Research
A Section 508–conformant HTML version of this article
is available at http://dx.doi.org/10.1289/ehp.1409111.
Low-Concentration PM2.5 and Mortality: Estimating Acute and Chronic
Effects in a Population-Based Study
Liuhua Shi,1 Antonella Zanobetti,1 Itai Kloog,1,2 Brent A. Coull,3 Petros Koutrakis,1 Steven J. Melly,1 and
Joel D. Schwartz1
1Department
of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA; 2Department of
Geography and Environmental Development, Ben-Gurion University of the Negev, Beer Sheva, Israel; 3Department of Biostatistics,
Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
Background: Both short- and long-term exposures to fine particulate matter (≤ 2.5 μm; PM2.5)
are associated with mortality. However, whether the associations exist at levels below the new U.S.
Environmental Protection Agency (EPA) standards (12 μg/m3 of annual average PM2.5, 35 μg/m3
daily) is unclear. In addition, it is not clear whether results from previous time series studies (fit in
larger cities) and cohort studies (fit in convenience samples) are generalizable.
Objectives: We estimated the effects of low-concentration PM2.5 on mortality.
Methods: High resolution (1 km × 1 km) daily PM2.5 predictions, derived from satellite aerosol
optical depth retrievals, were used. Poisson regressions were applied to a Medicare population
(≥ 65 years of age) in New England to simultaneously estimate the acute and chronic effects of
exposure to PM2.5, with mutual adjustment for short- and long-term exposure, as well as for
area-based confounders. Models were also restricted to annual concentrations < 10 μg/m3 or daily
concentrations < 30 μg/m3.
Results: PM2.5 was associated with increased mortality. In the study cohort, 2.14% (95% CI:
1.38, 2.89%) and 7.52% (95% CI: 1.95, 13.40%) increases were estimated for each 10-μg/m3
increase in short- (2 day) and long-term (1 year) exposure, respectively. The associations held for
analyses restricted to low-concentration PM2.5 exposure, and the corresponding estimates were
2.14% (95% CI: 1.34, 2.95%) and 9.28% (95% CI: 0.76, 18.52%). Penalized spline models of
long-term exposure indicated a larger effect for mortality in association with exposures ≥ 6 μg/m3
versus those < 6 μg/m3. In contrast, the association between short-term exposure and mortality
appeared to be linear across the entire exposure distribution.
Conclusions: Using a mutually adjusted model, we estimated significant acute and chronic effects
of PM2.5 exposure below the current U.S. EPA standards. These findings suggest that improving
air quality with even lower PM2.5 than currently allowed by U.S. EPA standards may benefit
public health.
Citation: Shi L, Zanobetti A, Kloog I, Coull BA, Koutrakis P, Melly SJ, Schwartz JD. 2016.
Low-concentration PM2.5 and mortality: estimating acute and chronic effects in a population-based
study. Environ Health Perspect 124:46–52; http://dx.doi.org/10.1289/ehp.1409111
Introduction
Many studies have found associations between
fine particulate matter [PM with aerodynamic
diameter ≤ 2.5 μm (PM2.5)] and increased
mortality (Dockery et al. 1993; Franklin
et al. 2007; Pope et al. 2002; Schwartz 1994;
Zanobetti and Schwartz 2009). Biological
evidence has been established for plausible
mechanisms between PM2.5 and mortality,
such as increased risk of ventricular arrhythmia
and thrombotic processes, increased system
inflammation and oxidative stress, increased
blood pressure, decreased plaque stability, and
reduced lung function, among others (Brook
et al. 2009; Gauderman et al. 2004; Gurgueira
et al. 2002; Suwa et al. 2002; Yue et al. 2007).
Based on evidence from epidemiological and
toxicological studies (Chen and Nadziejko
2005; Furuyama et al. 2006; Ohtoshi et al.
1998), National Ambient Air Quality
Standards (NAAQS) were implemented for
fine particulate matter. For example, the U.S.
Environmental Protection Agency (EPA)
revised the fine particle NAAQS in 1997,
2006, and 2012 in order to protect public
46
health (U.S. EPA 1997, 2006, 2013). Further
changes in the standards require additional
studies to elucidate whether health effects
occur at levels below the current annual and
daily U.S. EPA NAAQS of 12 and 35 μg/m3,
respectively. The Clean Air Act Amendments
of 1990 require the U.S. EPA to review
national air quality standards every 5 years to
determine whether they should be retained
or revised; thus, whether health effects can
be observed below the current standards is of
great interest and importance.
Previous studies have generally focused
on either long-term (Hart et al. 2011; Jerrett
et al. 2005; Puett et al. 2009; Schwartz
2000) or short-term (Dominici et al. 2006;
Katsouyanni et al. 1997; Samoli et al. 2008;
Schwartz and Dockery 1992) exposures
across the entire range of PM2.5 concentrations. In the case of time series analyses of
short-term exposures, the need to ensure the
relevance of the monitoring data as well as
the need to have a study population of a size
for sufficent power has limited analyses to
large cities; hence, exurbs, small cities, and
volume
rural areas are not generally represented in
the literature, which may compromise the
generalizability of the results. In addition,
there is spatial variability in PM2.5 concentrations within cities that time series studies
generally do not take into account, which
can introduce exposure measurement error
(Laden et al. 2006; Lepeule et al. 2012).
Chronic effects studies began using
comparisons across cities of mortality experiences of cohorts living in various communities
and the monitored air pollutant concentrations in those communities (Dockery et al.
1993; Pope et al. 1995). Again, these studies
suffered from exposure error due to failure
to capture within-city spatial variability in
exposure. Because the geographic exposure
gradient is the exposure contrast in these
studies, the failure to capture within-city
contrasts leads to classical measurement error
with expected downward bias. Studies with,
for example, land use regression estimates of
exposure have generally reported larger effect
sizes (Miller et al. 2007; Puett et al. 2009).
Previous cohort studies have not controlled
for the acute effects of particles when estimating chronic effects, raising the question of
whether there are independent chronic effects
that represent more than the cumulative
effects of acute responses.
In general, existing study cohorts are not
representative of the overall population. For
example, the American Cancer Society (ACS)
cohort has a higher level of education than
the U.S. population as a whole (Stellman
Address correspondence to L. Shi, Department of
Environmental Health, Harvard T.H. Chan School
of Public Health, Landmark Center, 401 Park Dr.,
Boston, MA 02215 USA. Telephone: (339) 2218486. E-mail: lis678@mail.harvard.edu
We thank P. Liu and Y. Wang for fruitful discussions.
This study was funded by the National Institute of
Environmental Health Sciences/National Institutes
of Health grant ES000002. This publication was also
made possible by U.S. Environmental Protection
Agency (EPA) grant RD-83479801.
The contents of this publication are solely the responsibility of the grantee and do not necessarily represent
the official views of the U.S. EPA. Further, the U.S.
EPA does not endorse the purchase of any commercial
products or services mentioned in the publication.
The authors declare they have no actual or potential
competing financial interests.
Received: 22 August 2014; Accepted: 13 May 2015;
Advance Publication: 3 June 2015; Final Publication:
1 January 2016.
124 number 1 January 2016 • Environmental Health Perspectives
Low-concentration PM2.5 and mortality
and Garfinkel 1986). Hence, few populationbased cohort studies have been conducted
until recently (Kloog et al. 2013).
Several time series studies examined the
concentration–response relationship between
PM2.5 and mortality below concentrations of
100 μg/m3; these studies generally reported
a linear concentration–response relationship
(Samoli et al. 2008; Schwartz and Zanobetti
2000). However, there have been few studies
focusing on exposures below the current daily
U.S. EPA standard of 35 μg/m3.
Many studies have examined the shape of
the concentration–response curve for longterm exposure versus short-term exposure, but
in general, they have not covered populationbased cohorts, or have only included very
low exposures (Schwartz et al. 2008; Crouse
et al. 2012).
We recently presented a new hybrid
method of assessing temporally and spatially
resolved PM2.5 exposure for epidemiological
studies by combining 1 km × 1 km resolution satellite-retrieved aerosol optical depth
(AOD) measurements with traditional land
use terms, meteorological variables, and
their interactions (Kloog et al. 2014a). This
approach allows for predicting daily PM2.5
concentrations at a 1 km × 1 km spatial resolution throughout the New England area of
the northeastern United States. We also validated our model’s performance in rural areas:
10-fold cross-validation (CV) of our model
in rural areas (using the IMPROVE stations)
resulted in a CV R2 of 0.92. Further details
have been published (Kloog et al. 2014a).
The present study aimed to simultaneously estimate acute and chronic health effects
of PM 2.5 in a population-based Medicare
cohort (≥ 65 years of age) encompassing the
New England region. We used high-spatialresolution exposure estimates based on satellite measurements that are available across
the region and not just in limited locations.
To make this study relevant to future assessments of current U.S. EPA standards, we
repeated the analysis after restricting the data
to long-term exposures (365-day moving
average) < 10 μg/m3 and repeated the time
series analysis of short-term exposures after
restricting the data to 2-day average exposures
< 30 μg/m3.
Methods
Study domain. The spatial domain of our
study included the New England area,
comprising the states of Connecticut, Maine,
Massachusetts, New Hampshire, Rhode
Island, and Vermont (Figure 1A).
Exposure data. A 3-stage statistical
modeling approach for predicting daily PM2.5
was previously reported incorporating AOD
and land use data for the New England region
(Kloog et al. 2011). Previous studies have
shown that using actual physical measurements in our prediction models improved
predictive accuracy over that of comparable land use or spatial smoothing models
(Kloog et al. 2011). With AOD retrieved by
the multi-angle implementation of atmospheric correction (MAIAC) algorithm, a
similar approach was applied for estimating
daily PM 2.5 exposures in New England at
a spatial resolution of 1 km × 1 km (Kloog
et al. 2014a). In this study, the same PM2.5
exposure predictions were employed.
Briefly, we calibrated the AOD–PM2.5
relationship on each day of the study period
(2003–2008) using data from grid cells with
both ground PM 2.5 monitors and AOD
measurements (stage 1), and we used inverse
probability weighting to address selection bias
due to nonrandom missingness patterns in
the AOD measurements. We then used the
AOD–PM2.5 relationship to predict PM2.5
concentrations for grid cells that lacked
monitors but had available AOD measurement data (stage 2). Finally, we used a generalized additive mixed model (GAMM) with
spatial smoothing and a random intercept for
each 1 km × 1 km grid cell to impute data for
grid cells/days for which AOD measurements
were not available (stage 3). The performance
of the estimated PM 2.5 was validated by
10-fold cross-validation. High out-of-sample
R2 (R2 = 0.89, year-to-year variation 0.88–
0.90 for the years 2003–2008) was found
for days with available AOD data. Excellent
performance held even in cells/days with no
available AOD (R2 = 0.89, year-to-year variation 0.87–0.91 for the years 2003–2008).
The 1-km model had better spatial (0.87)
A1 roads
PM2.5 (µg/m3)
Average PM2.5
0–0.5
1.01–1.95
0.6–1.3
1.96–2.54
1.4–2
2.55–3.08
2.1–2.6
3.09–3.57
3.58–4.11
4.12–4.75
2.7–3.3
0 10 20
3.4–3.9
4.76–5.34
4–4.5
5.35–5.94
4.6–5.1
5.95–6.53
5.2–5.9
6.54–7.12
6–12.8
40 kilometers
Boston
7.13–7.76
Springfield
7.77–8.4
8.41–8.99
9–9.63
9.64–10.2
10.3–10.7
10.8–11.1
11.2–11.6
11.7–12.1
0
100
200 kilometers
12.2–13.6
New York
Figure 1. (A) Mean PM2.5 concentrations in 2004 at a high resolution (1 km × 1 km) across New England predicted by the AOD models. (B) Predicted PM2.5 concentrations at a 1 km × 1 km grid for 15 November 2003.
Environmental Health Perspectives • volume 124 number 1 January 2016
47
Shi et al.
and temporal (0.87) out-of-sample R2 than
the previous 10-km model (0.78 and 0.84,
respectively). Details of the PM2.5 prediction
models are in Kloog et al. (2014a).
Figure 1A shows an example of mean
PM2.5 concentrations in 2004 at a 1 km × 1 km
spatial resolution across New England. By
averaging the estimated daily exposures at each
location, we generated long-term exposures.
Figure 1B (a subset of the study area)
shows that spatial variability existed even for
daily data and was not identical to the longterm pattern shown in Figure 1A. That is,
there was space–time variation in the PM2.5
exposure captured in this analysis, but not in
previous time-series analyses.
Because the deaths were coded at the ZIP
code level, both long- and short-term predictions were matched to ZIP codes by using
ArcGIS (ESRI, Redlands, CA) and SAS (SAS
Institute Inc., Cary, NC) to link the ZIP code
centroid to the nearest PM2.5 grid.
Traditionally, studies of acute air pollution
effects have controlled for temperature using
values taken from the nearest airport. This
approach is not feasible for the entire region
because many residences are distant from
airports. In addition, there is spatiotemporal
variation in temperature. We have applied a
similar 3-stage statistical modeling approach
to estimate daily ambient temperature at 1 km
× 1 km resolution in New England using
satellite-derived surface temperature (Kloog
et al. 2014b). To our knowledge, such fine
control for temperature has not previously
been used in air pollution epidemiology.
Mortality data. Individual mortality
records were obtained from the U.S. Medicare
program for all residents ≥ 65 years of age for
all available years during 2003–2008 (CMS
2013b). The Medicare cohort was used because
of the availability of ZIP code of residence
data, whereas National Center for Health
Statistics mortality data are only available at
the county level. Additionally, previous studies
found that elderly people are highly susceptible to the effects of particulate matter (Pope
2000). The Medicare beneficiary denominator file from the Centers for Medicare and
Medicaid services (CMS 2013a) lists all beneficiaries enrolled in the Medicare fee-for-service
(FFS) program and contains information on
beneficiaries’ eligibility and enrollment in
Medicare and the date of death. The Medicare
Provider Analysis and Review (MEDPAR) file
includes information on age, sex, race, ZIP
code of residence, and one record for each
hospital admission (CMS 2013c).
Daily mortality was first aggregated by
ZIP code and then matched with the corresponding PM2.5 exposure. We summarized the
mortality data by ZIP code and day because
that was the finest resolution we could obtain
for addresses. Because the mortality data sets
48
did not include changes of residence, we
assumed that the subjects lived at their current
address over the entire study period.
Covariates. We used daily 1-km temperature data estimated from surface temperature
measured by satellites (Kloog et al. 2014b).
All socioeconomic variables were obtained
through the U.S. Census Bureau 2000 Census
Summary File 3, which includes social,
economic, and housing characteristics (U.S.
Census Bureau 2000). ZIP code tabulation
area–level socioeconomic variables, including
race, education, and median household
income, were used. The county-level percentage
of people who currently smoke every day,
obtained from the CDC Behavioral Risk Factor
Surveillance survey for the entire country, was
also adjusted (CDC 2013). Dummy variables
were used to control for day of the week.
Statistical models. Conventionally, the
acute effects of air pollution are estimated by
Poisson log-linear models, and the chronic
effects of air pollution are estimated by Cox
proportional hazard models (Kloog et al.
2013; Laden et al. 2006). Laird and Olivier
(1981) noted the equivalence of the likelihood
of a proportional hazard model with piecewise
constant hazard for each year of follow-up and
a Poisson regression with a dummy variable
for each year of follow-up. We have taken
advantage of this equivalence to generalize
from dummy variables for each year to a
spline of time to represent the baseline hazard
and to aggregate subjects into counts per
person time at risk, and we obtained a mixed
Poisson regression model (Kloog et al. 2012).
This approach allows the rate of death as a
function of both long- and short-term exposures to be modeled simultaneously. By doing
so, we achieved the equivalence of a separate
time series analysis for each ZIP code, greatly
reducing the exposure error in that part of
the model, while simultaneously conducting
a survival analysis on the participants, and we
were also able to estimate the independent
effects of both exposures.
Most time series studies have reported
stronger associations with acute exposures
when exposures were defined as the mean
PM2.5 on the day of death and the previous
day (lag01) than when they were defined as
the mean PM2.5 on the current day only, or
for exposures with longer lags (Schwartz et al.
1996; Schwartz 2004). We used the lag01
average for our main analysis but performed
a sensitivity analysis on that choice. Longterm exposure was calculated as the 365-day
moving average ending on the date of death
so that our results were comparable with
those of previous studies (Lepeule et al. 2012;
Schwartz et al. 2008). Short-term exposure
was defined as the difference between the
2-day average and the long-term average,
ensuring that acute and chronic effects were
volume
independent. We subtracted the long-term
average from the short-term average to avoid
collinearity issues and to ensure that differences between ZIP codes in PM2.5 at a given
time did not contribute to the short-term
effect estimate. Thus, the short-term effect
could not be confounded by variables that
differed across ZIP codes.
Specifically, we fit a Poisson survival
analysis with a logarithmic link function and
a log (population) offset term and modeled
the expected daily death counts (μit) in the ith
ZIP code on the tth day as follows:
log(μit) = λi + β1PMit + β2∆PMit
+ λ(t) + temporal covariates
+ spatial covariates + offset,
[1]
where λi is a random intercept for each ZIP
code, PMit is the 365-day moving average
ending on day t in ZIP code i, ∆PMit is the
deviation of the 2-day average from its longterm average (PMit) in ZIP code i, λ(t) is a
smooth function of time, temporal covariates are temperature and day of the week,
and spatial covariates are socioeconomic
factors defined at the ZIP code level (percent
of people without high school education,
percent of white people, median household
income) and smoking data at the county
level. Additionally, a quasi-Poisson model was
used to control for possible overdispersion
(Ver Hoef and Boveng 2007).
We estimated λ(t) with a natural cubic
spline with 5 degrees of freedom (df) per
year to control for time and season trends.
The specific temporal and spatial covariates
that we used were a natural cubic spline for
temperature with 3 df in total; a categorical
variable for day of the week; linear variables
for percent of people without high school
education, percent of white people, median
household income, and percent of people
who currently smoke every day.
The number of deaths per ZIP code area
over the study period (2003–2008) averaged
319 with a standard deviation of 430. Because
the outcome was counts, we could not adjust
for age and sex as in a Cox model. Instead, we
adjusted for variables that varied by ZIP code.
The analyses were repeated without mutual
adjustment for short- and long-term PM2.5.
We modeled the association between allcause mortality and PM2.5 at low doses in
which the person-time at risk in each year of
follow-up in each ZIP code was used as the
offset. We also conducted effect modification
by population size by choosing the median
(4,628) of the ZIP code–level total population
as the cutoff between urban and rural areas.
Estimating the effects of low-level PM2.5.
For full cohort analyses with 10,938,852
person-years of follow-up, all observed
deaths were used. To estimate effects at low
124 number 1 January 2016 • Environmental Health Perspectives
Low-concentration PM2.5 and mortality
Results
Table 1 presents a summary of the predicted
exposures for both short- and long-term PM2.5
exposure across all grid cells in the study area.
Table 2 presents the estimated percent
change in all-cause mortality with 95% CIs for
a 10-μg/m3 increase in both short- and longterm PM2.5 in the restricted and full cohort.
In the restricted population, we found an estimated 9.28% increase in mortality (95% CI:
0.76, 18.52%) for every 10-μg/m3 increase in
long-term PM2.5 exposure. A 2.14% increase
in mortality (95% CI: 1.34, 2.95%) was
observed for every 10-μg/m3 increase in shortterm PM2.5 exposure. For long-term exposure,
the effect estimates were smaller when higher
pollution days were included (7.52%; 95% CI:
1.95, 13.40%), suggesting larger effects
between low-concentration long-term PM2.5
and mortality.
Without mutual adjustment, lower estimates were found for both acute and chronic
effects than for those with mutual adjustment.
In full-cohort analyses, a 2.08% (95% CI:
1.32, 2.84%) and a 6.46% (95% CI:
0.93, 12.30%) increase in mortality was found
for each 10-μg/m3 increase in short- and longterm PM2.5, respectively. In restricted analyses,
the corresponding effect estimates were 2.07%
(95% CI: 1.27, 2.89%) and 7.16% (95% CI:
–1.23, 16.27%), respectively.
Our results were robust to the choice of
lag period for acute exposure. We analyzed
different averaging periods (Figure 2): for
example, lag0 (day of death exposure) and
lag04 (a moving average of day of death
exposure and previous 4-day exposure). For
the acute effects, we found a significant but
smaller association for lag0 PM2.5 (1.71%;
95% CI: 1.09, 2.34%) and lag04 PM 2.5
(1.76%; 95% CI: 0.72, 2.81%) than for lag01
analysis. The lag period used for short-term
exposure did not affect estimates of chronic
effects. For example, estimated increases
in mortality with a 10-μg/m 3 increase in
long-term PM 2.5 were 7.35% (95% CI:
1.79, 13.21%) and 7.25% (95% CI:
1.69, 13.12%) when short-term PM2.5 was
classified using lag0 or lag04, respectively.
We also examined effect modification by population size. In the full cohort, a
significant interaction was found for chronic
effects (p < 0.01), with a larger effect of
12.56% (95% CI: 5.71, 19.85%) in urban
areas compared with 3.21% (95% CI:
–2.92, 9.72%) in rural areas. Such a significant
interaction, however, was not observed in the
restricted analysis (p = 0.16). Estimates were
14.27% (95% CI: 3.19, 26.53%) and 5.48%
(95% CI: –4.21, 16.16%) in urban and rural
areas, respectively. For short-term exposure,
population size did not modify the acute
effects in either the full cohort or the restricted
analysis (p = 0.74 and 0.46, respectively).
Table 1. Descriptive statistics for PM2.5 exposure and temperature in New England, 2003–2008.
Covariate
Lag01 PM2.5 (μg/m3)
1-year PM2.5 (μg/m3)
Temperature (˚C)
Mean
8.21
8.12
9.24
SD
5.10
2.28
6.50
Minimum Median Maximum
0.00
7.10
53.98
0.08
8.15
20.22
–36.79
9.81
41.51
Range
53.98
20.14
78.30
Q1
4.60
6.22
4.90
Q3
10.65
10.00
14.39
IQR
6.05
3.78
9.49
Table 2. Percent increase in mortality (95% CI) for a 10-μg/m3 increase for both short-term and long-term
PM2.5.
PM2.5 exposure
With mutual adjustment
Short-term PM2.5
Long-term PM2.5
Without mutual adjustment
Short-term PM2.5
Long-term PM2.5
Model
Percent increase
p-Value
Low daily exposurea
Full cohort
Low chronic exposureb
Full cohort
2.14 ± 0.81
2.14 ± 0.75
9.28 ± 8.88
7.52 ± 5.73
< 0.001
< 0.001
0.032
0.007
Low daily exposurea
Full cohort
Low chronic exposureb
Full cohort
2.07 ± 0.80
2.08 ± 0.76
7.16 ± 8.75
6.46 ± 5.69
< 0.001
< 0.001
0.109
0.026
The full cohort analysis had 551,024 deaths.
aThe analysis was restricted only to person time with daily PM < 30 μg/m3 (422,637 deaths). bThe analysis was restricted
2.5
only to person time with chronic PM2.5 < 10 μg/m3 (268,050 deaths).
Environmental Health Perspectives • volume 124 number 1 January 2016
In our penalized spline model for longterm exposure below the cutoff of 10 μg/m3
(Figure 3A), we found a nonlinear relationship between long-term PM2.5 and mortality.
The association was linear with evidence of
a smaller effect < 6 μg/m3. However, a large
confidence interval was observed; hence, we
could not be confident whether the slope of
the dose–response curve changed for longterm exposures < 6 μg/m3. When examining
the shape of the dose–response curve for
chronic effects, both a linear term for shortterm exposure (the difference) and a penalized
spline for long-term average exposure were
included in the model, resulting in a penalized spline with a df of 1.71. In contrast, we
only included the 2-day average in the penalized spline model of acute effects in order
to provide an interpretable dose–response
relationship (Figure 3B). The results of this
analysis indicated a linear association across
the exposure distribution, but we could not
be certain about the shape of the slope for
acute effects < 3 μg/m3.
Discussion
When we applied the predicted daily PM2.5
with 1-km spatial resolution from our
novel hybrid models, we observed that both
short- and long-term PM2.5 exposure were
significantly associated with all-cause mortality
among residents of New England ≥ 65 years
of age, even when restricted to ZIP codes
and times with annual exposures < 10 μg/m3
or with daily exposure < 30 μg/m3. Hence,
the association of particle exposure with
mortality exists for concentrations below the
current standards established by the United
States, the World Health Organization
(WHO) (10 μg/m3 of annual average PM2.5,
25 μg/m3 daily), and the European Union
(EU) (25 μg/m3 of annual average PM2.5) (EU
2013; WHO 2013). Notably, this analysis
includes all areas in New England and all
Medicare enrollees ≥ 65 years of age in this
region, and it provides chronic effect estimates
that are independent of acute effects. Based
Percent change in mortality
levels of exposure, we performed restricted
analyses: we conducted one analysis restricted
to annual exposures < 10 μg/m3, below the
current annual PM2.5 NAAQS of 12 μg/m3,
and another restricted to observations with
short-term exposure < 30 μg/m3, below the
current daily PM2.5 NAAQS of 35 μg/m3.
After these exclusions, the chronic analyses
were restricted to 268,050 deaths out of
551,024 deaths in total, and the acute
analyses were restricted to 422,637 deaths.
Assessing the dose–response relationship.
For both the acute and chronic analyses,
we fit penalized regression splines in the
restricted analyses to estimate the shape of the
dose–response curve below current U.S. EPA
standards. The degrees of freedom of the
penalized splines for PM2.5 were estimated by
generalized cross-validation (GCV).
2.5
2.0
1.5
1.0
0.5
0.0
lag0
lag1
lag2
lag3
lag4 lag01 lag02 lag03 lag04
PM2.5 lags
Figure 2. Percent change in mortality per 10-μg/m3
increase in short-term PM 2.5 with different lags
with mutual adjustment. Error bars indicate the
95% CIs.
49
Shi et al.
50
composition. The sources and composition of
the particles may differ between low-pollution
days and high-pollution days, which are likely
more affected by secondary aerosols. Compared
with the effect estimate for the full cohort, the
effect estimate from the restricted analysis was
closer to estimates published in the literature
that reported larger effect estimates, such as
those reported by the ESCAPE (European
Study of Cohorts for Air Pollution Effects)
study, the Harvard Six Cities study, and the
Women’s Health Initiative study (Beelen et al.
2014; Puett et al. 2008). Smaller effect estimates were also observed for chronic effects
without mutual adjustment.
To the best of our knowledge, this study
is the first of its kind to restrict exposure and
to explore the dose–response relationship
between PM2.5 below the current U.S. EPA
standards (12 μg/m3 of annual average PM2.5,
35 μg/m3 daily) and mortality. Moreover,
the use of the Medicare cohort means that
we studied the entire population of Medicare
enrollees ≥ 65 years of age and not a convenience sample. In addition, temperature was
controlled on a 1 km × 1 km fine geographic
scale. The acute and chronic effects observed
in analyses restricted to low PM2.5 exposure
were similar to or even higher than those of
the full cohort analyses. These results indicate
that the adverse health effects of PM2.5 are
at least retained, if not strengthened, at low
levels of exposure. However, the findings
from the penalized spline model did not
support a strong association at the lowest
range of PM2.5 concentrations. This finding
provides epidemiological evidence for the
reevaluation of U.S. EPA guidelines and standards, although more evidence is needed to
confirm the association < 6 μg/m3.
The Poisson survival analysis applied in
this study provided a novel method of simultaneously assessing acute and chronic effects.
As shown in our analysis, the chronic effect
estimate was much larger than the acute
effect estimate after controlling for the acute
estimate, indicating that there were chronic
effects of PM 2.5 , which cannot be solely
explained by the short-term exposure.
Another key component of this study
is that the application of high spatial
(1 km × 1 km) and temporal (daily) resolution
of PM2.5 concentrations reduced exposure
error to a certain extent. The out-of-sample R2
was higher than that for the predictions with
10 km × 10 km spatial resolution.
A potential limitation is the limited
availability of individual-level confounders,
such as smoking status, which could bias
the health effect estimates. We were able to
control for ZIP code–level education, median
income, race, and county-level smoking data.
However, Brochu et al. (2011) reported
that census tract–level socioeconomic indicators were uncorrelated with PM2.5 on the
subregional and local scale, providing some
assurance that confounding by socioeconomic status may not be much of an issue.
The results reported by Brochu et al. (2011)
suggest that those variables may not confound
the association, but the inability to control
for them remains an issue. Another limitation
is that we did not examine other pollutants
such as ozone (O3) or nitrogen dioxide (NO2)
owing to a lack of data at the same spatial
level as that of PM2.5.
Conclusions
In conclusion, the acute and chronic effects of
low-concentration PM2.5 were examined for a
Medicare population using a comprehensive
exposure data set obtained from a satellite-based
prediction model. Our findings show that both
short- and long-term exposure to PM2.5 were
associated with all-cause mortality, even for
exposure levels not exceeding the newly revised
U.S. EPA standards, suggesting that adverse
health effects occur at low levels of fine particles. The policy implication of these findings
is that improving the air quality at even lower
levels of PM2.5 than presently allowed by the
U.S. EPA standards can yield health benefits.
25
14
20
12
Percent change in mortality
Percent change in mortality
on a penalized spline model, the positive
dose–response relationship between chronic
exposure and mortality appears to be linear
for PM2.5 concentrations ≥ 6 μg/m 3, with
a positive (though smaller and less precise)
dose–response slope continuing below this
level. This lack of power is likely due to the
small exposed population in areas with annual
PM2.5 < 6 μg/m3, which were quite rural.
For acute effects, we found a 2.14%
(95% CI: 1.38, 2.89%) increase in all-cause
mortality per 10-μg/m3 increment in PM2.5
for the full cohort of our study, which is
higher than the effect size of most studies
using city averages obtained from monitors.
For instance, in a U.S. national study by
Zanobetti and Schwartz (2009), the effect size
was 0.98% (95% CI: 0.75, 1.22%). Similar
results were also obtained in a systematic
review, where researchers determined that
the overall summary estimate was 1.04%
(95% CI: 0.52, 1.56%) per 10-μg/m3 increment in PM2.5 (Atkinson et al. 2014). The
exposure data used in most previous studies
had low spatial resolution (citywide average,
not ZIP code), which introduced exposure
measurement error and likely resulted in a
downward bias in estimates; our results (for
the acute effect) are consistent with such a
phenomenon. Our restricted study estimated
a 2.14% (95% CI: 1.34, 2.95%) increase in
all-cause mortality per 10-μg/m3 increment
in PM2.5, which was close to the effect size
of the full cohort study, possibly because the
sample size of the restricted study for acute
effects was close to that of the full cohort.
Furthermore, the U.S. EPA daily standard
(35 μg/m 3) was almost never exceeded in
this study. In addition, lower effect estimates
for short-term exposure were observed with
mutual adjustment for both full cohort and
restricted analyses. This finding has important
implications for the interpretation of previous
studies without such mutual adjustment.
For chronic effects, the effect estimate
in our full cohort study was consistent with
findings of previous studies with comparable
sample sizes (Hoek et al. 2013; Laden et al.
2006; Lepeule et al. 2012). For example, an
ACS study comprising 500,000 adults from
51 U.S. cities reported a 6% (95% CI:
2, 11%) increase in all-cause mortality for each
10-μg/m3 increment in PM2.5 (Pope et al.
2002). A study of 13.2 million elderly Medicare
recipients across the eastern United States
found a 6.8% (95% CI: 4.9, 8.7%) increase
in all-cause mortality for each 10-μg/m3 increment in PM2.5 (Zeger et al. 2008). When
we restricted our analysis to annual concentrations < 10 μg/m3, a larger slope of 9.28%
(95% CI: 0.76, 18.52%) increase per 10 μg/m3
was observed. Our findings suggest a larger
effect at low concentrations among those
≥ 65 years of age, which may also reflect particle
15
10
5
0
–5
–10
10
8
6
4
2
0
–2
–15
0
2
4
6
8
365-day moving average PM2.5 (µg/m3)
10
0
5
10
15
20
25
30
Lag01 PM2.5 (µg/m3)
Figure 3. The dose–response relationship between long-term PM2.5 and mortality at low doses with mutual
adjustment (A) and the dose–response relationship between short-term PM2.5 and mortality at low doses
without mutual adjustment (B). Shaded areas indicate the 95% CIs.
volume
124 number 1 January 2016 • Environmental Health Perspectives
Low-concentration PM2.5 and mortality
References
Atkinson RW, Kang S, Anderson HR, Mills IC,
Walton HA. 2014. Epidemiological time series
studies of PM2.5 and daily mortality and hospital
admissions: a systematic review and meta-analysis.
Thorax 69:660–665.
Beelen R, Raaschou-Nielsen O, Stafoggia M,
Andersen ZJ, Weinmayr G, Hoffmann B, et al.
2014. Effects of long-term exposure to air pollution on natural-cause mortality: an analysis of 22
European cohorts within the multicentre ESCAPE
project. Lancet 383:785–795.
Brochu PJ, Yanosky JD, Paciorek CJ, Schwartz J,
Chen JT, Herrick RF, et al. 2011. Particulate air
pollution and socioeconomic position in rural and
urban areas of the Northeastern United States.
Am J Public Health 101(suppl 1):S224–S230.
Brook RD, Urch B, Dvonch JT, Bard RL, Speck M,
Keeler G, et al. 2009. Insights into the mechanisms and mediators of the effects of air pollution
exposure on blood pressure and vascular function
in healthy humans. Hypertension 54:659–667.
CDC (Centers for Disease Control and Prevention).
2013. Behavioral Risk Factor Surveillance System.
BRFSS 2013 Survey Data and Documentation.
Available: http://www.cdc.gov/brfss/annual_data/
annual_2013.html [accessed 15 November 2013].
Chen LC, Nadziejko C. 2005. Effects of subchronic
exposures to concentrated ambient particles
(CAPs) in mice: V. CAPs exacerbate aortic plaque
development in hyperlipidemic mice. Inhal toxicol
17:217–224.
Clean Air Act Amendments of 1990. 1990. Public Law
101-549.
CMS (Centers for Medicare & Medicaid Services).
2013a. Denominator File. Available: http://
www.cms.gov/Research-Statistics-Data-andSystems/Files-for-Order/IdentifiableDataFiles/
DenominatorFile.html [accessed 15 November
2013].
CMS. 2013b. Medicare Program—General Information.
Available: http://www.cms.gov/Medicare/
Medicare-General-Information/MedicareGenInfo/
index.html [accessed 15 November 2013].
CMS. 2013c. Medicare Provider Analysis and
Review (MEDPAR) File. Available: http://
www.cms.gov/Research-Statistics-Data-andSystems/Files-for-Order/IdentifiableDataFiles/
MedicareProviderAnalysisandReviewFile.html
[accessed 15 November 2013].
Crouse DL, Peters PA, van Donkelaar A, Goldberg MS,
Villeneuve PJ, Brion O, et al. 2012. Risk of nonaccidental and cardiovascular mortality in relation
to long-term exposure to low concentrations of
fine particulate matter: a Canadian national-level
cohort study. Environ Health Perspect 120:708–714;
doi:10.1289/ehp.1104049.
Dockery DW, Pope CA III, Xu X, Spengler JD, Ware JH,
Fay ME, et al. 1993. An association between air
pollution and mortality in six U.S. cities. New Engl J
Med 329:1753–1759.
Dominici F, Peng RD, Bell ML, Pham L, McDermott A,
Zeger SL, et al. 2006. Fine particulate air pollution
and hospital admission for cardiovascular and
respiratory diseases. JAMA 295:1127–1134.
EU (European Union). 2013. Air Quality Standards.
Available: http://ec.europa.eu/environment/air/
quality/standards.htm [accessed 15 November
2013].
Franklin M, Zeka A, Schwartz J. 2007. Association
between PM2.5 and all-cause and specific-cause
mortality in 27 US communities. J Expo Sci Environ
Epidemiol 17:279–287.
Furuyama A, Hirano S, Koike E, Kobayashi T. 2006.
Induction of oxidative stress and inhibition of
plasminogen activator inhibitor-1 production in
endothelial cells following exposure to organic
extracts of diesel exhaust particles and urban fine
particles. Arch Toxicol 80:154–162.
Gauderman WJ, Avol E, Gilliland F, Vora H, Thomas D,
Berhane K, et al. 2004. The effect of air pollution
on lung development from 10 to 18 years of age.
New Engl J Med 351:1057–1067.
Gurgueira SA, Lawrence J, Coull B, Murthy GK,
González-Flecha B. 2002. Rapid increases in the
steady-state concentration of reactive oxygen
species in the lungs and heart after particulate
air pollution inhalation. Environ Health Perspect
110:749–755.
Hart J, Garshick E, Dockery D, Smith T, Ryan L, Laden F.
2011. Long-term ambient multi-pollutant exposures
and mortality. Am J Respir Crit Care Med 183:73–78.
Hoek G, Krishnan RM, Beelen R, Peters A, Ostro B,
Brunekreef B, et al. 2013. Long-term air pollution
exposure and cardio-respiratory mortality: a review.
Environ Health 12:43. doi:10.1186/1476-069X-12-43.
Jerrett M, Burnett RT, Ma R, Pope CA III, Krewski D,
Newbold KB, et al. 2005. Spatial analysis of
air pollution and mortality in Los Angeles.
Epidemiology 16(6):727–736.
Katsouyanni K, Touloumi G, Spix C, Schwartz J,
Balducci F, Medina S, et al. 1997. Short-term
effects of ambient sulphur dioxide and particulate
matter on mortality in 12 European cities: results
from time series data from the APHEA project. Air
Pollution and Health: a European Approach. BMJ
314:1658–1663.
Kloog I, Chudnovsky AA, Just AC, Nordio F,
Koutrakis P, Coull BA, et al. 2014a. A new hybrid
spatio-temporal model for estimating daily multiyear PM2.5 concentrations across northeastern
USA using high resolution aerosol optical depth
data. Atmos Environ 95:581–590.
Kloog I, Coull BA, Zanobetti A, Koutrakis P,
Schwartz JD. 2012. Acute and chronic effects of
particles on hospital admissions in New-England.
PLoS ONE 7:e34664; doi:10.1371/journal.
pone.0034664.
Kloog I, Koutrakis P, Coull BA, Lee HJ, Schwartz J.
2011. Assessing temporally and spatially resolved
PM2.5 exposures for epidemiological studies using
satellite aerosol optical depth measurements.
Atmos Environ 45:6267–6275.
Kloog I, Nordio F, Coull BA, Schwartz J. 2014b.
Predicting spatiotemporal mean air temperature
using MODIS satellite surface temperature
measurem ents across the Northeastern USA.
Remote Sens Environ 150:132–139.
Kloog I, Ridgway B, Koutrakis P, Coull BA,
Schwartz JD. 2013. Long-and short-term exposure
to PM 2.5 and mortality: using novel exposure
models. Epidemiology 24:555–561.
Laden F, Schwartz J, Speizer FE, Dockery DW. 2006.
Reduction in fine particulate air pollution and
mortality: extended follow-up of the Harvard
Six Cities study. Am J Respir Crit Care Med
173:667–672.
Laird N, Olivier D. 1981. Covariance analysis of
censored survival data using log-linear analysis
techniques. J Am Stat Assoc 76:231–240.
Lepeule J, Laden F, Dockery D, Schwartz J. 2012.
Chronic exposure to fine particles and mortality:
an extended follow-up of the Harvard Six Cities
study from 1974 to 2009. Environ Health Perspect
120:965–970; doi:10.1289/ehp.1104660.
Miller KA, Siscovick DS, Sheppard L, Shepherd K,
Sullivan JH, Anderson GL, et al. 2007. Longterm exposure to air pollution and incidence of
Environmental Health Perspectives • volume 124 number 1 January 2016
cardiovascular events in women. New Engl J Med
356:447–458.
Ohtoshi T, Takizawa H, Okazaki H, Kawasaki S,
Takeuchi N, Ohta K, et al. 1998. Diesel exhaust
particles stimulate human airway epithelial cells to
produce cytokines relevant to airway inflammation
in vitro. J Allergy Clin Immunol 101(6 pt 1):778–785.
Pope CA III. 2000. Epidemiology of fine particulate
air pollution and human health: biologic mechanisms and who’s at risk? Environ Health Perspect
108:713–723.
Pope CA III, Burnett RT, Thun MJ, Calle EE, Krewski D,
Ito K, et al. 2002. Lung cancer, cardiopulmonary
mortality, and long-term exposure to fine particulate air pollution. JAMA 287:1132–1141.
Pope CA III, Thun MJ, Namboodiri MM, Dockery DW,
Evans JS, Speizer FE, et al. 1995. Particulate air
pollution as a predictor of mortality in a prospective study of U.S. Adults. Am J Respir Crit Care
Med 151:669–674.
Puett RC, Hart JE, Yanosky JD, Paciorek C, Schwartz J,
Suh H, et al. 2009. Chronic fine and coarse particulate exposure, mortality, and coronary heart
disease in the Nurses’ Health Study. Environ Health
Perspect 117:1697–1701; doi:10.1289/ehp.0900572.
Puett RC, Schwartz J, Hart JE, Yanosky JD,
Speizer FE, Suh H, et al. 2008. Chronic particulate
exposure, mortality, and coronary heart disease
in the Nurses’ Health Study. Am J Epidemiol
168:1161–1168.
Samoli E, Peng R, Ramsay T, Pipikou M, Touloumi G,
Dominici F, et al. 2008. Acute effects of ambient
particulate matter on mortality in Europe and North
America: results from the APHENA study. Environ
Health Perspect 116:1480–1486; doi:10.1289/
ehp.11345.
Schwartz J. 1994. Air pollution and daily mortality: a
review and meta analysis. Environ Res 64(1):36–52.
Schwartz J. 2000. Harvesting and long term exposure
effects in the relation between air pollution and
mortality. Am J Epidemiol 151:440–448.
Schwartz J. 2004. Is the association of airborne particles with daily deaths confounded by gaseous air
pollutants? An approach to control by matching.
Environ Health Perspect 112:557–561; doi:10.1289/
ehp.6431.
Schwartz J, Coull B, Laden F, Ryan L. 2008. The effect
of dose and timing of dose on the association
between airborne particles and survival. Environ
Health Perspect 116:64–69; doi:10.1289/ehp.9955.
Schwartz J, Dockery DW. 1992. Increased mortality
in Philadelphia associated with daily air pollution
concentrations. Am Rev Respir Dis 145:600–604.
Schwartz J, Dockery DW, Neas LM. 1996. Is daily
mortality associated specifically with fine particles? J Air Waste Manage Assoc 46:927–939.
Schwartz J, Zanobetti A. 2000. Using meta-smoothing
to estimate dose-response trends across multiple
studies, with application to air pollution and daily
death. Epidemiology 11:666–672.
Stellman SD, Garfinkel L. 1986. Smoking habits and tar
levels in a new American Cancer Society prospective study of 1.2 million men and women. J Natl
Cancer Inst 76(6):1057–1063.
Suwa T, Hogg JC, Quinlan KB, Ohgami A, Vincent R,
van Eeden SF. 2002. Particulate air pollution
induces progression of atherosclerosis. J Am Coll
Cardiol 39:935–942.
U.S. Census Bureau. 2000. Summary File 3 (SF 3).
Available: http://www.census.gov/census2000/
sumfile3.html [accessed 15 November 2013].
U.S. EPA (U.S. Environmental Protection Agency).
1997. 40 CFR Part 50. National Ambient Air Quality
Standards for particulate matter. Final rule. Fed
Reg 62:38652–38460.
51
Shi et al.
U.S. EPA. 2006. 40 CFR Part 50. National Ambient Air
Quality Standards for particulate matter. Final rule.
Fed Reg 71:61144–61233. Available: http://www.
epa.gov/ttn/amtic/files/ambient/pm25/pt5006.pdf
[accessed 17 October 2006].
U.S. EPA. 2013. 40 CFR Parts 50, 51, 52 53 and 58. National
Ambient Air Quality Standards for particulate matter.
Final rule. Fed Reg 78:3086–3287. Available: http://
www.gpo.gov/fdsys/pkg/FR-2013-01-15/pdf/201230946.pdf [accessed 15 January 2013].
Ver Hoef JM, Boveng PL. 2007. Quasi-Poisson vs.
52
negative binomial regression: how should we model
overdispersed count data? Ecology 88(11):2766–2772.
WHO (World Health Organization). 2013. Ambient
(Outdoor) Air Quality and Health. Available: http://
www.who.int/mediacentre/factsheets/fs313/en/
[accessed 15 November 2013].
Yue W, Schneider A, Stölzel M, Rückerl R, Cyrys J,
Pan X, et al. 2007. Ambient source-specific particles
are associated with prolonged repolarization and
increased levels of inflammation in male coronary
artery disease patients. Mutat Res 621:50–60.
volume
Zanobetti A, Schwartz J. 2009. The effect of fine and
coarse particulate air pollution on mortality:
a national analysis. Environ Health Perspect
117:898–903; doi:10.1289/ehp.0800108.
Zeger SL, Dominici F, McDermott A, Samet JM. 2008.
Mortality in the Medicare population and chronic
exposure to fine particulate air pollution in urban
centers (2000–2005). Environ Health Perspect
116:1614–1619; doi:10.1289/ehp.11449.
124 number 1 January 2016 • Environmental Health Perspectives