Air Pollution Reduction:
Is it Real and Does it Work?
C. Arden Pope III
Mary Lou Fulton Professor of Economics
Presented at:
Frontiers in Spatial Epidemiology
Imperial College of London, Natural History Museum of London
November 5-6, 2012
Illustration for Charles Dickens’
Sketches by Boz, by
George Cruikshank, 1838
The chimney, smoke Stack, and
smoke and fog (smog) have long
played important iconic roles in the art
and literature of London.
Over London by Rail, 1870
Gustave Doré
Depicts the congested and polluted
Industrial environment of 1870 London
Coffee Stall by
Gustave Doré,
In a description of a
representative London Fog day
(April 6, 1870) Doré’s friend
Blanchard Jerrold say’s of an
experience with Doré:
“We were well into the morning—
and it was as dark as the darkest
midnight. . . It was choking: it
made the eyes ache.”
From:
London: A pilgrimage, 1872 and
Doré’s London: All 180
illustrations from London, A
Pilgrimage, 2004
Claude
Monet,
Houses of
Parliament,
London,
1903
Claude
Monet,
Houses of
Parliament,
London,
1904
Claude Monet, Waterloo Bridge, London, 1903
Claude Monet, Waterloo Bridge, London, 1903—
Zoom in on background of pollution
Classical Economics
• Adam Smith— Wealth of Nations (1776),
free markets, competition, and the
metaphor of the “invisible hand”.
• “Laissez-faire”
• What about air pollution? Smells like
money to me!
Meuse Valley, Belgium Smog, December 1-5, 1930
--60 deaths, 10 time expected
Donora, PA
(Severe episode, Oct. 27-31, 1948)
From, Public Health Service,
Bulletin No. 306, 1949
Danora, PA. Noon, Oct 29, 1948.
http://www.weather.com/multimedia/videoplayer.html?clip=12603&collection=257&fro
Donora, PA—administering oxygen
--Approx. 20 deaths, 6 times expected
London, England (Dec. 5-9, 1952)
London, England (Dec. 5-9, 1952)
London Fog Episode, Dec. 1952
From: Brimblecombe P. The Big Smoke, Methuen 1987
Loss of Faith in the doctrine of Laissez-faire
• No longer did air pollution “smell like money” but it
smelled like something unhealthy.
• In the UK, US, and elsewhere there were public policy
efforts to clean up the air and eliminate the “Killer Smog
Episodes.” (Clean air act)
• There was success at eliminating or reducing the
extreme air pollution episodes but air pollution remained
and we learned that even moderate air pollution had
adverse health effects.
Brief methodological outline of air
pollution epidemiological studies:
Studies of short-term exposure
 Episode
 Population-based daily time-series
 Panel-based acute exposure
 Case-crossover
Studies of long-term exposure
 Population-based cross-sectional
 Cohort-based mortality
 Cohort- and panel-based morbidity
 Case-control studies
Intervention/natural experiment
Exploit
temporal
variability
Exploit
spatial
variability
Exploit both
temporal and
spatial
variability
Brief methodological outline of air
pollution epidemiological studies:
Studies of short-term exposure
 Episode
 Population-based daily time-series
 Panel-based acute exposure
 Case-crossover
Studies of long-term exposure
 Population-based cross-sectional
 Cohort-based mortality
 Cohort- and panel-based morbidity
 Case-control studies
Intervention/natural experiment
Exploit
temporal
variability
Exploit
spatial
variability
Exploit both
temporal and
spatial
variability
Adjusted Mortality for 1980 (Deaths/Yr/100,000)
Age-, sex-, and race- adjusted populationbased mortality rates in U.S. cities for 1980
plotted over various indices of particulate
air pollution (From Pope 2000).
1000
950
900
850
800
750
700
650
600
8
10
12
14
16
18
20
22
24
26
28
Median PM2.5 for aprox. 1980
30
32
34
Brief methodological outline of air
pollution epidemiological studies:
Studies of short-term exposure
 Episode
 Population-based daily time-series
 Panel-based acute exposure
 Case-crossover
Studies of long-term exposure
 Population-based cross-sectional
 Cohort-based mortality
 Cohort- and panel-based morbidity
 Case-control studies
Intervention/natural experiment
Exploit
temporal
variability
Exploit
spatial
variability
Exploit both
temporal and
spatial
variability
An Association Between Air Pollution and
Mortality in Six U.S. Cities
1993
Dockery DW, Pope CA III, Xu X, Spengler JD,
Ware JH, Fay ME, Ferris BG Jr, Speizer FE.
Methods:
14-16 yr prospective follow-up of 8,111 adults living in
six U.S. cities.
Monitoring of TSP PM10, PM2.5, SO4, H+, SO2, NO2, O3 .
Data analyzed using survival analysis, including
Cox Proportional Hazards Models.
Controlled for individual differences in: age, sex, smoking,
BMI, education, occupational exposure.
Clean
cities
Average
Polluted
cities
Highly
Polluted
cities
Cox Proportional Hazards Survival Model
Cohort studies of outdoor air pollution have commonly used the CPH Model
to relate survival experience to exposure while simultaneously controlling for
other well known mortality risk factors. The model has the form
 (t )   (t ) exp  x (t )
(l )
i
Hazard function
or instantaneous
probability of
death for the ith
subject in the lth
strata.
(l )
0
Baseline
hazard
function,
common to all
subjects within
a strata.
T
(l )
i
Regression equation that
modulates the baseline
hazard. The vector Xi(l)
contains the risk factor
information related to the
hazard function by the
regression vector β which
can vary in time.
Adjusted risk ratios (and 95% CIs) for
cigarette smoking and PM2.5
Cause of
Death
All
Lung
Cancer
Cardiopulmonary
All
other
Current Smoker,
25 Pack years
Most vs. Least
Polluted City
2.00
1.26
(1.51-2.65)
(1.08-1.47)
8.00
1.37
(2.97-21.6)
(0.81-2.31)
2.30
1.37
(1.56-3.41)
(1.11-1.68)
1.46
1.01
(0.89-2.39)
(0.79-1.30)
Particulate Air Pollution as a Predictor of Mortality in a
Prospective Study of U.S. Adults
Pope CA III, Thun MJ, Namboodiri MM,
Dockery DW, Evans JS, Speizer FE, Heath CW Jr.
Michael Thun
Methods: Linked and
analyzed ambient air
pollution data from 51151 U.S. metro areas
with risk factor data for
over 500,000 adults
enrolled in the ACSCPSII cohort.
1995
Clark Heath
Adjusted mortality risk ratios (and 95% CIs) for cigarette
smoking the range of sulfates and fine particles
Cause of
Death
All
Lung
Cancer
CardioPulmonary
All other
Current
Smoker
2.07
Sulfates
1.15
Fine
Particles
1.17
(1.75-2.43)
(1.09-1.22)
(1.09-1.26)
9.73
1.36
1.03
(5.96-15.9)
(1.11-1.66)
(0.80-1.33)
2.28
1.26
1.31
(1.79-2.91)
(1.16-1.37)
(1.17-1.46)
1.54
1.01
1.07
(1.19-1.99)
(0.92-1.11)
(0.92-1.24)
Legal uncertainty largely
resolved with 2001
unanimous ruling by the
U.S. Supreme Court.
Dan Krewski
Rick Burnett
Mark Goldberg
and 28 others
Cox Proportional Hazards Survival Model
Cohort studies of outdoor air pollution have commonly used the CPH Model
to relate survival experience to exposure while simultaneously controlling for
other well known mortality risk factors. The model has the form
 (t )   (t ) exp  x (t )
(l )
i
Hazard function
or instantaneous
probability of
death for the ith
subject in the lth
strata.
(l )
0
Baseline
hazard
function,
common to all
subjects within
a strata.
T
(l )
i
Regression equation that
modulates the baseline
hazard. The vector Xi(l)
contains the risk factor
information related to the
hazard function by the
regression vector β which
can vary in time.
Note: We have extended this basic model to deal with random effects (spatial
clustering), spatial autocorrelation, and allow for non-parametric spatial smoothing.
Further extension of the Cox Proportional Hazards model to allow for random
effects (spatial clustering) and to include spatial socioeconomic and
environmental (contextual) variables at two spatial levels (zip code and MSA).
Contextual variables included: Poverty (%), Income disparity (Gini), Median
Household income, Unemployment (%), Not white (%), Grade 12 education (%),
Air conditioning (%).
Results: Contextual variables do no reduce the PM2.5-mortality effect estimates
Comparisons of PM2.5-mortality effects estimate over time with improving methods.
Sources
Data/ periods
Model
Strata
Individual
risk factor
variables
Lave/Seskin
1977,
Ozkaynak/
Thurston
1987
Pope et al.
AJRCCM 1995
Krewski et al.
HEI 2000
U.S. pop.based
mortality
rates
OLS or WLS NA
cross-sectional
regression
models
None
ACS CPS-II
cohort/
1982-1989
Standard
Age,
Cox
Sex,
proportional
Race
hazards model
Pope et al.
JAMA 2002
ACS CPS-II
cohort/
1982-1998
Random
Age,
Effects
Sex,
Cox
Race
proportional
hazards model
Krewski et al.
HEI 2009
ACS CPS-II
cohort/
1982-2000
Random
Age,
Effects
Sex,
Cox
Race
proportional
hazards model
smoking,
education,
marital status,
BMI, alcohol,
dust/fume
exposure
smoking,
education,
marital status,
BMI, alcohol,
occupation/
occupational
exposure, diet
smoking,
education,
marital status,
BMI, alcohol,
occupation/
occupational
exposure, diet
Ecological/contextual
spatial variables
Spatial
smoothing?
% increase/10 µg/m3
AllCardioCause
pulmonary
Various including:
Population, population
density, percent elderly,
none white, poverty
levels, etc.
None
No
~9
----
No
7 (4-10)
12 (7-17)
None
Yes
6 (2-11)
9 (3-16)
Poverty,
income disparity,
household income,
unemployment,
none white, education, air
conditioning—at two
levels of spatial scale (Zip
Code, Metro)
No
6 (4-8)
13 (10-16)
70
60
50
40
30
20
10
0
-10
90
Geman women (Gehring et al. 2006)
French PAARC (Filleul et al. 2005)
11 CA county, elderly (Enstrom 2005)
VA, Hypertensive, Males (Lipfert et al. 2006)
AHSMOG, Males (McDonnell et al. 2000)
U.S. Medicare; East/Central/West (Zegar et al. 2008)
ACS Medicare (Efim et al. 2008)
ACS
Nurses' Health Study (Puett et al. 2008)
ACS
AHSMOG, Female (Chen et al. 2005)
Women's Health Initiative (Miller et al. 2007)
Nurses' Health Study (Puett et al. 2008)
ACS
Cal. Teachers Study (Ostro et al. 2009)
CVD
LA
Dutch Cohort (Beelen et al. 2008)
Oslo, men and women, 71-90 (Naess et al. 2007)
CPD
Oslo, men and women, 51-70 (Naess et al. 2007)
Six Cities
Other studies
Geman women (Gehring et al. 2006)
110
Cal. Teachers Study (Ostro et al. 2009)
French PAARC (Filleul et al. 2005)
AHSMOG, Males (McDonnell et al. 2000)
Six Cities
Cal. Teachers Study (Ostro et al. 2009)
All Cause
ACS
100
Pope et al. 1995; Krewski et al. 2000; Pope et al.
2002, 2004; Jerrett et al. 2005; Krewski et al. 2009)
Six-Cities
studies
ACS studies
LA
80
Six Cities (Dockery et al 1993;
Krewski et al. 2000; Laden et al. 2008)
Six Cities, Medicare (Efim et al. 2008)
180
120
Dutch Cohort (Beelen et al. 2008)
LA
Percent increase in mortalily risk (95% CI)
Other cohort studies have shown associations between exposure
to fine PM and increased risk of cardiovascular death.
190
IHD
Women’s Health Initiative Study
190
All Cause
180
120
110
Percent increase in mortalily risk (95% CI)
100
90
80
70
60
50
Lianne Sheppard
40
30
20
10
0
-10
Joel Kaufman
CPD
CVD
IHD
All Cause
CPD
190
180
120
110
Percent increase in mortalily risk (95% CI)
100
90
Nurses’ Health Study:
•Puett et al. Am. J. Epidemiology 2008
Stronger association with CVD than
with all cause.
80
Frank Speizer
70
60
50
40
30
20
10
0
-10
CVD
IHD
Lung Cancer
All Cause
CPD
190
Netherlands, Germany, Norway studies:
180 Beelen et al. EHP 2008
120 Gehring et al. Epidemiology 2006
Naess et al. Am. J. Epidemiology 2007
110
Percent increase in mortalily risk (95% CI)
Again, positive associations, generally
100 Stronger for cardiovascular disease.
90 Brunekreef (summary paper)
JESEE 2007
80
Bert Brunekreef
70
60
50
40
30
20
10
0
-10
CVD
IHD
Lung Cancer
U.S. Medicare Cohort Studies
190
All Cause
CPD
CVD
180
120
110
Percent increase in mortalily risk (95% CI)
100
U.S. Medicare Cohort studies:
•Eftim et al. Epidemiology 2008
•Zegar et al. EHP 2008
90
Cohorts of Medicare participants cities of the 6-cities and ACS study, plus all U.S.
80
70
60
50
40
30
20
10
0
-10
IHD
100
90
80
Survival Probability
70
60
50
Survival curves based on projected U.S life tables
and alternative excess risk assumptions
40
Baseline survival curve, life expect.=76.4
1.25 RR from Pollut, ages < 45 unaffected; life expect.=73.9
1.25 RR from Pollut., infants unaffected; life expect.=73.5
1.25 RR from Pollut., infants affected; life expect. = 73.4
2.00 RR from Smoking from age 20; life expect.=67.9
30
20
Note: 1.15 RR from Polut., infants uneffected; life expect=74.6
(curve not shown)
10
0
0
5
10
15
20
25
30
35
40
45
50
55
60
65
70
75
80
85
90
95
100 105
Age
Brunekreef 1997
Pope 2000
Brief methodological outline of air
pollution epidemiological studies:
Studies of short-term exposure
 Episode
 Population-based daily time-series
 Panel-based acute exposure
 Case-crossover
Studies of long-term exposure
 Population-based cross-sectional
 Cohort-based mortality
 Cohort- and panel-based morbidity
 Case-control studies
Intervention/natural experiment
Exploit
temporal
variability
Exploit
spatial
variability
Exploit both
temporal and
spatial
variability
Southern California Children’s Health Study
Effects of air pollution on
children’s health, especially
lung function growth.
David Bates, Advisor
W. James Gauderman
John Peters
Southern California Children’s Health Study, has shown that
air pollution impacts lung development in children.
Gauderman et al. 2007
Children living in cities with higher air pollution and
living near major traffic sources showed greater
deficits in lung function growth.
AJRCCM 2001
Brief methodological outline of air
pollution epidemiological studies:
Studies of short-term exposure
 Episode
 Population-based daily time-series
 Panel-based acute exposure
 Case-crossover
Studies of long-term exposure
 Population-based cross-sectional
 Cohort-based mortality
 Cohort- and panel-based morbidity
 Case-control studies
Intervention/natural experiment
Exploit
temporal
variability
Exploit
spatial
variability
Exploit both
temporal and
spatial
variability
Follow up analysis of the Six-Cities cohort:
Air pollution reductions
reduced risk of death
Kingston
Steubenville
Topeka
Portage
St. Louis
Watertown
Reduction in fine particulate air pollution:
Extended follow-up of the Harvard Six Cities Study
Laden, Schwartz,
Speizer, Dockery, 2006
Copper Smelter Strike, Mid July 1967 – March 1968
Adapted from: Trijonis, John.
Atmospheric Environment 1979
100
90
80
Survival Probability
70
60
50
Survival curves based on projected U.S life tables
and alternative excess risk assumptions
40
Baseline survival curve, life expect.=76.4
1.25 RR from Pollut, ages < 45 unaffected; life expect.=73.9
1.25 RR from Pollut., infants unaffected; life expect.=73.5
1.25 RR from Pollut., infants affected; life expect. = 73.4
2.00 RR from Smoking from age 20; life expect.=67.9
30
20
Note: 1.15 RR from Polut., infants uneffected; life expect=74.6
(curve not shown)
10
0
0
5
10
15
20
25
30
35
40
45
50
55
60
65
70
75
80
85
90
95
100 105
Age
Brunekreef 1997
Pope 2000
Do cities with bigger improvements in air quality have bigger
improvements in health, measured by life expectancy?
January 22, 2009
Fine-Particulate Air Pollution and Life
Expectancy in the United States
C. Arden Pope, III, Ph.D., Majid Ezzati, Ph.D., and Douglas W.
Dockery, Sc.D.
- Matching PM2.5 data for
1979-1983 and 1999-2000 in
51 Metro Areas
- Life Expectancy data for
1978-1982 and 1997-2001 in
211 counties in 51 Metro areas
- Evaluate changes in Life
Expectancy with changes in
PM2.5 for the 2-decade period
of approximately 1980-2000.
Covariates included in the regression models
Changes in socio-economic and demographic variables (from U.S.
Census Data):
Per capita income
Population
5-yr in-migration
High-school graduates
Urban population
Black proportion of population
Hispanic proportion of population
Proxy cigarette smoking variables—available for all 211 counties
COPD mortality rates
Lung Cancer mortality rates
Survey-based metro-area estimates of smoking prevalence
National Health Interview Survey (1978-1980)
Behavioral Risk Factor Surveillance System (1998-2000)
Matching data available for only 24 of 51 metro areas
Clustered standard errors (clustered by the 51 metro areas) were estimated for all models
except for analysis that included only the 51 largest counties in each metro area.
A 10 µg/m3 decrease in PM2.5 was associated with
a 7.3 (± 2.4) month increase in life expectancy.
This increase in life expectancy persisted even after controlling for
socio-economic, demographic, or smoking variables
Residual changes in LE controlling for covariates
2.5
B
2.0
1.5
1.0
20
0.5
26
23
10
2134
42
5111
38
41 49
13
24
3 36
33 927
14
6 39
50
43
219
16
22
29
4
12
30
1
7
35
18
0.0
3240
48
-0.5
37
8
31
25
28
44
17
47
45
-1.0
-1.5
5
46
15
-2.0
-2.5
0
2
4
6
8
10
Reduction in PM2.5, 1980-2000
12
14
Air Pollution Reduction:
Is it Real and Does it Work?
Yes