American Journal of Epidemiology
© The Author 2014. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of
Public Health. All rights reserved. For permissions, please e-mail: [email protected].
Vol. 180, No. 7
DOI: 10.1093/aje/kwu183
Advance Access publication:
August 19, 2014
Original Contribution
Ozone and Daily Mortality Rate in 21 Cities of East Asia: How Does Season Modify
the Association?
Renjie Chen, Jing Cai, Xia Meng, Ho Kim, Yasushi Honda, Yue Leon Guo, Evangelia Samoli,
Xin Yang, and Haidong Kan*
* Correspondence to Dr. Haidong Kan, School of Public Health, Fudan University, P.O. Box 249, 130 Dong-An Road, Shanghai 200032, China
(e-mail: [email protected]).
Initially submitted February 6, 2014; accepted for publication June 13, 2014.
Previous studies in East Asia have revealed that the short-term associations between tropospheric ozone and
daily mortality rate were strongest in winter, which is opposite to the findings in North America and Western Europe.
Therefore, we investigated the season-varying association between ozone and daily mortality rate in 21 cities of
East Asia from 1979 to 2010. Time-series Poisson regression models were used to analyze the association
between ozone and daily nonaccidental mortality rate in each city, testing for different temperature lags. The
best-fitting model was obtained after adjustment for temperature in the previous 2 weeks. Bayesian hierarchical
models were applied to pool the city-specific estimates. An interquartile-range increase of the moving average concentrations of same-day and previous-day ozone was associated with an increase of 1.44% (95% posterior interval
(PI): 1.08%, 1.80%) in daily total mortality rate after adjustment for temperature in the previous 2 weeks. The corresponding increases were 0.62% (95% PI: 0.08%, 1.16%) in winter, 1.46% (95% PI: 0.89%, 2.03%) in spring,
1.60% (95% PI: 1.03%, 2.17%) in summer, and 1.12% (95% PI: 0.73%, 1.51%) in fall. We found significant associations between short-term exposure to ozone and higher mortality rate in East Asia that varied considerably from
season to season with a significant trough in winter.
air pollution; East Asia; mortality rate; ozone; season; temperature
Abbreviations: IQR, interquartile range; NMMAPS, National Morbidity, Mortality, and Air Pollution Study; PI, posterior interval;
PM10, particulate matter with aerodynamic diameter of 10 μm or less.
and population growth (6). Single-city studies in Hong Kong,
Shanghai, and Suzhou, China, have reported stronger ozone–
mortality rate associations in the cool period or in winter (7–
9). A recent 4-city study in the Pearl River Delta of Southern
China estimated higher mortality effects of ozone during the
nonpeak period compared with the peak period (10). Another
recent multicity study in Japan indicated significant effects of
short-term ozone exposure on daily total and cardiovascular
deaths among the elderly during spring and fall but not during winter and summer (11).
Given the inconsistent findings regarding these seasonal
patterns in East Asia versus those obtained in North America
and Western Europe, there was a need to reevaluate this issue
in more cities of East Asia. Therefore, the main objective of
this multicenter study was to investigate the modifying effect
Tropospheric ozone (herein referred to as ozone) is a photochemical pollutant formed through a complex series of
reactions involving the action of sunlight on nitric oxide
and hydrocarbons. The health effects associated with shortterm exposure to ambient ozone have been well documented
worldwide (1–3). In general, studies from North America and
Western Europe have reported significant associations primarily restricted to the summer or warm periods, when
ozone concentrations were high (1, 3–5). However, the association between ozone and mortality rates during cold seasons
in these regions was not investigated extensively because the
ozone concentrations were much lower or not routinely monitored in some cities.
East Asia is a region with one of the highest levels of ozone
pollution in the world because of rapid economic development
729
Am J Epidemiol. 2014;180(7):729–736
730 Chen et al.
of season on the association of ozone with daily mortality rate
in 21 cities from mainland China, Hong Kong, Taiwan, Korea,
and Japan. Temperature has been demonstrated to modify the
ozone–mortality rate association (12), and its health effects
may be lagged for days or weeks (13); however, previous
East Asian analyses tended to adjust only for current-day
temperature. Thus, a secondary objective of this study was
to examine the appropriate temperature control for analyzing
the ozone–mortality rate association in this region.
METHODS
Data collection
Our study included the following 21 cities throughout East
Asia, which were chosen according to data availability: Beijing, Suzhou, and Shanghai in mainland China; Hong Kong;
Taipei, Taichung, and Kaohsiung in Taiwan; Busan, Daegu,
Daejeon, Gwangju, Incheon, Seoul, and Ulsan in Korea; and
Fukuoka, Kitakyushu, Nagoya, Osaka, Sapporo, Sendai, and
Tokyo in Japan. These cities, with a total population of approximately 80 million urban residents, covered a wide range
of geographical and socioeconomic features and are characteristic of East Asian countries.
The sources of daily mortality data were the Municipal
Center for Disease Control and Prevention for each city in
mainland China; the Census and Statistics Department in
Hong Kong; the Department of Health in Taiwan; the Ministry of Health, Labor, and Welfare for each city in Japan; and
the Korea National Statistics Office for each city in Korea.
The total nonaccidental mortality rate was coded according
to International Classification of Diseases, Ninth Revision,
codes 001–799 and International Classification of Diseases,
Tenth Revision, codes A00–R99.
We collected data on daily concentrations of ozone, as well
as particulate matter with aerodynamic diameter of 10 μm or
less (PM10), sulfur dioxide, and nitrogen dioxide from fixedsite air monitoring stations in each city. Ozone was measured
year-round in all cities, and few data were missing. We computed daily maximum 8-hour averages (from 10 AM to 6 PM)
for ozone and the daily 24-hour mean for the other pollutants.
In each city, the daily air pollutants’ concentrations were averaged from the available monitoring measurements across various stations. We also obtained daily mean temperature and
relative humidity data from the meteorological departments
in each city to allow adjustment for the potential confounding
effects of weather conditions on daily mortality rate.
Statistical analyses
We applied 2-stage Bayesian hierarchical models to estimate the regional average associations of ozone and daily
mortality rate. In the first stage, we fitted time-series regression
to estimate the annual average and season-specific associations
of mortality rate and ozone in each city; in the second stage, we
pooled the city-specific estimates at the country level and regional level using Bayesian hierarchical models (3).
In the first stage, we initially applied a “main” model that
assumed the associations of ozone were constant throughout
the year. We included, a priori, the average of the same
day’s and previous day’s ozone concentrations (lag 01) in all
models because this lag was most commonly used in previous
studies and was shown to be more strongly associated with the
observed health effects (3, 4, 14). In each city, we used quasiPoisson generalized additive models because daily deaths
typically followed an overdispersed Poisson distribution. To
control for potential confounders, we incorporated the following covariates in the generalized additive models: 1) a natural
cubic spline of calendar time with 7 degrees of freedom per
year to control for unmeasured long-term and seasonal trends in
daily mortality rate, 2) natural cubic splines of mean temperature and relative humidity to adjust for the potential confounding
effects of weather conditions on mortality rate, 3) an indicator
variable for day of the week, and 4) influenza epidemics (15).
Hence, the formula for the “main” model is
logEðYt Þ ¼ β1 Ztl þ nsðt; dfÞ þ nsðtemperaturet ; 6Þ
þ nsðhumidityt ; 3Þ þ Σβ2 DOWt
þ β3 influenza,
ð1Þ
where Yt is the daily count of total nonaccidental deaths; β1 is the
regression coefficient of ozone on mortality rate; Zt−l is ozone
on day t at a lag; ns(t, df) is the natural spline smoothing function of calendar time with 7 df per year; ns(temperaturet, 6) is the
natural spline of temperature on day t with 6 df for the whole
study period; ns(humidityt, 3) is the natural spline of relative humidity on day t with 3 df for the whole study period; DOWt is
the indicator of day of the week on day t; β2 is the regression
coefficient relating day of the week; influenza is a dummy variable taking the value of 1 when the 7-day moving average of
the respiratory mortality rate was greater than the 90th percentile
of its city-specific distribution (16); and β3 is the regression coefficient of influenza. Ozone was introduced into the model as a
linear term (17). Because temperature’s effects on health may be
lagged over days or weeks (13, 18), we alternatively included
different lag structures of temperature in our model, namely
lag 0, the moving average of lags 0–3, lags 0–7, lags 0–14,
and lags 0–28 days. We also controlled for 2 temperature
terms as proposed in the National Morbidity, Mortality, and
Air Pollution Study (NMMAPS) (17) (i.e., the present-day temperature and the average of the 3 previous days’ temperatures).
To determine the optimal temperature control for the main analyses, we further compared the different models using fit parameters, including the adjusted R 2 value and the generalized
cross-validation criteria. The optimal terms for lagged temperature were determined by maximizing the R 2 value and minimizing the scores of generalized cross-validation. We incorporated
only the present-day relative humidity in the models because
previous studies have shown no evidence of confounding by
this variable in air pollution epidemiology.
Then, we constructed the “seasonal” model to estimate the
season-specific associations of ozone (19–21). This model
was similar to the “main” model except for the following 2
replacements:
1. We replaced the β with an ozone × season interaction term
β ¼ βw Iw þ βsp Isp þ βsu Isu þ βf If ;
ð2Þ
Am J Epidemiol. 2014;180(7):729–736
Ozone and Mortality Rate in East Asia 731
where Iw, Isp, Isu, and If are indicators of winter, spring,
summer, and fall, respectively, and βw, βsp, βsu, and βf
are the respective regression coefficients of ozone’s
associations with mortality rate in winter (December–
February), spring (March–May), summer (June–August),
and fall (September–November), respectively.
2. To control the temporal trend in each season, we replaced
the ns(t, df ) in equation 1 with
nsðt; dfÞ ¼ nsðt; dfÞIw þ nsðt; dfÞIsp
þ nsðt; dfÞIsu þ nsðt; dfÞIf :
ð3Þ
In the second stage, we applied the Bayesian hierarchical models to pool the city-specific estimates while accounting for
within-city statistical error and between-city variability (i.e.,
heterogeneity) of the “true” estimates. For each association
under investigation, we estimated a city-specific maximum
likelihood estimate ^β, which is a scalar for the “main” model
and a vector of length 4 for the “seasonal” model. ^β is assumed
to be normally distributed around the true city-specific β with
covariance matrix V, estimated within each city. We applied
this hierarchical model by using 2-level normal independent
sampling estimation with uniform priors (21). This provides
a sample from the posterior distribution from which one can
calculate posterior means and variances of the overall and cityspecific ozone estimates. Then, we presented the estimates as
the posterior means of the percentage increases in daily mortality rate and their 95% posterior intervals.
We evaluated the between-city heterogeneity of all-year
estimates and the within-city heterogeneity of season-specific
estimates by calculating the I 2 statistic. The I 2 statistic provides an estimate of the percentage of variability in magnitude across estimates that is likely due to true differences as
opposed to chance. We considered I 2 > 0.50 as indicating significant heterogeneity and 0.25 < I 2 < 0.50 as indicating moderate heterogeneity.
We investigated the sensitivity of our results to the simultaneous exposure to other pollutants by alternatively fitting
2-pollutant models with PM10, sulfur dioxide, and nitrogen
dioxide. We also estimated the all-year and season-specific
associations of ozone at single-day lags of 0–6 days and a
7-day moving average to examine whether the seasonal pattern of ozone’s association with mortality rate changed when
using alternative lags of ozone.
Table 1. Basic Descriptive Information in 21 East Asian Cities, 1979–2010
City by Region
Study Dates
Mean No. of
Daily Deaths
Mean Ozone Level, μg/m3
All Year
Winter
Spring
Summer
Fall
Mean
Temperature, °C
Mainland Chinaa
Beijing
2007–2008
120.7
43.5
15.2
48.3
81.8
34.6
12.0
Hong Kong
1996–2002
84.2
36.6
34.3
35.9
25.3
51.0
23.7
Shanghai
2001–2004
119.0
63.1
36.4
70.3
81.3
64.1
17.7
Suzhou
2005–2008
34.2
55.8
30.6
73.1
59.8
56.3
17.2
Kaohsiung
1994–2008
41.6
132.5
138.7
132.2
90.1
169.3
25.3
Taichung
1994–2008
32.1
118.8
98.4
124.7
109.2
142.7
23.7
Taipei
1994–2008
76.5
95.5
71.6
106.3
117.8
85.9
23.2
Fukuoka
1983–2009
19.9
88.7
77.1
109.5
82.9
85.4
17.1
Kitakyushu
1983–2009
22.2
51.2
47.2
68.3
43.6
45.6
17.1
Nagoya
1979–2009
38.1
70.5
49.7
89.0
83.9
58.9
15.8
Osaka
1979–2009
57.4
86.6
60.0
98.8
106.5
80.5
16.9
Sapporo
1992–2009
31.0
71.0
60.1
99.6
63.8
60.1
9.2
Sendai
1984–2009
13.5
81.8
76.1
104.1
76.3
70.3
12.5
Tokyo
1979–2009
150.2
78.1
54.7
93.4
97.9
65.4
16.3
Busan
1999–2010
51.3
82.2
60.9
99.0
86.5
82.3
15.0
Daegu
1999–2010
31.4
83.7
54.2
105.9
98.7
74.9
14.7
Daejeon
1999–2010
16.1
80.1
53.2
102.9
90.8
73.1
13.2
Gwangju
1999–2010
16.2
78.3
52.5
97.4
86.8
75.9
14.3
Incheon
1999–2010
30.2
77.9
51.4
92.5
93.0
73.7
12.9
Seoul
1999–2010
105.0
73.2
40.2
89.7
98.1
63.6
13.0
Ulsan
1999–2010
11.1
81.5
61.5
99.2
86.7
78.6
14.7
Taiwan
Japan
Korea
a
Including Hong Kong.
Am J Epidemiol. 2014;180(7):729–736
0.3718
0.3734
0.3728
0.3736
0.3740
0.3737
0 and 1–3
0–3
0–7
0–14
0–28
Adjusted R
0
Abbreviations: GCV, generalized cross-validation; PI, posterior interval.
a
Estimates are presented as percent increases (posterior mean and 95% PIs) of daily total mortality rate associated with an interquartile-range increase of the average ozone concentrations
of the same day and previous day.
0.73, 1.51
0.92, 1.75
0.70, 1.48
1.09
0.96, 2.03
1.50
0.92, 2.04
1.48
0.20, 1.24
0.72
1.07, 1.75
1.41
1.1237
1.12
1.03, 2.17
1.60
0.89, 2.03
1.46
0.08, 1.16
0.62
1.08, 1.80
1.44
1.1231
0.99, 1.83
1.41
1.33
0.62. 1.63
0.41, 1.73
1.07
1.13
1.05, 2.18
1.20, 2.28
1.74
1.61
0.20, 1.41
0.54, 1.44
0.99
0.81
1.13, 1.84
1.10, 1.76
1.43
1.49
1.1254
1.1260
95% PI
0.62, 1.27
0.94
0.50, 1.64
1.07
0.28, 1.24
0.76
0.15, 0.86
0.50
0.58, 1.09
0.84
1.1248
1.00
Estimate
95% PI
0.10, 0.78
0.44
Estimate
95% PI
0.64, 1.54
1.09
Estimate
95% PI
0.71, 1.84
1.28
Estimate
95% PI
0.67, 1.15
0.91
Estimate
GCV
1.1282
Fall
Summer
Spring
Winter
All Year
2
Model Fit
Table 1 shows descriptive data on the study period, daily
total deaths, ozone concentrations, and mean temperature
from the 21 large East Asian cities included in our analysis.
The study period varied from city to city according to the data
availability, but it was longer than 10 years in most cities. The
mean daily total deaths ranged from 11 in Ulsan to 150 in
Tokyo. Cities in Taiwan had the highest mean concentration
of ozone (115.6 μg/m3), followed by Korea (59.5 μg/m3),
Japan (56.4 μg/m3), and mainland China (49.8 μg/m3). Generally, the annual average concentrations of ozone in East
Asia were on a par with those reported in developed Western
countries. The mean ozone levels varied greatly across seasons with higher concentrations in spring (73.3 μg/m3) and
summer (68.4 μg/m3) than in fall (55.7 μg/m3) and winter
(39.9 μg/m3), except for the cities in Taiwan, where the highest concentrations were observed in fall. The magnitude
of variation in the daily ozone concentrations also differed
across seasons. The averaged IQRs of ozone concentration
were 22.0 μg/m3 in winter, 34.4 μg/m3 in spring, 44.6 μg/
m3 in summer, and 31.9 μg/m3 in fall (Web Table 1, available
at http://aje.oxfordjournals.org/). The annual mean temperatures ranged from 9.2°C in Sapporo to 25.3°C in Kaohsiung,
reflecting distinct climatic characteristics among these cities.
Overall, ozone was weakly associated with PM10, sulfur
dioxide, and nitrogen dioxide (mean correlation coefficients =
0.21, 0.04, and −0.01, respectively), and it was positively correlated with temperature (mean correlation coefficient = 0.29)
and negatively correlated with humidity (mean correlation
coefficient = −0.30). Specifically, the mean correlation coefficients were −0.05 in winter, 0.33 in spring, 0.12 in summer,
and 0.28 in fall.
Table 2 presents the model fit criteria following different
temperature control with different moving lag days and 2
temperature terms. The highest R 2 values and lowest generalized cross-validation scores were obtained under the moving
average of temperature at lags 0–14. Hence, we controlled for
temperature using lags of 0–14 days in the main analyses.
Table 2 also shows the annual and season-specific associations of ozone and mortality rate at lags 0–1 for different lag
structures of temperature. The all-year estimates of ozone increased after controlling for multiple temperature lags. For
example, when adjusting for the moving average temperature
of lag 0–14 days, we estimated a 1.44% (95% posterior interval (PI): 0.76%, 1.19%) increase in daily total mortality rate
per an IQR increase of ozone, compared with a corresponding increase of 0.91% (95% PI: 0.67%, 1.15%) when
Temperature Lag
RESULTS
Table 2. Model Fit Parameters and Estimatesa of Association With Ozone in Models Adjusted for Different Temperature Lags in 21 East Asian Cities, 1979–2010
The statistical tests were 2-sided, and associations of P <
0.05 were considered statistically significant. All models
were fitted with R statistical software, version 2.15.1 (R
Foundation for Statistical Computing, Vienna, Austria) using
the mgcv package for generalized additive models, and tlnise
for Bayesian hierarchical models. To allow for the comparison of season-specific estimates, we calculated excess logrelative mortality rates per increases of ozone equal to the
city-specific and season-specific interquartile ranges (IQRs)
in the main analysis. We also reported the ozone–mortality
rate associations using a 10-μg/m3 increase.
0.62, 1.38
732 Chen et al.
Am J Epidemiol. 2014;180(7):729–736
Ozone and Mortality Rate in East Asia 733
Table 3. Percent Increase of Daily Total Mortality Rate Associated With an Interquartile-Range Increase of Ozone (Lags 0–1) in China, Japan,
Korea, and Overall, Adjusted for Temperature With Lags 0–14 in 21 East Asian Cities, 1979–2010
All Year
Region
Estimate
Winter
95% PI
Estimate
Spring
95% PI
Estimate
Summer
95% PI
Estimate
Fall
95% PI
Estimate
95% PI
Mainland China
2.35
0.78, 3.92
1.49
−2.48, 5.47
2.28
0.93, 3.63
3.12
0.94, 5.31
2.42
0.50, 4.34
Taiwan
0.84
0.23, 1.44
0.30
−0.54, 1.15
0.75
−0.23, 1.72
1.07
−0.51, 2.66
1.06
0.05, 2.07
a
Japan
0.84
0.33, 1.36
0.25
−0.63, 1.13
1.16
0.45, 1.86
1.06
0.19. 1.93
0.55
0.26, 0.84
Korea
2.05
1.53, 2.58
0.23
−0.69, 1.15
2.47
1.86, 3.07
2.38
1.44, 3.32
1.64
0.82, 2.46
1.44
1.08, 1.80
0.62
0.08, 1.16
1.46
0.89, 2.03
1.60
1.03, 2.17
1.12
0.73, 1.51
Overall
Abbreviation: PI, posterior interval.
a
Including Hong Kong.
controlling for the current-day temperature only. Seasonspecific estimates were attenuated in winter from 1.28% when
controlling for the same-day temperature to 0.62% when controlling for lags 0–14, but they increased drastically in summer from 0.44% to 1.60%, and they also increased slightly
in transitional seasons. The seasonal pattern of ozone’s associations with mortality rate varied by different temperature
controls—from a winter peak when controlling for same-day
temperature to a winter trough when using multiple lags. For
example, when adjusting for same-day temperature, we estimated a 1.28% increased daily mortality rate in winter, 1.09%
in spring, 0.44% in summer, and 1.00% in fall; however, the
corresponding increases were 0.62%, 1.46%, 1.60%, and
1.12% when controlling for temperature with lags of 0–14
days. We also obtained a similar pattern with a trough in winter when using 2 temperature terms.
Table 3 presents the pooled percent increase of daily mortality rate associated with an IQR increase of ozone in 4 regions (Taiwan, Japan, Korea, and mainland China including
Hong Kong), adjusting for temperature at lags 0–14 days.
The seasonal pattern remained the same across regions with
an appreciable trough in winter. This seasonal pattern was
also evident in city-specific results (Web Table 2).
We observed significant heterogeneity for the all-year associations of ozone and mortality rate among the 21 cities,
with the lowest heterogeneity in Korean cities. For the
season-specified estimates of ozone associations, the overall
differences for the seasonal estimates were significant (averaged I 2 > 0.50); the seasonal difference was most significant
for mainland Chinese cities (data not shown).
Results from 2-pollutant models (Figure 1) indicated that
ozone’s associations were fairly stable after adjusting in turn
for PM10, sulfur dioxide, and nitrogen dioxide. Figure 2 illustrates that our main lag (lag 01) generated the largest estimates
compared with single-day lags of 0–6 days and the moving
average lag of 0–6 days. Ozone’s association with mortality
rate varied greatly by season, with an obvious trough in winter using alternative lags for ozone, which was also evident
when using 10 μg/m3 as the ozone increment for the estimation of mortality rate change (Web Table 3).
% Increase In Mortality Rate
2.5
2.0
1.5
1.0
0.5
0.0
-
+PM10 +SO2 +NO2
-
All Year
+PM10 +SO2 +NO2
Winter
-
+PM10 +SO2 +NO2
Spring
-
+PM10 +SO2 +NO2
Summer
-
+PM10 +SO2 +NO2
Fall
Ozone
Figure 1. Percent increase ( posterior mean and 95% posterior intervals) of daily total mortality rate associated with an interquartile-range increase
of ozone, adjusted for copollutants in 2-pollutant models in 21 East Asian cities, 1979–2010. Dashes under the x-axis indicate results obtained from
single-pollutant models of ozone. NO2, nitrogen dioxide; PM10, particulate matter with aerodynamic diameter of 10 μm or less; SO2, sulfur dioxide.
Am J Epidemiol. 2014;180(7):729–736
734 Chen et al.
% Increase In Mortality Rate
2.5
2.0
1.5
1.0
0.5
0.0
–0.5
L01 L0 L1 L2 L3 L4 L5 L6 L06 L01 L0 L1 L2 L3 L4 L5 L6 L06 L01 L0 L1 L2 L3 L4 L5 L6 L06 L01 L0 L1 L2 L3 L4 L5 L6 L06 L01 L0 L1 L2 L3 L4 L5 L6 L06
All Year
Winter
Spring
Summer
Fall
Lag Days
Figure 2. Percent increase ( posterior mean and 95% posterior intervals) of daily total mortality rate associated with an interquartile-range increase
of ozone at alternative lag days in 21 East Asian cities, 1979–2010.
DISCUSSION
This multicenter time-series study in 21 large cities across
East Asia provided robust evidence of a significant association between short-term ozone exposure and mortality rate.
Results were generally robust to adjustment for copollutants,
the use of alternative lags for ozone, and the choice of units
for the estimate, but they were sensitive to how we adjusted
for temperature. Our study did not support previous findings
that the associations of ozone with daily mortality rate peaked
in winter or in cool months in East Asia.
On average across the 21 cities, we estimated a 1.44%
(95% PI: 1.08%, 1.80%) increase in daily total mortality
rate for an IQR increase in the average of the same and previous day’s ozone concentrations. To allow for the comparison of studies with different ozone metrics, we used a
relationship of 20:15:8 for the 1-hour maximum: 8-hour
maximum: daily average (22). The pooled all-year estimate
in our study was a 0.30% increase in total mortality rate associated with a 10-μg/m3 increase in 8-hour maximum ozone.
This estimate was generally consistent with previously reported results from meta-analyses and multicity studies
worldwide (1–4, 10, 14, 17, 22–26). Meta-analyses have estimated 0.21%–0.34% increases of mortality rate per 10-μg/
m3 increase in 8-hour maximum ozone, and multicity studies
have estimated a wider range of 0.03%–1.13%.
Consistent with previous multicity studies (3, 17), our
study found significant heterogeneity for the estimates of
ozone’s associations with mortality rate among the 21 cities,
which may be attributable to a number of factors. Exposure
characteristics, such as building ventilation, insulation, use of
air conditioning, and time-location activity patterns, as well
as population susceptibility factors including age structure
and socioeconomic status, might vary across these cities.
For example, a higher proportion of elderly residents was expected in Japan compared with mainland China. However,
the overall results strongly support a positive association between ozone exposure and mortality rate, although the cityspecific estimates were significantly heterogeneous.
It has been widely reported that the levels of ambient
ozone vary seasonally and, thus, the associated health effects
may also be different across seasons (27). A number of studies have recorded significant associations of ozone with mortality rate in summer but weak or null associations in other
seasons. In the Air Pollution and Health: A European Approach project, Gryparis et al. (4) found that a 10-μg/m3 increase in 8-hour maximum ozone was associated with a
0.31% (95% confidence interval: 0.17%, 0.52%) increase in
daily total mortality rate in the summer period; however, this
estimate was 0.12% (95% confidence interval: −0.12%,
0.37%) in the winter season. A meta-analysis of 144 effect
estimates by Bell et al. (1), mainly from North America
and Western Europe, also showed a larger effect of ozone
during the warm period (May–October) than in the whole
year. Similarly, a US multicity study by Zanobetti and
Schwartz (5) detected a significant effect of ozone on daily
mortality rate in summer, whereas the effect decreased to
null in the fall and winter. More recently, in an intercontinental study including 23 European cities, 12 Canadian cities,
and 86 US cities, Peng et al. (3) found appreciably stronger
ozone–mortality rate associations during the warm period
(April–September) compared with the all-year estimates.
Contrary to the findings in North America and Western Europe, several previous studies in East Asia have consistently
reported significant and larger mortality effects of ozone in
the cold season. In Hong Kong, Shanghai, and Suzhou,
China, significant associations were reported during the
cold period (October–March) but not the warm period
(April–September) (7–9). Tao et al. (10) observed a significant effect of ozone on respiratory mortality rate during the
nonpeak period (December–August) but not the peak period
Am J Epidemiol. 2014;180(7):729–736
Ozone and Mortality Rate in East Asia 735
(September–November) in 4 cities of the Pearl River Delta of
Southern China. Outside China, Kim et al. (28) also reported
a moderately smaller effect of ozone in the summer than
in the whole year in Seoul, Korea. A multicity time-series
study in Japan suggested that significant effects of short-term
ozone exposure on total and cardiovascular mortality rate
among the elderly were restricted to the moderate season
(not the warm or cool season), whereas the effect on respiratory mortality rate was strongest in winter (11).
Heterogeneity tests for season-specific estimates also supported significant seasonal patterns for ozone’s associations
with mortality rate, but the magnitude of the heterogeneity
varied. Heterogeneity of ozone’s effect estimates among different temperature levels was also reported in the NMMAPS,
which suggested that high temperature enhanced the effects
of ozone, particularly in the northern regions (12).
It is plausible that the ozone–mortality rate association
peaks in the summer and warm seasons because of better
population exposure characterization when people spend
more time outdoors. Our results suggest that previously reported higher effect estimates in the winter are likely due to
the inadequate control of temperature as an important confounder. Numerous studies have illustrated that the health effects associated with changes in temperature could be lagged
for days or even weeks. Specifically, shorter lags (up to several days) are associated with hot temperatures and longer
lags (up to 3 or 4 weeks) with cold temperatures (18, 29).
We based our choice of optimal temperature control in the
main analysis on model fit performance. Previous Asian studies tended to adjust only for same-day temperature when exploring ozone’s effects, whereas studies from North America
and Europe have included separate terms of lagged temperatures in the models (3, 4, 17).
Nonetheless, several limitations should be noted. First, as
in most previous time-series studies, we averaged pollutant
measurements across monitors within a city. This results in
measurement error, which is difficult to quantify, especially
in 2-pollutant models. This measurement error may further
vary in magnitude and direction across these cities (30). Second, the lengths of study period, population size, and climatic
characteristics varied in different cities, which might decrease
the between-city comparability of our estimates. Third, we
were not able to investigate the observed heterogeneity more
thoroughly because there was little available information on
possible effect modifiers. Fourth, in the 2-pollutant models,
we did not evaluate the sensitivity of our results to the adjustment for fine particles because it was not a pollutant measured
routinely during our study periods; nevertheless, a prior study
in China has shown that ozone’s health effects might be independent from fine particles (31). Fifth, the definitions and
cutoffs of seasons may vary appreciably across regions; however, most of our cities were located in subtropical and warm
temperate zones with an IQR of annual average temperature
less than 5°C.
In summary, we found significant associations between
short-term exposure to ozone and increased daily mortality
rates in 21 cities of East Asia, which was generally consistent
with the results reported in North America and Western Europe. The ozone–mortality rate association varied considerably from season to season with a significant trough in
Am J Epidemiol. 2014;180(7):729–736
winter. Previous Asian findings of stronger associations in
the cool period were likely caused by inadequate control of
the confounding effects of temperature. Our results may contribute to targeted environmental and public health policies,
as well as improvements in the health risk assessment of
ozone pollution in East Asian countries.
ACKNOWLEDGMENTS
Author affiliations: Department of Environmental Health,
School of Public Health, Key Laboratory of Public Health
Safety of the Ministry of Education, Fudan University,
Shanghai, China (Renjie Chen, Jing Cai, Xia Meng, Haidong
Kan); Shanghai Key Laboratory of Atmospheric Particle Pollution and Prevention, Fudan University, Shanghai, China
(Renjie Chen, Jing Cai, Xia Meng, Xin Yang, Haidong
Kan); Department of Biostatistics, Graduate School of Public
Health, Seoul National University, Seoul, Korea (Ho Kim);
Department of Health Care Policy and Management, University of Tsukuba, Ibaraki, Japan (Yasushi Honda); Institute of
Occupational Medicine and Industrial Hygiene, School of
Public Health, National Taiwan University, Taipei, Taiwan
(Yue Leon Guo); and Department of Hygiene, Epidemiology, and Medical Statistics, Medical School, University of
Athens, Athens, Greece (Evangelia Samoli).
R.C., J.C., and X.M. contributed equally to this work and
should all be considered first author.
The study was supported by the National Basic Research
Program (973 program) of China (grant 2011CB503802);
the Global Research Laboratory (grant K2100400000110A0500-00710) through the National Research Foundation
of Korea funded by the Ministry of Science, Information,
Communication Technologies, and Future Planning; and
Health-X Seed Funding (CMB grant 13-131) of the Global
Health Institute, Fudan University.
Conflict of interest: none declared.
REFERENCES
1. Bell ML, Dominici F, Samet JM. A meta-analysis of time-series
studies of ozone and mortality with comparison to the National
Morbidity, Mortality, and Air Pollution Study. Epidemiology.
2005;16(4):436–445.
2. Stieb DM, Judek S, Burnett RT. Meta-analysis of time-series
studies of air pollution and mortality: update in relation to the
use of generalized additive models. J Air Waste Manag Assoc.
2003;53(3):258–261.
3. Peng RD, Samoli E, Pham L, et al. Acute effects of ambient
ozone on mortality in Europe and North America: results from
the APHENA Study. Air Qual Atmos Health. 2013;6(2):
445–453.
4. Gryparis A, Forsberg B, Katsouyanni K, et al. Acute effects
of ozone on mortality from the “Air Pollution and Health: a
European Approach” project. Am J Respir Crit Care Med.
2004;170(10):1080–1087.
5. Zanobetti A, Schwartz J. Is there adaptation in the ozone
mortality relationship: a multi-city case-crossover analysis.
Environ Health. 2008;7:22.
736 Chen et al.
6. Anenberg SC, Horowitz LW, Tong DQ, et al. An estimate
of the global burden of anthropogenic ozone and fine
particulate matter on premature human mortality using
atmospheric modeling. Environ Health Perspect. 2010;118(9):
1189–1195.
7. Zhang Y, Huang W, London SJ, et al. Ozone and daily mortality
in Shanghai, China. Environ Health Perspect. 2006;114(8):
1227–1232.
8. Wong CM, Ma S, Hedley AJ, et al. Effect of air pollution on
daily mortality in Hong Kong. Environ Health Perspect. 2001;
109(4):335–340.
9. Yang C, Yang H, Guo S, et al. Alternative ozone metrics
and daily mortality in Suzhou: the China Air Pollution and
Health Effects Study (CAPES). Sci Total Environ. 2012;426:
83–89.
10. Tao Y, Huang W, Huang X, et al. Estimated acute effects of
ambient ozone and nitrogen dioxide on mortality in the Pearl
River Delta of southern China. Environ Health Perspect. 2012;
120(3):393–398.
11. Ng CF, Ueda K, Nitta H, et al. Seasonal variation in the acute
effects of ozone on premature mortality among elderly
Japanese. Environ Monit Assess. 2013;185(10):8767–8776.
12. Ren C, Williams GM, Mengersen K, et al. Temperature
enhanced effects of ozone on cardiovascular mortality in
95 large US communities, 1987–2000: assessment using
the NMMAPS data. Arch Environ Occup Health. 2009;
64(3):177–184.
13. Armstrong B. Models for the relationship between ambient
temperature and daily mortality. Epidemiology. 2006;17(6):
624–631.
14. Wong CM, Vichit-Vadakan N, Kan H, et al. Public Health and
Air Pollution in Asia (PAPA): a multicity study of short-term
effects of air pollution on mortality. Environ Health Perspect.
2008;116(9):1195–1202.
15. Peng RD, Dominici F, Louis TA. Model choice in time series
studies of air pollution and mortality. J R Stat Soc Ser A Stat
Soc. 2006;169(2):179–203.
16. Samoli E, Aga E, Touloumi G, et al. Short-term effects
of nitrogen dioxide on mortality: an analysis within the
APHEA Project. Eur Respir J. 2006;27(6):1129–1138.
17. Bell ML, McDermott A, Zeger SL, et al. Ozone and short-term
mortality in 95 US urban communities, 1987–2000. JAMA.
2004;292(19):2372–2378.
18. Anderson BG, Bell ML. Weather-related mortality: How heat,
cold, and heat waves affect mortality in the United States.
Epidemiology. 2009;20(2):205–213.
19. Peng RD, Dominici F, Pastor-Barriuso R, et al. Seasonal
analyses of air pollution and mortality in 100 US cities. Am J
Epidemiol. 2005;161(6):585–594.
20. Bell ML, Ebisu K, Peng RD, et al. Seasonal and regional
short-term effects of fine particles on hospital admissions in 202
US counties, 1999–2005. Am J Epidemiol. 2008;168(11):
1301–1310.
21. Chen R, Peng RD, Meng X, et al. Seasonal variation in the acute
effect of particulate air pollution on mortality in the China
Air Pollution and Health Effects Study (CAPES). Sci Total
Environ. 2013;450-451:259–265.
22. Thurston GD, Ito K. Epidemiological studies of acute ozone
exposures and mortality. J Expo Anal Environ Epidemiol. 2001;
11(4):286–294.
23. Ito K, De Leon SF, Lippmann M. Associations between ozone
and daily mortality: analysis and meta-analysis. Epidemiology.
2005;16(4):446–457.
24. Levy JI, Carrothers TJ, Tuomisto JT, et al. Assessing the public
health benefits of reduced ozone concentrations. Environ
Health Perspect. 2001;109(12):1215–1226.
25. Anderson HR, Atkinson RW, Peacock JL, et al. Meta-Analysis
of Time-Series Studies and Panel Studies of Particulate Matter
(PM) and Ozone (O3). Report of a WHO Task Group.
Copenhagen, Denmark: World Health Organization; 2004.
26. Yan M, Liu Z, Liu X, et al. Meta-analysis of the Chinese studies
of the association between ambient ozone and mortality.
Chemosphere. 2013;93(6):899–905.
27. Dufour G, Eremenko M, Orphal J, et al. IASI observations
of seasonal and day-to-day variations of tropospheric ozone
over three highly populated areas of China: Beijing, Shanghai,
and Hong Kong. Atmos Chem Phys. 2010;10(8):3787–3801.
28. Kim SY, Lee JT, Hong YC, et al. Determining the threshold
effect of ozone on daily mortality: an analysis of ozone and
mortality in Seoul, Korea, 1995–1999. Environ Res. 2004;
94(2):113–119.
29. Guo Y, Barnett AG, Pan X, et al. The impact of temperature on
mortality in Tianjin, China: a case-crossover design with a
distributed lag nonlinear model. Environ Health Perspect.
2011;119(12):1719–1725.
30. Zeka A, Schwartz J. Estimating the independent effects of
multiple pollutants in the presence of measurement error: an
application of a measurement-error–resistant technique.
Environ Health Perspect. 2004;112(17):1686–1690.
31. Kan H, London SJ, Chen G, et al. Differentiating the effects of
fine and coarse particles on daily mortality in Shanghai, China.
Environ Int. 2007;33(3):376–384.
Am J Epidemiol. 2014;180(7):729–736