covariate effects in regression analysis of correlated data. Stat Med.
2003;22:2591–2602.
28.Elliott P, Shaddick G, Wakefield JC, de Hoogh C, Briggs DJ. Long-term
associations of outdoor air pollution with mortality in Great Britain. Thorax. 2007;62:1088–1094.
29.Forbes LJ, Patel MD, Rudnicka AR, et al. Chronic exposure to outdoor
air pollution and diagnosed cardiovascular disease: meta-analysis of three
large cross-sectional surveys. Environ Health. 2009;8:30.
30.Jerrett M, Arain A, Kanaroglou P, et al. A review and evaluation of intraurban air pollution exposure models. J Expo Anal Environ Epidemiol.
2005;15:185–204.
31.Beelen R, Voogt M, Duyzer J, Zandveld P, Hoek G. Comparison of the
performances of land use regression modelling and dispersion modelling
in estimating small-scale variations in long-term air pollution concentrations in a Dutch urban area. Atmos Environ. 2010;44:4614–4621.
32. Gulliver J, Morris C, Lee K, Vienneau D, Briggs D, Hansell A. Land use
regression modeling to estimate historic (1962–1991) concentrations of
black smoke and sulfur dioxide for Great Britain. Environ Sci Technol.
2011;45:3526–3532.
33.Vienneau D, de Hoogh K,Beelen R, Fischer P, Hoek G, Briggs D. Comparison of land-use regression models between Great Britain and the
Netherlands. Atmos Environ. 2010;44:688–696.
34.Armstrong BG. Effect of measurement error on epidemiological studies of environmental and occupational exposures. Occup Environ Med.
1998;55:651–656.
35.Chen LH, Knutsen SF, Shavlik D, et al. The association between fatal
coronary heart disease and ambient particulate air pollution: Are females
at greater risk? Environ Health Perspect. 2005;113:1723–1729.
36.Beelen R, Hoek G, Houthuijs D, et al. The joint association of air pollution and noise from road traffic with cardiovascular mortality in a cohort
study. Occup Environ Med. 2009;66:243–250.
37. Rosenlund M, Berglind N, Pershagen G, Hallqvist J, Jonson T, Bellander
T. Long-term exposure to urban air pollution and myocardial infarction.
Epidemiology. 2006;17:383–390.
38.Jerrett M, Burnett RT, Brook J, et al. Do socioeconomic characteristics
modify the short term association between air pollution and mortality?
Evidence from a zonal time series in Hamilton, Canada. J Epidemiol
Community Health. 2004;58:31–40.
39. Carey IM, Dewilde S, Harris T, Whincup PH, Cook DG. Spurious trends
in coronary heart disease incidence: unintended consequences of the new
GP contract? Br J Gen Pract. 2007;57:486–489.
40.Gallagher AM, Puri S, van Staa TP. Linkage of the General Practice Research Database (GPRD) with other data sources (Abstract). Pharmacoepidemiol Drug Saf. 2011;20, S1–S364.
www.epidem.com | 53