Estimated Acute Effects of Ambient Ozone and Nitrogen Dioxide on Mortality in the Pearl River Delta of Southern China

Background and objectives: Epidemiologic studies have attributed adverse health effects to air pollution; however, controversy remains regarding the relationship between ambient oxidants [ozone (O3) and nitrogen dioxide (NO2)] and mortality, especially in Asia. We conducted a four-city time-series study to investigate acute effects of O3 and NO2 in the Pearl River Delta (PRD) of southern China, using data from 2006 through 2008. Methods: We used generalized linear models with Poisson regression incorporating natural spline functions to analyze acute mortality in association with O3 and NO2, with PM10 (particulate matter ≤ 10 μm in diameter) included as a major confounder. Effect estimates were determined for individual cities and for the four cities as a whole. We stratified the analysis according to high- and low- exposure periods for O3. Results: We found consistent positive associations between ambient oxidants and daily mortality across the PRD cities. Overall, 10-μg/m3 increases in average O3 and NO2 concentrations over the previous 2 days were associated with 0.81% [95% confidence interval (CI): 0.63%, 1.00%] and 1.95% (95% CI: 1.62%, 2.29%) increases in total mortality, respectively, with stronger estimated effects for cardiovascular and respiratory mortality. After adjusting for PM10, estimated effects of O3 on total and cardiovascular mortality were stronger for exposure during high-exposure months (September through November), whereas respiratory mortality was associated with O3 exposure during nonpeak exposure months only. Conclusions: Our findings suggest significant acute mortality effects of O3 and NO2 in the PRD and strengthen the rationale for further limiting the ambient pollution levels in the area.

Substantial evidence supports the association between ambient air pollution, including particulate matter (PM), ozone (O 3 ), and nitrogen dioxide (NO 2 ), and mortality and morbidity from cardiopulmonary diseases (Bell et al. 2004;Brook et al. 2010;Dockery et al. 1993;Pope and Dockery 2006;Pope et al. 2002;Samet et al. 2000). However, although higher concentrations of air pollution are often found in Asia, published Asian research is still limited, making region-specific results difficult to interpret and compare with findings for populations in the developed world (Cao et al. 2011;Chen et al. 2004;Wong et al. 2008;Zhou et al. 2010).
Chinese megacities are some of the most air-polluted cities in the world (Chan and Yao 2008). There has been growing concern about air-pollution-related health effects in the Pearl River Delta (PRD), which recently has undergone rapid economic development and urbanization. The PRD lies in the coastal part of southern China and comprises nine cities in Guangdong Province and the special administrative regions of Hong Kong and Macao. This region accounts for only 0.5% of the Chinese geographic area but holds 4% of the Chinese population and produces about one-fifth of the total gross domestic product. Remarkable problems of traffic and photochemical air pollution have emerged in the PRD, in large part due to vehicle emissions and high sun exposure year-round (Shao et al. 2009). Photochemical reactions involving nitrogen oxides, volatile organic compounds, and hydroxyl radicals produce O 3 , a secondary air pollutant, in the presence of sunlight [Hofzumahaus et al. 2009;World Health Organization (WHO) 2005]. The PRD was one of the first regions in China to experience serious photochemical smog pollution: up to 171 ppb O 3 was measured in a suburb of Hong Kong, with sizable contributions likely from air mass transported from Guangdong cities in the PRD (So and Wang 2003). The PRD Regional Air Quality Monitoring Network, with 16 monitoring stations across the region, was established to facilitate research on pollution in the PRD (Hua et al. 2008;Zhang et al. 2008a).
Ambient O 3 and NO 2 , two major oxidants involved in photochemical processes, have been associated with adverse health effects. A systematic review indicates that the reaction between O 3 and biomolecules to form ozonides and free radicals triggers inflammatory responses and systemic oxidative stress in the cardiorespiratory system (Srebot et al. 2009). NO 2 is a highly reactive and nitrogencentered free radical that can induce airway inflammation (Kelly et al. 1996). Short-term exposure to ambient O 3 and NO 2 has adverse effects on pulmonary function, particularly in asthmatics, and may increase airway allergic inflammatory reactions, hospital admissions, and mortality (WHO 2004). Many epidemiologic studies suggested both short-and long-term effects of exposure to O 3 (Bell et al. 2004(Bell et al. , 2005Gryparis et al. 2004;Jerrett et al. 2005) and NO 2 (Hoek et al. 2002;Samoli et al. 2006) on total and cause-specific mortality. A few studies focused on both oxidants as reactive components of the photochemical air pollution mixture and reported significant mortality effects of short-term oxidant exposure (Saez et al. 2002;Touloumi et al. 1997).
Air monitoring data collected through the PRD monitoring network have made it possible to assess the health effects of ambient air pollution on a regional scale. In this multicity study, we estimated associations between ambient O 3 and NO 2 and mortality in the PRD (2006PRD ( -2008, using a time-series approach.

Materials and Methods
Study site description. Our study cities included Guangzhou, Foshan, Zhongshan, and Zhuhai in the PRD ( Figure 1). Guangzhou, the capital city of Guangdong Province, is a megacity that had 7.8 million urban residents in 2008; Foshan, Zhongshan, and Zhuhai had 3.8, 1.5, and 1.0 million residents, respectively. These cities have a typical monsoon-influenced climate with wet, hot summers and dry, Background and oBjectives: Epidemiologic studies have attributed adverse health effects to air pollution; however, controversy remains regarding the relationship between ambient oxidants [ozone (O 3 ) and nitrogen dioxide (NO 2 )] and mortality, especially in Asia. We conducted a fourcity time-series study to investigate acute effects of O 3 and NO 2 in the Pearl River Delta (PRD) of southern China, using data from 2006 through 2008. Methods: We used generalized linear models with Poisson regression incorporating natural spline functions to analyze acute mortality in association with O 3 and NO 2 , with PM 10 (particulate matter ≤ 10 μm in diameter) included as a major confounder. Effect estimates were determined for individual cities and for the four cities as a whole. We stratified the analysis according to high-and low-exposure periods for O 3 . results: We found consistent positive associations between ambient oxidants and daily mortality across the PRD cities. Overall, 10-μg/m 3 increases in average O 3 and NO 2 concentrations over the previous 2 days were associated with 0.81% [95% confidence interval (CI): 0.63%, 1.00%] and 1.95% (95% CI: 1.62%, 2.29%) increases in total mortality, respectively, with stronger estimated effects for cardiovascular and respiratory mortality. After adjusting for PM 10 , estimated effects of O 3 on total and cardiovascular mortality were stronger for exposure during high-exposure months (September through November), whereas respiratory mortality was associated with O 3 exposure during nonpeak exposure months only. conclusions: Our findings suggest significant acute mortality effects of O 3 and NO 2 in the PRD and strengthen the rationale for further limiting the ambient pollution levels in the area. cool to mild winters. The annual average precipitation is approximately 1,800 mm, the annual average temperature is 22-23°C, and the annual average relative humidity (RH) is 72-80%. The predominant southern or southeastern winds from the South China Sea during spring and summer bring relatively clean oceanic air, whereas northeastern winds carry air pollutants from close-vicinity northern cities in autumn and winter. These cities are typical of the PRD cities with respect to geographical, meteorological, and cultural conditions, although they vary in size and industrial structure.
Environmental data. Air pollution data were measured at five monitoring stations in the area: two in Guangzhou (one regional monitoring station in Wanqingsha and one in the central city park Luhu), and one station each in Foshan, Zhongshan, and Zhuhai ( Figure 1). All monitoring stations except the Luhu station are in urban areas with mixed residential and commercial activities. The monitors sample air about 10-20 m above ground level.
We obtained hourly concentrations of NO 2 , O 3 , sulfur dioxide (SO 2 ), carbon monoxide (CO), and PM 10 (PM with aerodynamic diameter ≤ 10 μm) from each station. NO 2 , SO 2 , O 3 , and CO were measured using chemiluminescence, fluorescence, ultraviolet, and infrared instruments, respectively; PM 10 was measured by tapered element oscillating microbalance. We calculated 24-hr average concentrations for PM 10 , NO 2 , SO 2 , and CO for days that had measurements for at least 18 of 24 hr. We calculated 8-hr (from 1000 hours to 1800 hours) average concentrations for O 3 for days with at least six of eight hourly measurements available. If monitor data for an individual pollutant were insufficient to calculate a daily average, all measurements from that day were excluded for that pollutant and monitor. Data from the two monitoring stations were averaged to derive concentrations for Guangzhou. Daily temperature and RH data for each city were obtained from the Chinese Academy of Meteorological Sciences. Missing data were not imputed.
Statistical analysis. Because daily mortality counts typically follow a Poisson distribution, we used Poisson regression models to evaluate the association between mortality and air pollution controlling for temperature, RH, seasonal patterns, and long-term trends using natural spline smoothers (Burnett et al. 2004;Samet and Katsouyanni 2006;Wood 2006). Degrees of freedom of the natural spline smoothers were determined by Akaike's information criterion (Hurvich et al. 1998) and generalized cross-validation. If there was overdispersion in the variance, we used the partial autocorrelation function (PACF) of the residuals to guide the selection of degrees of freedom until the absolute values of sum of PACF for lags up to 30 days reached minimum. Analyses were also adjusted for year, day of the week (DOW), and public holidays using categorical indicator variables. We adjusted for influenza epidemics by including an indicator variable that was assigned a value of 1 when the 7-day moving average of the respiratory mortality was greater than the 90th percentile of its city-specific distribution, and 0 otherwise (Samoli et al. 2006). Because the influenza variable was based on the distribution of respiratory mortality, and because previous studies (Braga et al. 2000;Touloumi et al. 2005) suggest that omitting control for influenza is unlikely to influence the association between air pollution and respiratory mortality, we adjusted for influenza only in models of total mortality and cardiovascular mortality . Residuals of each model were examined for discernible patterns and autocorrelation using residual plots and PACF plots, respectively [Health Effects Institute (HEI) 2010b].
Associations between mortality and average air pollutant concentrations on individual days (lag 0 to lag 6) and 2-day periods (lag 0-1 days and lag 1-2 days) were first examined in single-pollutant models. Preliminary analyses indicated that the largest pollutant effects were usually observed at lag 1-2 days (data not shown). Therefore, we report the excess risk (ER) of mortality and its 95% confidence interval (CI) associated with a 10-μg/m 3 increase in the average concentration of each pollutant during the previous 1-2 days. Single-day temperature and RH (lag 0 or lag 1 day) were used in our analyses, and the reported results were from models with lag 1 day covariates.
After establishing the final model that controlled for time trend, temperature, RH, year, DOW, public holiday, and influenza epidemics, we calculated city-specific estimates by fitting Poisson regression models for each city separately. We calculated Q-statistics to test the homogeneity of effect estimates among the study cities (α = 0.05) (DerSimonian and Laird 1986). Because the cities are all located along the estuary of the Pearl River, in close proximity to each other (Figure 1), and are similar with regard to natural and social factors, and because measurements from the five monitoring stations were correlated (Pearson correlation coefficients ranged from 0.64 to 0.88 for O 3 and from 0.43 to 0.83 for NO 2 ), we also generated combined effect estimates by summing the mortality data across the four cities and averaging the environmental data from the five monitoring stations. In the city-merged analyses, both singleand two-pollutant models were applied to estimate the effects of O 3 , NO 2 , and PM 10 adjusted for confounding by other pollutants. Two-pollutant models were restricted to pollutants with Pearson correlation coefficients < 0.6 to avoid multicollinearity. Furthermore, we stratified O 3 exposure as exposure during peak (September through November) and nonpeak (December through August) exposure periods identified previously for the area (Zhang et al. 2008b;Zheng et al. 2010) by using 2-df splines to control for time trend during the peak period and 6-df splines for the nonpeak period (Zanobetti and Schwartz 2008). In the stratified analysis, we adjusted for PM 10 , because stratum-specific effects of O 3 could be confounded by PM 10 (Bell et al. 2007).
Finally, we conducted a series of sensitivity analyses focused on O 3 to assess the impact of dropping model covariates, increasing or decreasing the degrees of freedom for time and meteorological spline variables by 25%, lagging temperature by 2-3 days or 4-6 days, and excluding days with daily concentrations of O 3 above the 95th or below the 5th percentile.
Analyses were performed using R (version 2.13.0; R Foundation for Statistical Computing, Vienna, Austria) with the mgcv package (version 1.5-5; Comprehensive R Archive Network, http://cran.r-project.org). Statistical significance was defined as p < 0.05. Table 1 summarizes the mortality data for the four PRD cities from 2006 through 2008. Average daily nonaccidental death counts were 83, 25, 21, and 9 for Guangzhou, Foshan, Zhongshan, and Zhuhai, respectively. About 56% of all non accidental deaths occurred in males, and 54-63% were attributed to cardiovascular or respiratory diseases. Table 2 presents the air pollution levels of the four cities from 2006 through 2008. Foshan had the highest average concentrations of PM 10 , NO 2 , SO 2 , and CO, and Zhongshan had the highest average concentration of O 3 . Monthly O 3 concentrations were highest during the peak exposure period from September through November (Figure 2). Based on combined data for the four cities, O 3 was the least correlated with other pollutants (Pearson correlation coefficients, -0.06 to 0.17), whereas NO 2 was highly correlated with PM 10 , SO 2 , and CO (correlation coefficients, 0.72-0.82). Temperature was positively correlated with O 3 only, whereas RH was negatively correlated with all the pollutants (Table 3).

Results
In the individual cities, adjusted ERs of total mortality in association with a 10-μg/m 3 increase in exposure (1-2 day lag, singlepollutant models) ranged from 0.22% to 0.64% for O 3 , 1.22% to 1.87% for NO 2 , and 0.37% to 0.74% for PM 10 (Table 4). In general, we observed stronger associations of O 3 , NO 2 , and PM 10 with cardiovascular and respiratory mortality than with total mortality. Although there was heterogeneity for the associations between PM 10 and respiratory mortality and also some variation in Foshan and Zhuhai, no significant heterogeneity was found in the effect estimates of O 3 and NO 2 among the four cities, which supported our analyses with city-merged data.
In the city-merged analyses, effect estimates for O 3 were moderately reduced but still significant after adjustment for PM 10 , NO 2 , SO 2 , and CO. Effect estimates for NO 2 and PM 10 were also attenuated but still significant after adjustment for O 3 ( Table 5). The variation of effect estimates among pollutants became smaller when assessment was made per interquartile range (IQR) increase. For instance, the effect estimates of O 3 and NO 2 on total mortality were 0.81% and 1.95% per 10-μg/m 3 increase for each pollutant ( . We also observed significant increased risk in cardiorespiratory mortality per unit increase in air pollution: a 10-μg/m 3 increase in O 3 exposure was associated with 0.79% (95% CI: 0.36, 1.22%), 1.17% (95% CI: 0.65, 1.70%), and 1.16% (95% CI: 0.56, 1.77%) increases in mortality from coronary, stroke, and COPD diseases, respectively (see Supplemental Material, Table 2).
Average concentrations of O 3 were 117.4 μg/m 3 in the peak exposure period (September through November) and 66.9 μg/m 3 in the nonpeak period. O 3 exposure was significantly associated with total mortality and cardiovascular mortality in both periods (Table 6). After adjustment for PM 10 , effect estimates for total and cardiovascular mortality increased in the peak period but decreased in the nonpeak period. Respiratory mortality was significantly associated with O 3 exposure during the nonpeak period only, and there was no evidence of an association during the peak period after adjusting for PM 10 .

Discussion
We estimated significant acute mortality effects associated with exposure to ambient oxidants in the PRD. In general, associations between O 3 and NO 2 exposure and daily mortality were homogeneous among the study cities. The estimated effects of O 3 were robust to adjustment for other pollutants (PM 10 , NO 2 , SO 2 , CO), and effect estimates for NO 2 were robust to adjustment for O 3 . Effect estimates for O 3 were larger for exposure during the peak period for total and cardiovascular mortality after adjustment for PM 10 , whereas for respiratory mortality, the association appeared to be limited to exposure during the nonpeak period. 10.6 ± 4.0 (1-28) 3.1 ± 2.3 (0-15) 2.2 ± 1.6 (0-9) 1.7 ± 1.4 (0-8) Respiratory 15.2 ± 5.2 (5-44) 5.3 ± 3.0 (0-25) 3.7 ± 2.2 (0-13) 1.1 ± 1.1 (0-6) COPD 7.7 ± 3.5 (0-27) 2.2 ± 1.6 (0-11) 2.3 ± 1.7 (0-11) 0.8 ± 0.9 (0-5) We observed much higher concentrations of O 3 in the PRD cities (annual mean, 70-85 μg/m 3 ) than those observed in North American cities (14-38 μg/m 3 ) (Samet and Katsouyanni 2006). Our analysis indicated significant increases of 0.81% (95% CI: 0.63%, 1.00%) in total mortality, 1.01% (95% CI: 0.71%, 1.32%) in cardiovascular mortality, and 1.33% (95% CI: 0.89%, 1.76%) in respiratory mortality, per 10-μg/m 3 increase in lag 1-2 day O 3 level in the PRD. Consistently, a multisite time-series study of 95 large U.S. urban communities estimated that a 10-ppb (≈ 20 μg/m 3 ) increase in the previous week's O 3 was associated with a 0.52% (95% CI: 0.27%, 0.77%) increase in daily mortality and a 0.64% (95% CI: 0.31%, 0.98%) increase in cardiovascular and respiratory mortality (Bell et al. 2004). A metaanalysis of 144 effect estimates from 39 timeseries studies also provided strong evidence of a short-term association between O 3 exposure and mortality, with larger estimated effects on cardiovascular and respiratory mortality than on total mortality (Bell et al. 2005). However, several other studies reported acute O 3 exposure effects on cardiovascular mortality but not respiratory mortality, including an analysis of seven Spanish cities in the EMECAM (Spanish Multicenter Study on the Relationship between Air Pollution and Mortality) project (Saez et al. 2002) and studies in Asia (HEI 2010b; Zhang et al. 2006). Wong et al. (2008) reported that effect estimates of O 3 were significant for total and cardiovascular mortality but only marginally significant for respiratory mortality in four Asian cities. In addition, a recent review of Asian studies reported positive but much smaller and inconsistent effect estimates for O 3 and mortality across cities (HEI 2010a). Considering the limited number of estimates available for meta-analysis, future studies should be conducted in more Asian cities in order to address the inconsistency.
Consistent with previous studies (Burnett et al. 2004;Michelozzi et al. 1998;Touloumi et al. 1997), our analysis indicated significant associations between short-term change in NO 2 and mortality in the PRD, with estimated increases of 1.95% (95% CI: 1.62%, 2.29%) in total mortality, 2.12% (95% CI: 1.58%, 2.65%) in cardiovascular mortality, and 3.48% (95% CI: 2.73%, 4.23%) in respiratory mortality, per 10-μg/m 3 increase in lag 1-2 day NO 2 concentrations. However, the magnitude of NO 2 effect estimates in our study was much greater than comparable estimates reported for Western populations (pooled estimates, 0.30-0.50% increases in total, cardiovascular, and respiratory mortality per 10-μg/m 3 increase in lag 0-1 day NO 2 concentrations) (Samoli et al. 2006). Our estimates were also larger than comparable  estimates for other Asian cities (HEI 2010b) and summary estimates of 0.98% (95% CI: 0.54%, 1.42%), 1.74% (95% CI: 0.85%, 2.63%), and 1.08% (95% CI: 0.59%, 1.56%) for total, cardiovascular, and respiratory mortality per 10-μg/m 3 increase in NO 2 based on a meta-analysis of Asian time-series studies (HEI 2010a). Ambient oxidants of O 3 and NO 2 are major air pollutants in the PRD. Estimated effects of O 3 were moderately reduced but still significant after adjustment for PM 10 , SO 2 , and CO in two-pollutant models. This is probably because O 3 , as a secondary pollutant, has formation paths in the environment different from those of PM 10 , SO 2 , and CO and thus does not typically covary with these pollutants. With weak correlations observed between O 3 and other pollutants, the mortality effect of O 3 exposure is at least partially independent of other pollutants, which is consistent with findings reported previously (Bell et al. 2007;Gryparis et al. 2004). Unlike O 3 , NO 2 could be a proxy marker of exposure to pollutants (including PM 10 , SO 2 , and CO) generated by the same sources, such as vehicle emissions, or may reflect the combined effects of pollutants related to traffic and atmospheric photochemical pollution in the PRD. The significant associations estimated for O 3 and NO 2 suggest that ambient oxidants may play important roles in initiating air pollutionrelated mortality effects in the PRD.
A major finding of our study was the differential mortality effects of O 3 estimated for exposures during the peak and nonpeak exposure periods. The PRD lies on the border between subtropical and tropical zones, and the highest O 3 concentrations often occur in autumn (September through November), in contrast with other geographic areas studied, where peak O 3 period occurs in summer (June through August), (Bell et al. 2005;Zanobetti and Schwartz 2008). Correlation analysis indicated that O 3 concentrations were positively correlated with temperature and negatively correlated with RH in the area. Although the formation of O 3 relies largely on sunlight, the diffusion and clearance of O 3 can be accelerated by precipitation in summer.
Therefore, peak levels of O 3 in the PRD are often observed in autumn, when it is drier than in the summer (Shao et al. 2009). In the stratified analysis of O 3 after adjustment for PM 10 , we observed higher risk of O 3 exposure for total and cardiovascular mortality in the peak period, which was consistent with many Western studies that reported significant associations in summer or warm months (April through September) when the O 3 level is higher (Bell et al. 2005;Gryparis et al. 2004;Ito et al. 2005;Schwartz 2005). Although the significantly higher association observed between O 3 and respiratory mortality in the nonpeak exposure period differed from those observed in Western studies, similar results have been reported for Hong Kong, where associations between respiratory mortality and O 3 exposure were stronger in cold seasons with lower O 3 levels (Wong et al. 1999(Wong et al. , 2001. People with respiratory diseases may be more sensitive to high O 3 exposure than are people with other diseases; thus, with very high mean level of 117.4 μg/m 3 measured in the peak period, the risks of respiratory mortality could be reduced because vulnerable subjects may have died before the O 3 concentration reached higher levels (Wong et al. 2001).
Limitations should be noted in interpreting the results of our study. First, exposure data were obtained from only one or two air pollution monitoring stations in each city. Air pollution exposure may be spatially autocorrelated across our study cities, which may violate the assumption of independent exposures among cities for the merged regression analyses. The limited number of air pollution monitoring stations and exposure data to assess spatial structure of air pollution between cities may yield biased variance of parameter estimates and inefficient significance tests in our study (Jerrett et al. 2005). Second, no PM 2.5 and component data were available for the study period. Compared with PM 10 , PM 2.5 is a better index of combustion source air pollution and has  a Two-pollutant models were limited to pollutants with Pearson correlation coefficients < 0.6; other covariates controlled for were the same as those in Table 4.  Table 4. *Significantly different from the peak period (p < 0.01).
volume 120 | number 3 | March 2012 • Environmental Health Perspectives a larger proportion of secondary particles. Franklin and Schwartz (2008) suggested that some secondary particle, such as particulate sulfate, may be partly responsible for observed O 3 effects. Therefore, adjusting for PM 10 , instead of PM 2.5 or its components, may overestimate the effect of O 3 .

Conclusion
We estimated significant increases in mortality associated with O 3 and NO 2 exposures in the PRD. The evidence of differential effects of O 3 on mortality from different diseases supports the need for further investigation of the pathophysiological mechanisms of O 3 -associated cardiovascular and respiratory effects. Our findings strengthen the rationale for further limiting ambient oxidant pollution in the PRD.