Environmental Research 136 (2015) 396–404
Contents lists available at ScienceDirect
Environmental Research journal homepage: www.elsevier.com/locate/envres
Smog episodes, fine particulate pollution and mortality in China Maigeng Zhou a,1, Guojun He b,1, Maoyong Fan c, Zhaoxi Wang d, Yang Liu e, Jing Ma d,f, Zongwei Ma e, Jiangmei Liu a, Yunning Liu a, Linhong Wang a,n, Yuanli Liu d,g,nn a
National Center for Chronic and Non-communicable Disease Control and Prevention, Beijing, China Division of Social Science, Division of Environment, and Economics Department, The Hong Kong University of Science and Technology, Hong Kong c Department of Economics, Ball State University, Muncie, IN, USA d Harvard School of Public Health, Cambridge, MA, USA e Rollins School of Public Health, Emory University, Atlanta, GA, USA f Harvard Medical School, Boston, MA, USA g Peking Union School of Public Health, Peking Union Medical School, Beijing, China b
art ic l e i nf o
a b s t r a c t
Article history: Received 11 June 2014 Received in revised form 30 August 2014 Accepted 19 September 2014
Background: Starting from early January 2013, northern China was hit by multiple prolonged and severe smog events which were characterized by extremely high-level concentrations of ambient fine particulate matter (PM2.5) with hourly peaks of PM2.5 over 800 mg/m3. However, the consequences of this severe air pollution are largely unknown. This study investigates the acute effect of the smog episodes and PM2.5 on mortality for both urban and rural areas in northern China. Data and methods: We collected PM2.5, mortality, and meteorological data for 5 urban city districts and 2 rural counties in Beijing, Tianjin and Hebei Province of China from January 1, 2013 through December 31, 2013. We employed the generalized additive models to estimate the associations between smog episodes or PM2.5 and daily mortality for each district/county. Results: Without any meteorological control, the smog episodes are positively and statistically significantly associated with mortality in 5 out of 7 districts/counties. However, the findings are sensitive to the meteorological factors. After controlling for temperature, humidity, dew point and wind, the statistical significance disappears in all urban districts. In contrast, the smog episodes are consistently and statistically significantly associated with higher total mortality and mortality from cardiovascular/respiratory diseases in the two rural counties. In Ji County, a smog episode is associated with 6.94% (95% Confidence Interval, 0.20 to 14.58) increase in overall mortality, and in Ci County it is associated with a 19.26% (95% CI, 6.66–33.34) increase in overall mortality. The smog episodes kill people primarily through its impact on cardiovascular and respiratory diseases. On average, a smog episode is associated with 11.66% (95% CI, 3.12–20.90) increase in cardiovascular and respiratory mortality in Ji County, and it is associated with a 22.23% (95% CI, 8.11–38.20) increase in cardiovascular and respiratory mortality in Ci County. A 10 μg/m3 increase in PM2.5 concentration is associated with 0.88% (95% CI, 0.3–1.46) increase in overall mortality and 1.2% (95% CI, 0.55–1.85) in Ji County. A 10 μg/m3 increase in PM2.5 concentration is associated with 0.55% (95% CI, 0.02 to 1.13) increase in overall mortality in Ci County. The findings suggest that the smog episodes and fine particulate have bigger and more detrimental impacts on rural residents, especially for those living close to big and polluted cities. Conclusions: The smog episodes and PM2.5 are statistically associated with mortality in rural areas of China. The associations for urban areas are not statistically significant. & 2014 Elsevier Inc. All rights reserved.
Keywords: Smog episode Ambient fine particulate Beijing air pollution Mortality
1. Introduction n Corresponding author at: National Center for Chronic and Non-communicable Disease Control and Prevention, Chinese Center for Disease Control and Prevention, Nanwei Road, Xicheng District, Beijing 100050, China. nn Corresponding author at: Peking Union School of Public Health, Room 833, Tower 2 Sun Dong An Plaza, No. 138 Wangfujin St. Dongcheng District, Beijing 100006, China. 1 These authors contribute equally to this manuscript.
http://dx.doi.org/10.1016/j.envres.2014.09.038 0013-9351/& 2014 Elsevier Inc. All rights reserved.
Air pollution, undoubtedly, is one of China's most pressing public health issues today. Many industrialized countries have had their share of severe air pollution problems in the past (such as Meuse Valley in Belgium in 1930 (Nemery et al., 2001), Pennsylvania in 1948 (Schreniz et al., 1949), and London in 1952 (U.K. Ministry of Health, 1954)). However, the intensity of the recent
M. Zhou et al. / Environmental Research 136 (2015) 396–404
serious air pollution events in China, their geographic extent, and the number of people affected are unprecedented. Starting from early January 2013, Beijing experienced multiple prolonged periods of severe smog. An important measure of the intensity of these mixtures of smoke and fog is the peak hourly concentrations of ambient fine particulate matter (PM2.5) which reached over 800 mg/m3 or more than 32 times higher than the World Health Organization's (WHO) recommended level (25 mg/ m3). These smog episodes affected a region that extended far beyond the Beijing metropolitan area. There is growing awareness that these episodes can cause health problems. For example, as Beijing suffered a sixth day of hazardous-level air pollution in February 2014, the World Health Organization urged the residents of the capital city to stay indoors. However, the WHO officials cautioned that they could not link recent pollution levels with local media reports of specific cases of lung cancer and other ailments. “We know it has an impact on health, but we don't know how much”, said Shih Young-Soo, the WHO's Western Regional Director. Did the high levels of air pollution during 2013 kill as many people as the 1952 London smog incident?2 Indeed, the consequences of China's severe air pollution are largely unknown. Several previous studies have examined the associations between short-term air pollution and health outcomes in China (Aunan and Pan, 2004; Chan and Yao, 2008; Chen et al., 2012a,, 2012b; Guo et al., 2009,, 2013,, 2010; He et al., 2002; Kan et al., 2008; Tao et al., 2012; Venners et al., 2003; Wong et al., 2008,, 2002). These studies generally found that higher mortality rates were associated with elevated air pollution. However, previous studies either (1) used PM10 measure, (2) focused on a single city, (3) analyzed less polluted episodes, or (4) studied only urban areas. Their results may lack external validity because none of them analyzed the smog events from 2013, and none of them looked at the air pollution effects in both urban and rural areas. Our study contributes to the literature by evaluating the acute effect of the smog episodes and PM2.5 concentrations on mortality for 7 districts/counties in 3 northern Chinese provinces, where large-scale smog episodes took place in early 2013. To the best of our knowledge, this is the first study examining the relationship between air pollution and the health of rural residents in China. To estimate the health effects of the severe air pollution events in 2013, we combined the most recent PM2.5 data published by the Ministry of Environmental Protection (MEP) with death records data from the Chinese Center for Disease Control and Prevention (CDC), and conducted this first study on estimating the health effects of the severe air pollution in 2013. In contrast to previous studies, all of which drew data only from cities, we include two rural counties to examine if differential air pollution effects exist. To control for possible confounding factors, we compared models on the bivariate association between air pollution and mortality with those containing other meteorological variables including daily temperature, humidity, dew point, and wind speed.
2 During the 5-day London Smog episode from December 5 to 9, 1952, the daily number of deaths registered tripled. In the weeks that followed, the medical services compiled statistics and found that the fog had killed roughly 4000 people, and another 8000 died in the weeks and months that followed (Greater London Authority, http://legacy.london.gov.uk/mayor/environment/air_quality/docs/ 50_years_on.pdf).
397
2. Data and methods 2.1. Data sources 2.1.1. Mortality data Mortality data are drawn from the Disease Surveillance Point System (DSPS) in the Chinese CDC. The DSPS, initiated in 1978, collected all death records for the surveillance locations each year. The data collected covered 71 districts/counties in 29 provinces from 1980 to 1989 and 145 districts/counties in 31 provinces from 1990 to 2000. The DSPS was overhauled following the SARS outbreak in 2003 and has since covered 161 districts/counties. In the event of a death, the doctor or decedent's family is required to fill out a death certificate and submit it to the DSPS. The mortality data includes basic demographic characteristics of the decedent and the cause of death. The causes of death are coded in the International Classification of Diseases 10 (ICD-10). Total mortality is classified by causes of death: cardiovascular diseases (I00-99), respiratory disease (J00-99), and all other diseases. For this study, we have daily number of deaths by age group, gender, and cause of death for all the DSPS districts/counties located in Beijing, Tianjin and Hebei Province. 2.1.2. Air pollution data PM2.5 concentrations are published by the MEP's ground monitoring stations and collected by the research team. In December 2012, PM2.5 was included in the Chinese National Ambient Air Quality Standard (CNAAQS, GB3095-2012). Since then, 24-h PM2.5 concentrations from all major Chinese cities have been published by the China National Environmental Monitoring Center (CNEMC) of MEP. We collected daily 24-h PM2.5 concentrations from all the monitoring stations located in Beijing, Tianjin and Hebei Province, then matched the monitoring sites with DSPS points based on the geodesic distance calculated from the longitude and latitude coordinates. Each DSPS point is assigned to its nearest monitoring site. If the distance between a DSPS point and its nearest air quality monitoring station was more than 30 km, we dropped it. 2.1.3. Weather data We collected daily weather information from January 1, 2013 to December 31, 2013, for all 7 city districts and counties in the study. These data were collected from The Weather Company's internet service known as the Weather Underground3 which is the world's largest network of weather stations (almost 10,000 stations in the United States and over 3000 across the rest of the world). As such it provides the most localized weather data currently available. We downloaded the following data for each location: daily precipitation, daily minimum, maximum and mean temperature, dew point, humidity, visibility miles and wind speed. 2.1.4. Study area We successfully matched mortality data, air pollution data and weather data for five city districts and two rural counties in Beijing, Tianjin and Hebei Province. Dongcheng District (city center) in Beijing (DC, BJ), Tongzhou District in Beijing (outer suburb) (TZ, BJ), Hongqiao District (inner city) in Tianjin (HQ, TJ), Haigang District in Qinhuangdao (HG, QHD), and Qiaodong District in Zhangjiakou (QD, ZJK) are urban the five city districts. Ji County in Tianjin (JC, TJ) and Ci County in Handan (CC, HD) are the two urban counties. The total populations in those districts/counties are 71,3602, 922,890, 647,673, 602,006, 287,249, 1,013,006, and 641,000 respectively for the year 2013. The male to female ratio ranges from 3
The data are available at www.weatherunderground.com.
398
M. Zhou et al. / Environmental Research 136 (2015) 396–404
estimate the degrees of freedom for each meteorological factor (Gu, 2013). The prediction results suggest that we use 7 degrees of freedom for dew point, 1 degree of freedom for temperature, relative humidity, and wind speed. To check the robustness of the results, we also conducted the analysis using up to 3 degrees of freedom for the smooth functions Si(Zit ) for each meteorological factor. We used the moving averages of current-day and previous-day concentrations of PM2.5 in the regressions, since the 2-day moving averages can pick up the health effects of both current day and previous day's exposure to air pollution. We also examined the air pollution effects with different lag structures (l = 0, 1, 2). We conducted the analyses separately for different age groups to explore the potential heterogeneous effects. We divided the sample into four age groups: o60, 60–69, 70–79, and Z80. Previous literature suggests that elder people are more vulnerable to fine particulate pollution, and we investigated if this is true in China during the smog episodes. We also investigated whether the smog episodes have heterogeneous effects on male and female mortality. Fig. 1. Geographic location of the study sites. The number associated with each district is the yearly averages of PM2.5 (mg/m3) in 2013.
0.95 to 1.05. Ji County in Tianjin and Ci County in Handan are rural counties, and the rest are urban districts. The geographical location of sampled districts/counties is shown in Fig. 1. Because the Chinese government did not start to monitor and publish PM2.5 concentrations until 2013, we only focused on a one-year window from January 1, 2013 through December 31, 2013. 2.2. Statistical methods We used a set of generalized additive models to estimate the associations between PM2.5 and daily mortality. Two basic models are adopted to examine the air pollution effects. The first model includes a set of dummy variables that indicates all the smog episodes. Any period during which PM2.5 concentrations exceed 100 mg/m3 for three or more consecutive days is defined as a smog episode. The models compare the mortality differences between the smog episodes and the regular days for each location. Specifically, the following model is estimated: p
log(E[yt ]) = α + βsmogt +
∑ Si(Zit , dfZi ) i=1
where yt is the number of deaths on day t ; smogt is a dummy indicator for the smog episodes. β represents the log-relative rate of mortality associated with a smoggy day. Zit are meteorological factors that are correlated with air pollution levels, and Si are their natural spline smooth functions. The second model uses specific PM2.5 concentrations in the regressions. For each location, we estimated:
3. Results Table 1 summarizes the descriptive statistics of daily number of deaths, PM2.5 concentrations, and weather conditions across all the locations. Overall, there were 11.26 deaths per day, including 6.50 from cardiovascular diseases and respiratory diseases. Of the daily deaths in our sample approximately 2.22 people were less than 60 years, 4.92 were between 60 and 79, and 4.13 were older than 80. The daily mean of PM2.5 concentration across all locations was 85.81 μg/m3, with a standard deviation of 74.47. The daily mean temperature, relative humidity, dew point and wind speed across all districts/counties were 53.98 °F, 51.74%, 36.26 °F and 5.37 mph, respectively. The air quality was particularly bad in January 2013. For example, in Beijing's Dongcheng District, the daily mean PM2.5 concentration was 211.13 μg/m3 in January 2013. The trends of PM2.5 concentration and daily mortality across all locations are shown in Fig. 2. The PM2.5 and mortality trends for each location is shown in Appendix A. PM2.5 concentration is positively associated with mortality. Both PM2.5 concentration and daily mortality are the highest during winter, and lowest during the summer. The fluctuations of PM2.5 concentration are dramatic, with the lowest value less than 10 μg/m3 and the highest value more than 350 μg/m3. The thick and short dashes indicate all the smog episodes throughout the year. Smog episodes are also positively associated with higher levels of mortality. Table 2 reports the results from regressing daily mortality on air pollution levels using the generalized additive models. The variable of interest is the smog episode indicator. Panel A presents the results on overall mortality. Without controlling for any Table 1 Summary statistics.
p
log(E[yt ]) = α + βxt − l +
∑ Si(Zit , dfZi ) i=1
where xt − l is the PM2.5 concentration on day t − l , and l is a day lag. We include daily average temperature, relative humidity, dew point, and wind speed in the regressions. We chose the degrees of freedom for each meteorological factor based on its best prediction for air pollution levels. Using degrees of freedom that predict best for air pollution levels is advantageous because they will produce unbiased or asymptotically unbiased estimates of the pollution log-relative risk (Dominici et al., 2004). We used the generalized cross-validation method to
Variable
Mean Std. Dev. Min
Daily death counts Daily cardiovascular and respiratory death counts Daily death counts (age o 60) Daily death counts (age 60–79) Daily death counts (age Z 80) PM2.5 concentrations (μg/m3) Temperature (°F) Humidity (%) Dew point (°F) Wind speed (mph)
11.26 6.50
6.42 4.53
2.22 1.89 4.92 3.14 4.13 3.27 85.81 74.47 53.40 20.92 51.53 20.81 35.23 24.48 5.75 2.96
0 0
Max 41 33
0 14 0 22 0 19 3 607 4 92 5 97 25 80 0 20
M. Zhou et al. / Environmental Research 136 (2015) 396–404
399
Fig. 2. The trends of daily PM2.5 and mortality across all locations. (Note: If PM2.5 concentrations exceed 100 mg/m3 for more than 3 consecutive days (including 3 days), we define it as a smog episode.) Table 2 Percent change of daily mortality associated with a smog episode. Panel A. Percent change in mortality associated with a smog episode Smog 95% CI Smog 95% CI DC, BJ TZ, BJ HQ, TJ HG, QHD QD, ZJK JC, TJ CC, HD Weather controls
9.82n 9.11n 6.71nn 2.66 12.07 23.42n 27.05n No
3.59 1.60 0.88 6.60 2.38 15.20 14.15
16.43 17.18 14.89 12.83 28.65 32.23 41.40
0.22 0.28 2.72 1.08 8.39 6.94nn 19.26n Yes
6.57 7.90 5.64 8.94 6.78 0.20 6.66
6.56 7.97 11.82 12.20 26.04 14.58 33.34
Panel B. Percent change in CVR and non-CVR mortality associated with a smog episode CVR DC, BJ TZ, BJ HQ, TJ HG, QHD QD, ZJK JC, TJ CC, HD Weather controls
1.25 6.11 0.76 3.63 16.05 11.66n 22.23n Yes
Non-CVR 7.78 4.35 9.27 9.98 4.95 3.12 8.11
11.17 17.71 11.90 19.31 41.68 20.90 38.20
1.88 7.83 5.99 1.58 1.02 2.35 13.46 Yes
11.42 17.92 6.76 15.99 19.58 13.79 3.50
8.68 3.50 20.49 15.31 26.89 10.60 33.40
The 95% CI are reported in the parenthesis. n
5% statistically significant. 10% statistically significant.
nn
weather conditions, the association between the smog episode indicator and daily mortality is positive in all 7 locations. For Dongcheng District of Beijing, Tongzhou District of Beijing, Ji County in Tianjin, and Ci County in Handan, the associations between smog and mortality are statistically significant at the 5% level. For Hongqiao District in Tianjin, the association is statistically significant at the 10% level. These results are consistent with the patterns in Fig. 2. However, when we include smooth functions from the meteorological factors, the coefficients of the smog indicator become insignificant for all five urban locations. The significant associations remain for the two rural counties. A smog episode is associated with a 6.94% (95% Confidence Interval, 0.20 to 14.98) increase in daily mortality for Ji County in Tianjin, and the association is statistically significant at the 10% level. A smog episode is associated with a 19.26 (95% CI, 6.66–33.34)% increase in daily mortality for Ci County in Handan, and the results are significant at the 5% level. Panel B of Table 2 presents the results on cardiovascular-respiratory (CVR) and non-CVR mortality separately. For CVR mortality,
the results are similar to the overall mortality in Panel A. Only in the two rural counties can we find that the smog episodes are statistically associated with daily CVR mortality. A smog episode is associated with an 11.66% (95% CI, 3.12–20.90) and a 22.23% (95% CI, 8.11– 38.20) increase in Ji County and Ci County, respectively. However, the smog episodes are not associated with deaths caused by non-CVR diseases. Thus the excessive deaths during the smog episodes in both rural counties are mainly from CVR diseases. No statistically significant associations are found for any urban districts. In Table 3, we focused on the two rural counties, Ji County and Ci County, where the associations between smog episodes and mortality are robust and statistically significant. We investigate the heterogeneous effects of smog episodes across different age groups or genders by separately estimating the models for each age group and gender. In Panel A, the results show that the smog episodes are statistically significant and associated with higher levels of mortality for the youngest age group (o 60) in both counties. In Ji County, smog episodes are associated with a 13.18% Table 3 Heterogeneous effects of smog on mortality. JC, TJ Smog
CC, HD 95% CI
Smog
Panel A. Heterogeneous effects by age Overall mortality Age o60 13.18n 2.13 30.88 0.47 Age 60–79 1.50 7.81 11.74 Age Z80 1.12 25.42 CVR mortality Age o60 Age 60–79 Age Z80 0.87
23.10n 9.92 9.54
0.39 1.88 2.54
52.12 23.14 23.12
17.11nn
95% CI
2.52
20.89n
0.86
26.64n 17.11nn
1.72 2.52
20.36nn 44.18 33.77 12.62nn 47.42 63.18 33.77 24.70nn 54.14
Panel B. Heterogeneous effects by gender Male 6.06 Female CVR mortality Male 9.07 Female
7.57
1.58
17.57
8.48n
1.24
19.15
11.62n
0.15
24.78
11.36n
0.02
24.04
The 95% CI are reported in the parenthesis. n
10% statistically significant. 5% statistically significant.
nn
14.77n
15.37n
0.34
21.38nn 38.92 32.17
1.82
27.21nn 48.35 35.57
400
M. Zhou et al. / Environmental Research 136 (2015) 396–404
4. Discussion
Table 4 Percent change in mortality associated with a 10 mg/m3 change in PM2.5.
DC, BJ TZ, BJ HQ, TJ HG, QHD QD, ZJK JC, TJ CC, HD Weather controls
PM2.5
95% CI
0.40n 0.77nn 0.59nn 0.64 1.70 1.31nn 1.18nn No
0.00 0.31 0.07 0.35 0.72 0.74 0.72
0.79 1.23 1.12 1.64 4.13 1.88 1.64
PM2.5
95% CI
0.01 0.45 0.58* 0.57 1.91 0.88** 0.55* Yes
0.47 0.11 0.06 0.51 1.19 0.30 0.02
0.48 1.01 1.22 1.65 5.02 1.46 1.13
The 95% CI are reported in the parenthesis. n
10% statistically significant. 5% statistically significant.
nn
increase in mortality for people less than 60 years of age. The estimate is even higher in Ci County, a 20.36% increase. For age group 60–79, the estimates are typically smaller. In fact in Ji County the smog episodes are not associated with the mortality for age group 60–79 at all. For the oldest age group (Age Z80), smog episodes are positively associated with mortality and the results are statistically significant. For CVR mortality, the qualitative results are similar to the overall mortality. The smog episodes have the largest impact on the youngest age group (o 60) and the oldest age group (Age Z80). In Panel B, we study the association between smog and mortality separately for males and females. In Ji County, the results from males are quite similar to those for females. In Ci County, the smog episodes have a greater impact on male mortality. A smog episode is associated with a 21.38% increase in overall male mortality, and a 14.77% increase in female mortality. For both counties and genders, air pollution has greater impacts on CVR mortality than overall mortality. To complement the analysis from the smog model, we provided the results from the PM2.5 in Table 4, in which we regressed daily death counts on 2-day moving averages of current-day and previous-day concentrations of PM2.5. The associations between PM2.5 and daily mortality vary by locations, but the qualitative results are similar to the smog model in Panel A of Table 2. We observe positive and statistically significant associations of PM2.5 with overall mortality in 5 out of 7 locations: Dongcheng District of Beijing, Tongzhou District of Beijing, Hongqiao District in Tianjin, Ji County in Tianjin, and Ci County in Handan. However, the results are sensitive to the inclusion of meteorological factors especially for the urban locations. After we adjust for the meteorological confounders, the coefficients of PM2.5 are still statistically significant only for the two rural counties: Ji County and Ci County. A 10 μg/m3 increase in 2-day average of PM2.5 concentration is associated with a 0.88% (95% CI, 0.30–1.46) and a 0.55% (95% CI, 0.02 to 1.13) increase in daily mortality for Ji County and Ci County, respectively. We also find weak evidence that PM2.5 is associated with daily mortality in Hongqiao District in Tianjin, which is statistically significant at the 10% level. Table 5 Social-economic conditions (2006–2010 average). District
Share of Ag. (%)
DC&TZ, BJ 0.90 HG, QHD 1.41 HQ, TJ 1.44 QJ, ZZK 2.00 CC, TS 9.10 JC, TJ 13.49
GDP PC (Yuan)
Population density (Person/km2)
# Of hospitals
# Of doctors
Region
62,390 46,187 61,197 27,688 21,552 17,645
944 2231 1067 2347 630 519
599 54 286 46 24 33
52,463 3095 23,497 2609 1055 4046
Urban Urban Urban Urban Rural Rural
We conducted an exploratory study on health effects of severe air pollution during 2013 in northern China in selected city districts and rural counties of Beijing, Tianjin, and Hebei Province. We collected comprehensive data on mortality, air pollution and weather conditions in both urban and rural areas in China. To the best of our knowledge, this is the first study to include an examination of the relationship between air pollution and the health of rural residents in China. In the matching process, we used more accurate readings of PM2.5 concentration for our study locations, while most previous studies simply average the data from various monitor stations or use a single reading in the U.S. embassy and four consulates as a proxy for population exposure level to particulate matter. As is evident in our data, the concentration of PM2.5 can differ dramatically across monitor sites on a daily basis. The approximating or averaging process will potentially introduces substantial measurement error bias and the direction of the bias is uncertain (Sarnat et al., 2005). Our measure of population exposure is more accurate because we match the monitor station with the DSPS locations within 30 km, and the estimated results suffer less from the potential measurement error bias. Our statistical analysis suggests that, without taking into account the weather conditions, the smog episodes are typically positively and statistically significantly associated with higher mortality. However, the associations in urban areas are sensitive to the meteorological factors. After we adjust for a flexible spline of temperature, humidity, dew point and wind, the statistical significance disappears for all urban areas. We also investigate the association between mortality and 2-day moving average of PM2.5 concentrations. The results are consistent with the smog model. Overall the results in urban areas suggest that the health effects of the smog might not be as bad as people speculated. In sharp contrast to the urban areas, we find that the smog episodes are consistently and statistically significantly associated with higher levels of mortality in the two rural counties, Ji County of Tianjin and Ci County of Handan. The results are robust to the inclusion of the meteorological factors in the model and across statistical specifications. On average, a smog episode is associated with 6.94% higher overall mortality in Ji County, and 19.26% higher mortality in Ci County. The different results between rural and urban areas can be explained by several factors. First, urban residents are typically more aware of the potential air pollution effects on health. During the smoggy days, vulnerable people (elderly and sick) were warned to stay indoors to reduce outdoor air pollution exposure. In contrast, rural residents are less aware of the potential harmful effects of air pollution. Second, the protective measures, such as wearing masks and installing air filters at home and workplace, have become very common in urban areas. Compared with urban residents, rural residents are less likely to take protective measures simply because they cannot afford them. Finally, rural residents lack immediate access to emergent medical assistance. When the smog events trigger heart attacks or acute respiratory diseases, rural residents are more likely to die. As shown in Table 5, the share of agricultural industry in GDP is the highest and the GDP per capita is the lowest in the two rural counties, Ji County and Ci County. Compared with the urban districts, the population density in the rural counties is smaller, and they have fewer hospitals and doctors. The heterogeneous effects of air pollution in the two rural counties reveal another striking result: the smog episodes have a greater impact on people who are less than 60 years of age in rural areas. The results seem to contradict the conventional wisdom that the elderly are more likely to be affected by air pollution than
M. Zhou et al. / Environmental Research 136 (2015) 396–404
Fig. A1. The trends of PM2.5 and Daily Mortality in Each Location.
401
402
M. Zhou et al. / Environmental Research 136 (2015) 396–404
Fig. A1. (continued)
the young. This controversial results may be explained by the fact that a young adult's exposure to air pollution can be very high in rural areas. In fact, the young adult population is the primary labor
force in rural areas. The majority of rural residents have to do outdoor farming work during the day. If the outdoor air quality is worse than the indoor air quality in those areas, it is possible that
Table B1 The association between smog episode and mortality using different degrees of freedom. Smog DC, BJ TZ, BJ HQ, TJ JC, TJ HG, QHD CC, HD QD, ZJK Temperature Humidity Dew Point Wind Speed
95% CI
0.02 6.12 0.45 6.97 3.87 4.38 8.13** 1.12 0.18 9.32 16.32** 4.22 8.22 6.41 Linear function Linear function Linear function Linear function
Smog 6.56 8.46 12.84 15.62 10.68 29.84 25.15
0.18 0.05 2.17 7.07* 1.09 17.30** 9.41 Natural Natural Natural Natural
95% CI
spline spline spline spline
6.36 7.40 6.17 0.02 8.94 4.86 5.90 with 2 df with 2 df with 2 df with 2 df
Smog 6.41 8.09 11.25 14.66 12.21 31.22 27.22
0.57 0.74 2.53 6.86* 1.46 17.89** 8.76 Natural Natural Natural Natural
95% CI
spline spline spline spline
6.79 6.86 5.92 0.27 8.66 5.47 6.51 with 3 df with 3 df with 3 df with 3 df
6.06 8.95 11.74 14.50 12.70 31.78 26.53
M. Zhou et al. / Environmental Research 136 (2015) 396–404
403
Table C1 The association between mortality and current day, 1-day lag and 2-day lag PM2.5 concentration. Panel A: without control Current
95% CI
DC, BJ TZ, BJ HQ, TJ JC, TJ HG, QHD CC, HD QD, ZJK Weather
0.08 0.19 0.21 0.40 0.26 0.71 0.85 No
0.42** 0.60** 0.66** 0.90** 0.53 1.12** 1.16 No
0.76 1.00 1.11 1.41 1.32 1.54 3.17 No
1-Day lag
95% CI
0.15 0.58** 0.21 1.10** 0.54 1.00** 1.29
0.20 0.17 0.26 0.60 0.25 0.57 0.71
1-Day lag
95% CI
0.01 0.26 0.13 0.77** 0.43 0.35 0.89
0.39 0.18 0.39 0.31 0.38 0.16 1.50
0.51 0.98 0.68 1.60 1.33 1.44 3.30
2-Day lag
95% CI
0.03 0.27 0.10 1.08** 0.05 0.67** 1.19
0.33 0.14 0.57 0.58 0.87 0.20 0.76
2-Day lag
95% CI
0.18 0.13 0.27 0.60** 0.41 0.19 0.64
0.53 0.54 0.77 0.17 1.25 0.72 1.43
0.38 0.68 0.38 1.58 0.77 1.14 3.15
Panel B: with control
DC, BJ TZ, BJ HQ, TJ JC, TJ HG, QHD CC, HD QD, ZJK Weather
Current
95% CI
0.03 0.42 0.72** 0.54** 0.52 0.60** 1.03 Yes
0.42 0.11 0.18 0.03 0.37 0.08 1.37 Yes
0.47 0.95 1.26 1.04 1.41 1.12 3.44 Yes
the smog episodes kill more outdoor laborers than elderly people who spend most or all of their time indoors. Death of a young person in a rural family could be disastrous if the person is the major income earner. Therefore, alerting the rural residents about the danger of smog and educating them how to take preventive measures might have huge positive social impacts in rural areas. Overall, findings in this study suggest that the smog episodes and fine particulates air pollution have greater and more detrimental impacts on rural residents, especially those living close to big polluted cities. Compared with urban residents who are constantly alerted by the media about the severe air pollution, the rural residents are much less aware of and prepared for air pollution and don't usually take avoidance measures. As a consequence, the estimates in rural areas are closer to the true effects of air pollution. Previous studies only focus on urban areas where both air pollution data and mortality data are available (see Shang et al., 2013, for a recent review). Given the fact that more than half of Chinese people live in rural areas, understanding the health risks air pollution imposed on rural people is long overdue. We hope this preliminary study can motivate more academic research in this underexplored area. We conducted a variety of sensitivity checks and the results are reported as online Appendices. For the smog model, we checked whether the results are robust to the inclusion of control variables and the choices of different degrees of freedom in the spline function. We started by analyzing a linear function of each meteorological factor, and then extended the analysis by changing the choice degrees of freedom. The regression results, as summarized in Appendix B, are quite robust to the choices of different degrees of freedom in the spline functions, as long as some minimal representation is included. For the PM2.5 model, we experimented with different lag structures of PM2.5 for each location. The results are reported in Appendix C. In most cases mortality is either statistically significantly associated with current day PM2.5 or with one-day lagged PM2.5. So using a 2-day moving average picks up the average effect of both. The estimates using single day PM2.5 are either slightly smaller or slightly larger than those using the twoday moving averages, depending on the underlying dynamics of air pollution effects. For districts/counties showing no statistically significant associations between two-day moving average PM2.5 concentrations and mortality, a different lag structure made no difference.
0.36 0.70 0.66 1.24 1.24 0.85 3.28
0.17 0.28 0.22 1.04 0.43 0.34 2.71
5. Conclusion We find a significant association between the smog episodes (or PM2.5 concentration) and mortality in rural areas of China. The associations are robust to inclusion of meteorological factors and model specifications. Young rural residents ( o60) are also more likely to die during the smog period than young urban residents. The smog episodes and fine particulates air pollution had bigger and more detrimental impacts on rural residents, especially those living close to big polluted cities.
Acknowledgments This study is partially funded by a seed grant from the China Medical Board, United States (Grant No. 13-137) to Peking Union School of Public Health, and the National Basic Research Program of China (“973 Program”) (No. 2012CB955500).
Appendix A See appendix Fig. A1.
Appendix B See appendix Table B1.
Appendix C See appendix Table C1.
References Aunan, K., Pan, X.-C., 2004. Exposure-response functions for health effects of ambient air pollution applicable for China – a meta-analysis. Sci. Total Environ. 329, 3–16. Chan, C.K., Yao, X., 2008. Air pollution in mega cities in China. Atmos. Environ. 42, 1–42.
404
M. Zhou et al. / Environmental Research 136 (2015) 396–404
Chen, R., et al., 2012a. Short-term exposure to sulfur dioxide and daily mortality in 17 Chinese cities: the China air pollution and health effects study (CAPES). Environ. Res. 118, 101–106. Chen, R., et al., 2012b. Association of particulate air pollution with daily mortality: the China air pollution and health effects study. Am. J. Epidemiol. 175, 1173–1181. Dominici, F., et al., 2004. Improved semiparametric time series models of air pollution and mortality. J. Am. Stat. Assoc. 99, 938–948. Gu, C., 2013. Smoothing Spline ANOVA Models, vol. 297, Springer. Guo, Y., et al., 2009. The association between fine particulate air pollution and hospital emergency room visits for cardiovascular diseases in Beijing, China. Sci. Total Environ. 407, 4826–4830. Guo, Y., et al., 2013. The burden of air pollution on years of life lost in Beijing, China, 2004–08: retrospective regression analysis of daily deaths. Br. Med. J. 347, f7139. Guo, Y., et al., 2010. The relationship between particulate air pollution and emergency hospital visits for hypertension in Beijing, China. Sci. Total Environ. 408, 4446–4450. He, K., et al., 2002. Urban air pollution in china: current status, characteristics, and progress. Ann. Rev. Energy Environ. 27, 397–431. Kan, H., et al., 2008. Season, sex, age, and education as modifiers of the effects of outdoor air pollution on daily mortality in Shanghai, China: The Public Health and Air Pollution in Asia (PAPA) Study. Environ. Health Perspect. 116, 1183–1188. Nemery, B., et al., 2001. The Meuse Valley fog of 1930: an air pollution disaster. Lancet 357, 704–708.
Sarnat, J.A., et al., 2005. Ambient gas concentrations and personal particulate matter exposures: implications for studying the health effects of particles. Epidemiology 16, 385–395 (10.1097/01.ede.0000155505.04775.33). Schreniz, H., et al., 1949. Air Pollution in Donora, PA., Epidemiology of the Unusual Smog Episode of October 1948; Preliminary Report. Public Health Service Bulletin, PHS. Shang, Y., et al., 2013. Systematic review of Chinese studies of short-term exposure to air pollution and daily mortality. Environ. Int. 54, 100–111. Tao, Y., et al., 2012. Estimated acute effects of ambient ozone and nitrogen dioxide on mortality in the Pearl River Delta of southern China. Environ. Health Perspect. 120, 393–398. U.K. Ministry of Health, 1954. Mortality and Morbidity during the London Fog of December 1952. Reports on Public Health and Medical Subjects no. 95. Ministry of Health, London. Venners, S.A., et al., 2003. Particulate matter, sulfur dioxide, and daily mortality in Chongqing, China. Environ. Health Perspect. 111, 562–567. Wong, C.-M., et al., 2008. Public Health and Air Pollution in Asia (PAPA): a multicity study of short-term effects of air pollution on mortality. Environ. Health Perspect. 116, 1195–1202. Wong, T.W., et al., 2002. Associations between daily mortalities from respiratory and cardiovascular diseases and air pollution in Hong Kong, China. Occup. Environ. Med. 59, 30–35.
Contents lists available at ScienceDirect
Environmental Research journal homepage: www.elsevier.com/locate/envres
Smog episodes, fine particulate pollution and mortality in China Maigeng Zhou a,1, Guojun He b,1, Maoyong Fan c, Zhaoxi Wang d, Yang Liu e, Jing Ma d,f, Zongwei Ma e, Jiangmei Liu a, Yunning Liu a, Linhong Wang a,n, Yuanli Liu d,g,nn a
National Center for Chronic and Non-communicable Disease Control and Prevention, Beijing, China Division of Social Science, Division of Environment, and Economics Department, The Hong Kong University of Science and Technology, Hong Kong c Department of Economics, Ball State University, Muncie, IN, USA d Harvard School of Public Health, Cambridge, MA, USA e Rollins School of Public Health, Emory University, Atlanta, GA, USA f Harvard Medical School, Boston, MA, USA g Peking Union School of Public Health, Peking Union Medical School, Beijing, China b
art ic l e i nf o
a b s t r a c t
Article history: Received 11 June 2014 Received in revised form 30 August 2014 Accepted 19 September 2014
Background: Starting from early January 2013, northern China was hit by multiple prolonged and severe smog events which were characterized by extremely high-level concentrations of ambient fine particulate matter (PM2.5) with hourly peaks of PM2.5 over 800 mg/m3. However, the consequences of this severe air pollution are largely unknown. This study investigates the acute effect of the smog episodes and PM2.5 on mortality for both urban and rural areas in northern China. Data and methods: We collected PM2.5, mortality, and meteorological data for 5 urban city districts and 2 rural counties in Beijing, Tianjin and Hebei Province of China from January 1, 2013 through December 31, 2013. We employed the generalized additive models to estimate the associations between smog episodes or PM2.5 and daily mortality for each district/county. Results: Without any meteorological control, the smog episodes are positively and statistically significantly associated with mortality in 5 out of 7 districts/counties. However, the findings are sensitive to the meteorological factors. After controlling for temperature, humidity, dew point and wind, the statistical significance disappears in all urban districts. In contrast, the smog episodes are consistently and statistically significantly associated with higher total mortality and mortality from cardiovascular/respiratory diseases in the two rural counties. In Ji County, a smog episode is associated with 6.94% (95% Confidence Interval, 0.20 to 14.58) increase in overall mortality, and in Ci County it is associated with a 19.26% (95% CI, 6.66–33.34) increase in overall mortality. The smog episodes kill people primarily through its impact on cardiovascular and respiratory diseases. On average, a smog episode is associated with 11.66% (95% CI, 3.12–20.90) increase in cardiovascular and respiratory mortality in Ji County, and it is associated with a 22.23% (95% CI, 8.11–38.20) increase in cardiovascular and respiratory mortality in Ci County. A 10 μg/m3 increase in PM2.5 concentration is associated with 0.88% (95% CI, 0.3–1.46) increase in overall mortality and 1.2% (95% CI, 0.55–1.85) in Ji County. A 10 μg/m3 increase in PM2.5 concentration is associated with 0.55% (95% CI, 0.02 to 1.13) increase in overall mortality in Ci County. The findings suggest that the smog episodes and fine particulate have bigger and more detrimental impacts on rural residents, especially for those living close to big and polluted cities. Conclusions: The smog episodes and PM2.5 are statistically associated with mortality in rural areas of China. The associations for urban areas are not statistically significant. & 2014 Elsevier Inc. All rights reserved.
Keywords: Smog episode Ambient fine particulate Beijing air pollution Mortality
1. Introduction n Corresponding author at: National Center for Chronic and Non-communicable Disease Control and Prevention, Chinese Center for Disease Control and Prevention, Nanwei Road, Xicheng District, Beijing 100050, China. nn Corresponding author at: Peking Union School of Public Health, Room 833, Tower 2 Sun Dong An Plaza, No. 138 Wangfujin St. Dongcheng District, Beijing 100006, China. 1 These authors contribute equally to this manuscript.
http://dx.doi.org/10.1016/j.envres.2014.09.038 0013-9351/& 2014 Elsevier Inc. All rights reserved.
Air pollution, undoubtedly, is one of China's most pressing public health issues today. Many industrialized countries have had their share of severe air pollution problems in the past (such as Meuse Valley in Belgium in 1930 (Nemery et al., 2001), Pennsylvania in 1948 (Schreniz et al., 1949), and London in 1952 (U.K. Ministry of Health, 1954)). However, the intensity of the recent
M. Zhou et al. / Environmental Research 136 (2015) 396–404
serious air pollution events in China, their geographic extent, and the number of people affected are unprecedented. Starting from early January 2013, Beijing experienced multiple prolonged periods of severe smog. An important measure of the intensity of these mixtures of smoke and fog is the peak hourly concentrations of ambient fine particulate matter (PM2.5) which reached over 800 mg/m3 or more than 32 times higher than the World Health Organization's (WHO) recommended level (25 mg/ m3). These smog episodes affected a region that extended far beyond the Beijing metropolitan area. There is growing awareness that these episodes can cause health problems. For example, as Beijing suffered a sixth day of hazardous-level air pollution in February 2014, the World Health Organization urged the residents of the capital city to stay indoors. However, the WHO officials cautioned that they could not link recent pollution levels with local media reports of specific cases of lung cancer and other ailments. “We know it has an impact on health, but we don't know how much”, said Shih Young-Soo, the WHO's Western Regional Director. Did the high levels of air pollution during 2013 kill as many people as the 1952 London smog incident?2 Indeed, the consequences of China's severe air pollution are largely unknown. Several previous studies have examined the associations between short-term air pollution and health outcomes in China (Aunan and Pan, 2004; Chan and Yao, 2008; Chen et al., 2012a,, 2012b; Guo et al., 2009,, 2013,, 2010; He et al., 2002; Kan et al., 2008; Tao et al., 2012; Venners et al., 2003; Wong et al., 2008,, 2002). These studies generally found that higher mortality rates were associated with elevated air pollution. However, previous studies either (1) used PM10 measure, (2) focused on a single city, (3) analyzed less polluted episodes, or (4) studied only urban areas. Their results may lack external validity because none of them analyzed the smog events from 2013, and none of them looked at the air pollution effects in both urban and rural areas. Our study contributes to the literature by evaluating the acute effect of the smog episodes and PM2.5 concentrations on mortality for 7 districts/counties in 3 northern Chinese provinces, where large-scale smog episodes took place in early 2013. To the best of our knowledge, this is the first study examining the relationship between air pollution and the health of rural residents in China. To estimate the health effects of the severe air pollution events in 2013, we combined the most recent PM2.5 data published by the Ministry of Environmental Protection (MEP) with death records data from the Chinese Center for Disease Control and Prevention (CDC), and conducted this first study on estimating the health effects of the severe air pollution in 2013. In contrast to previous studies, all of which drew data only from cities, we include two rural counties to examine if differential air pollution effects exist. To control for possible confounding factors, we compared models on the bivariate association between air pollution and mortality with those containing other meteorological variables including daily temperature, humidity, dew point, and wind speed.
2 During the 5-day London Smog episode from December 5 to 9, 1952, the daily number of deaths registered tripled. In the weeks that followed, the medical services compiled statistics and found that the fog had killed roughly 4000 people, and another 8000 died in the weeks and months that followed (Greater London Authority, http://legacy.london.gov.uk/mayor/environment/air_quality/docs/ 50_years_on.pdf).
397
2. Data and methods 2.1. Data sources 2.1.1. Mortality data Mortality data are drawn from the Disease Surveillance Point System (DSPS) in the Chinese CDC. The DSPS, initiated in 1978, collected all death records for the surveillance locations each year. The data collected covered 71 districts/counties in 29 provinces from 1980 to 1989 and 145 districts/counties in 31 provinces from 1990 to 2000. The DSPS was overhauled following the SARS outbreak in 2003 and has since covered 161 districts/counties. In the event of a death, the doctor or decedent's family is required to fill out a death certificate and submit it to the DSPS. The mortality data includes basic demographic characteristics of the decedent and the cause of death. The causes of death are coded in the International Classification of Diseases 10 (ICD-10). Total mortality is classified by causes of death: cardiovascular diseases (I00-99), respiratory disease (J00-99), and all other diseases. For this study, we have daily number of deaths by age group, gender, and cause of death for all the DSPS districts/counties located in Beijing, Tianjin and Hebei Province. 2.1.2. Air pollution data PM2.5 concentrations are published by the MEP's ground monitoring stations and collected by the research team. In December 2012, PM2.5 was included in the Chinese National Ambient Air Quality Standard (CNAAQS, GB3095-2012). Since then, 24-h PM2.5 concentrations from all major Chinese cities have been published by the China National Environmental Monitoring Center (CNEMC) of MEP. We collected daily 24-h PM2.5 concentrations from all the monitoring stations located in Beijing, Tianjin and Hebei Province, then matched the monitoring sites with DSPS points based on the geodesic distance calculated from the longitude and latitude coordinates. Each DSPS point is assigned to its nearest monitoring site. If the distance between a DSPS point and its nearest air quality monitoring station was more than 30 km, we dropped it. 2.1.3. Weather data We collected daily weather information from January 1, 2013 to December 31, 2013, for all 7 city districts and counties in the study. These data were collected from The Weather Company's internet service known as the Weather Underground3 which is the world's largest network of weather stations (almost 10,000 stations in the United States and over 3000 across the rest of the world). As such it provides the most localized weather data currently available. We downloaded the following data for each location: daily precipitation, daily minimum, maximum and mean temperature, dew point, humidity, visibility miles and wind speed. 2.1.4. Study area We successfully matched mortality data, air pollution data and weather data for five city districts and two rural counties in Beijing, Tianjin and Hebei Province. Dongcheng District (city center) in Beijing (DC, BJ), Tongzhou District in Beijing (outer suburb) (TZ, BJ), Hongqiao District (inner city) in Tianjin (HQ, TJ), Haigang District in Qinhuangdao (HG, QHD), and Qiaodong District in Zhangjiakou (QD, ZJK) are urban the five city districts. Ji County in Tianjin (JC, TJ) and Ci County in Handan (CC, HD) are the two urban counties. The total populations in those districts/counties are 71,3602, 922,890, 647,673, 602,006, 287,249, 1,013,006, and 641,000 respectively for the year 2013. The male to female ratio ranges from 3
The data are available at www.weatherunderground.com.
398
M. Zhou et al. / Environmental Research 136 (2015) 396–404
estimate the degrees of freedom for each meteorological factor (Gu, 2013). The prediction results suggest that we use 7 degrees of freedom for dew point, 1 degree of freedom for temperature, relative humidity, and wind speed. To check the robustness of the results, we also conducted the analysis using up to 3 degrees of freedom for the smooth functions Si(Zit ) for each meteorological factor. We used the moving averages of current-day and previous-day concentrations of PM2.5 in the regressions, since the 2-day moving averages can pick up the health effects of both current day and previous day's exposure to air pollution. We also examined the air pollution effects with different lag structures (l = 0, 1, 2). We conducted the analyses separately for different age groups to explore the potential heterogeneous effects. We divided the sample into four age groups: o60, 60–69, 70–79, and Z80. Previous literature suggests that elder people are more vulnerable to fine particulate pollution, and we investigated if this is true in China during the smog episodes. We also investigated whether the smog episodes have heterogeneous effects on male and female mortality. Fig. 1. Geographic location of the study sites. The number associated with each district is the yearly averages of PM2.5 (mg/m3) in 2013.
0.95 to 1.05. Ji County in Tianjin and Ci County in Handan are rural counties, and the rest are urban districts. The geographical location of sampled districts/counties is shown in Fig. 1. Because the Chinese government did not start to monitor and publish PM2.5 concentrations until 2013, we only focused on a one-year window from January 1, 2013 through December 31, 2013. 2.2. Statistical methods We used a set of generalized additive models to estimate the associations between PM2.5 and daily mortality. Two basic models are adopted to examine the air pollution effects. The first model includes a set of dummy variables that indicates all the smog episodes. Any period during which PM2.5 concentrations exceed 100 mg/m3 for three or more consecutive days is defined as a smog episode. The models compare the mortality differences between the smog episodes and the regular days for each location. Specifically, the following model is estimated: p
log(E[yt ]) = α + βsmogt +
∑ Si(Zit , dfZi ) i=1
where yt is the number of deaths on day t ; smogt is a dummy indicator for the smog episodes. β represents the log-relative rate of mortality associated with a smoggy day. Zit are meteorological factors that are correlated with air pollution levels, and Si are their natural spline smooth functions. The second model uses specific PM2.5 concentrations in the regressions. For each location, we estimated:
3. Results Table 1 summarizes the descriptive statistics of daily number of deaths, PM2.5 concentrations, and weather conditions across all the locations. Overall, there were 11.26 deaths per day, including 6.50 from cardiovascular diseases and respiratory diseases. Of the daily deaths in our sample approximately 2.22 people were less than 60 years, 4.92 were between 60 and 79, and 4.13 were older than 80. The daily mean of PM2.5 concentration across all locations was 85.81 μg/m3, with a standard deviation of 74.47. The daily mean temperature, relative humidity, dew point and wind speed across all districts/counties were 53.98 °F, 51.74%, 36.26 °F and 5.37 mph, respectively. The air quality was particularly bad in January 2013. For example, in Beijing's Dongcheng District, the daily mean PM2.5 concentration was 211.13 μg/m3 in January 2013. The trends of PM2.5 concentration and daily mortality across all locations are shown in Fig. 2. The PM2.5 and mortality trends for each location is shown in Appendix A. PM2.5 concentration is positively associated with mortality. Both PM2.5 concentration and daily mortality are the highest during winter, and lowest during the summer. The fluctuations of PM2.5 concentration are dramatic, with the lowest value less than 10 μg/m3 and the highest value more than 350 μg/m3. The thick and short dashes indicate all the smog episodes throughout the year. Smog episodes are also positively associated with higher levels of mortality. Table 2 reports the results from regressing daily mortality on air pollution levels using the generalized additive models. The variable of interest is the smog episode indicator. Panel A presents the results on overall mortality. Without controlling for any Table 1 Summary statistics.
p
log(E[yt ]) = α + βxt − l +
∑ Si(Zit , dfZi ) i=1
where xt − l is the PM2.5 concentration on day t − l , and l is a day lag. We include daily average temperature, relative humidity, dew point, and wind speed in the regressions. We chose the degrees of freedom for each meteorological factor based on its best prediction for air pollution levels. Using degrees of freedom that predict best for air pollution levels is advantageous because they will produce unbiased or asymptotically unbiased estimates of the pollution log-relative risk (Dominici et al., 2004). We used the generalized cross-validation method to
Variable
Mean Std. Dev. Min
Daily death counts Daily cardiovascular and respiratory death counts Daily death counts (age o 60) Daily death counts (age 60–79) Daily death counts (age Z 80) PM2.5 concentrations (μg/m3) Temperature (°F) Humidity (%) Dew point (°F) Wind speed (mph)
11.26 6.50
6.42 4.53
2.22 1.89 4.92 3.14 4.13 3.27 85.81 74.47 53.40 20.92 51.53 20.81 35.23 24.48 5.75 2.96
0 0
Max 41 33
0 14 0 22 0 19 3 607 4 92 5 97 25 80 0 20
M. Zhou et al. / Environmental Research 136 (2015) 396–404
399
Fig. 2. The trends of daily PM2.5 and mortality across all locations. (Note: If PM2.5 concentrations exceed 100 mg/m3 for more than 3 consecutive days (including 3 days), we define it as a smog episode.) Table 2 Percent change of daily mortality associated with a smog episode. Panel A. Percent change in mortality associated with a smog episode Smog 95% CI Smog 95% CI DC, BJ TZ, BJ HQ, TJ HG, QHD QD, ZJK JC, TJ CC, HD Weather controls
9.82n 9.11n 6.71nn 2.66 12.07 23.42n 27.05n No
3.59 1.60 0.88 6.60 2.38 15.20 14.15
16.43 17.18 14.89 12.83 28.65 32.23 41.40
0.22 0.28 2.72 1.08 8.39 6.94nn 19.26n Yes
6.57 7.90 5.64 8.94 6.78 0.20 6.66
6.56 7.97 11.82 12.20 26.04 14.58 33.34
Panel B. Percent change in CVR and non-CVR mortality associated with a smog episode CVR DC, BJ TZ, BJ HQ, TJ HG, QHD QD, ZJK JC, TJ CC, HD Weather controls
1.25 6.11 0.76 3.63 16.05 11.66n 22.23n Yes
Non-CVR 7.78 4.35 9.27 9.98 4.95 3.12 8.11
11.17 17.71 11.90 19.31 41.68 20.90 38.20
1.88 7.83 5.99 1.58 1.02 2.35 13.46 Yes
11.42 17.92 6.76 15.99 19.58 13.79 3.50
8.68 3.50 20.49 15.31 26.89 10.60 33.40
The 95% CI are reported in the parenthesis. n
5% statistically significant. 10% statistically significant.
nn
weather conditions, the association between the smog episode indicator and daily mortality is positive in all 7 locations. For Dongcheng District of Beijing, Tongzhou District of Beijing, Ji County in Tianjin, and Ci County in Handan, the associations between smog and mortality are statistically significant at the 5% level. For Hongqiao District in Tianjin, the association is statistically significant at the 10% level. These results are consistent with the patterns in Fig. 2. However, when we include smooth functions from the meteorological factors, the coefficients of the smog indicator become insignificant for all five urban locations. The significant associations remain for the two rural counties. A smog episode is associated with a 6.94% (95% Confidence Interval, 0.20 to 14.98) increase in daily mortality for Ji County in Tianjin, and the association is statistically significant at the 10% level. A smog episode is associated with a 19.26 (95% CI, 6.66–33.34)% increase in daily mortality for Ci County in Handan, and the results are significant at the 5% level. Panel B of Table 2 presents the results on cardiovascular-respiratory (CVR) and non-CVR mortality separately. For CVR mortality,
the results are similar to the overall mortality in Panel A. Only in the two rural counties can we find that the smog episodes are statistically associated with daily CVR mortality. A smog episode is associated with an 11.66% (95% CI, 3.12–20.90) and a 22.23% (95% CI, 8.11– 38.20) increase in Ji County and Ci County, respectively. However, the smog episodes are not associated with deaths caused by non-CVR diseases. Thus the excessive deaths during the smog episodes in both rural counties are mainly from CVR diseases. No statistically significant associations are found for any urban districts. In Table 3, we focused on the two rural counties, Ji County and Ci County, where the associations between smog episodes and mortality are robust and statistically significant. We investigate the heterogeneous effects of smog episodes across different age groups or genders by separately estimating the models for each age group and gender. In Panel A, the results show that the smog episodes are statistically significant and associated with higher levels of mortality for the youngest age group (o 60) in both counties. In Ji County, smog episodes are associated with a 13.18% Table 3 Heterogeneous effects of smog on mortality. JC, TJ Smog
CC, HD 95% CI
Smog
Panel A. Heterogeneous effects by age Overall mortality Age o60 13.18n 2.13 30.88 0.47 Age 60–79 1.50 7.81 11.74 Age Z80 1.12 25.42 CVR mortality Age o60 Age 60–79 Age Z80 0.87
23.10n 9.92 9.54
0.39 1.88 2.54
52.12 23.14 23.12
17.11nn
95% CI
2.52
20.89n
0.86
26.64n 17.11nn
1.72 2.52
20.36nn 44.18 33.77 12.62nn 47.42 63.18 33.77 24.70nn 54.14
Panel B. Heterogeneous effects by gender Male 6.06 Female CVR mortality Male 9.07 Female
7.57
1.58
17.57
8.48n
1.24
19.15
11.62n
0.15
24.78
11.36n
0.02
24.04
The 95% CI are reported in the parenthesis. n
10% statistically significant. 5% statistically significant.
nn
14.77n
15.37n
0.34
21.38nn 38.92 32.17
1.82
27.21nn 48.35 35.57
400
M. Zhou et al. / Environmental Research 136 (2015) 396–404
4. Discussion
Table 4 Percent change in mortality associated with a 10 mg/m3 change in PM2.5.
DC, BJ TZ, BJ HQ, TJ HG, QHD QD, ZJK JC, TJ CC, HD Weather controls
PM2.5
95% CI
0.40n 0.77nn 0.59nn 0.64 1.70 1.31nn 1.18nn No
0.00 0.31 0.07 0.35 0.72 0.74 0.72
0.79 1.23 1.12 1.64 4.13 1.88 1.64
PM2.5
95% CI
0.01 0.45 0.58* 0.57 1.91 0.88** 0.55* Yes
0.47 0.11 0.06 0.51 1.19 0.30 0.02
0.48 1.01 1.22 1.65 5.02 1.46 1.13
The 95% CI are reported in the parenthesis. n
10% statistically significant. 5% statistically significant.
nn
increase in mortality for people less than 60 years of age. The estimate is even higher in Ci County, a 20.36% increase. For age group 60–79, the estimates are typically smaller. In fact in Ji County the smog episodes are not associated with the mortality for age group 60–79 at all. For the oldest age group (Age Z80), smog episodes are positively associated with mortality and the results are statistically significant. For CVR mortality, the qualitative results are similar to the overall mortality. The smog episodes have the largest impact on the youngest age group (o 60) and the oldest age group (Age Z80). In Panel B, we study the association between smog and mortality separately for males and females. In Ji County, the results from males are quite similar to those for females. In Ci County, the smog episodes have a greater impact on male mortality. A smog episode is associated with a 21.38% increase in overall male mortality, and a 14.77% increase in female mortality. For both counties and genders, air pollution has greater impacts on CVR mortality than overall mortality. To complement the analysis from the smog model, we provided the results from the PM2.5 in Table 4, in which we regressed daily death counts on 2-day moving averages of current-day and previous-day concentrations of PM2.5. The associations between PM2.5 and daily mortality vary by locations, but the qualitative results are similar to the smog model in Panel A of Table 2. We observe positive and statistically significant associations of PM2.5 with overall mortality in 5 out of 7 locations: Dongcheng District of Beijing, Tongzhou District of Beijing, Hongqiao District in Tianjin, Ji County in Tianjin, and Ci County in Handan. However, the results are sensitive to the inclusion of meteorological factors especially for the urban locations. After we adjust for the meteorological confounders, the coefficients of PM2.5 are still statistically significant only for the two rural counties: Ji County and Ci County. A 10 μg/m3 increase in 2-day average of PM2.5 concentration is associated with a 0.88% (95% CI, 0.30–1.46) and a 0.55% (95% CI, 0.02 to 1.13) increase in daily mortality for Ji County and Ci County, respectively. We also find weak evidence that PM2.5 is associated with daily mortality in Hongqiao District in Tianjin, which is statistically significant at the 10% level. Table 5 Social-economic conditions (2006–2010 average). District
Share of Ag. (%)
DC&TZ, BJ 0.90 HG, QHD 1.41 HQ, TJ 1.44 QJ, ZZK 2.00 CC, TS 9.10 JC, TJ 13.49
GDP PC (Yuan)
Population density (Person/km2)
# Of hospitals
# Of doctors
Region
62,390 46,187 61,197 27,688 21,552 17,645
944 2231 1067 2347 630 519
599 54 286 46 24 33
52,463 3095 23,497 2609 1055 4046
Urban Urban Urban Urban Rural Rural
We conducted an exploratory study on health effects of severe air pollution during 2013 in northern China in selected city districts and rural counties of Beijing, Tianjin, and Hebei Province. We collected comprehensive data on mortality, air pollution and weather conditions in both urban and rural areas in China. To the best of our knowledge, this is the first study to include an examination of the relationship between air pollution and the health of rural residents in China. In the matching process, we used more accurate readings of PM2.5 concentration for our study locations, while most previous studies simply average the data from various monitor stations or use a single reading in the U.S. embassy and four consulates as a proxy for population exposure level to particulate matter. As is evident in our data, the concentration of PM2.5 can differ dramatically across monitor sites on a daily basis. The approximating or averaging process will potentially introduces substantial measurement error bias and the direction of the bias is uncertain (Sarnat et al., 2005). Our measure of population exposure is more accurate because we match the monitor station with the DSPS locations within 30 km, and the estimated results suffer less from the potential measurement error bias. Our statistical analysis suggests that, without taking into account the weather conditions, the smog episodes are typically positively and statistically significantly associated with higher mortality. However, the associations in urban areas are sensitive to the meteorological factors. After we adjust for a flexible spline of temperature, humidity, dew point and wind, the statistical significance disappears for all urban areas. We also investigate the association between mortality and 2-day moving average of PM2.5 concentrations. The results are consistent with the smog model. Overall the results in urban areas suggest that the health effects of the smog might not be as bad as people speculated. In sharp contrast to the urban areas, we find that the smog episodes are consistently and statistically significantly associated with higher levels of mortality in the two rural counties, Ji County of Tianjin and Ci County of Handan. The results are robust to the inclusion of the meteorological factors in the model and across statistical specifications. On average, a smog episode is associated with 6.94% higher overall mortality in Ji County, and 19.26% higher mortality in Ci County. The different results between rural and urban areas can be explained by several factors. First, urban residents are typically more aware of the potential air pollution effects on health. During the smoggy days, vulnerable people (elderly and sick) were warned to stay indoors to reduce outdoor air pollution exposure. In contrast, rural residents are less aware of the potential harmful effects of air pollution. Second, the protective measures, such as wearing masks and installing air filters at home and workplace, have become very common in urban areas. Compared with urban residents, rural residents are less likely to take protective measures simply because they cannot afford them. Finally, rural residents lack immediate access to emergent medical assistance. When the smog events trigger heart attacks or acute respiratory diseases, rural residents are more likely to die. As shown in Table 5, the share of agricultural industry in GDP is the highest and the GDP per capita is the lowest in the two rural counties, Ji County and Ci County. Compared with the urban districts, the population density in the rural counties is smaller, and they have fewer hospitals and doctors. The heterogeneous effects of air pollution in the two rural counties reveal another striking result: the smog episodes have a greater impact on people who are less than 60 years of age in rural areas. The results seem to contradict the conventional wisdom that the elderly are more likely to be affected by air pollution than
M. Zhou et al. / Environmental Research 136 (2015) 396–404
Fig. A1. The trends of PM2.5 and Daily Mortality in Each Location.
401
402
M. Zhou et al. / Environmental Research 136 (2015) 396–404
Fig. A1. (continued)
the young. This controversial results may be explained by the fact that a young adult's exposure to air pollution can be very high in rural areas. In fact, the young adult population is the primary labor
force in rural areas. The majority of rural residents have to do outdoor farming work during the day. If the outdoor air quality is worse than the indoor air quality in those areas, it is possible that
Table B1 The association between smog episode and mortality using different degrees of freedom. Smog DC, BJ TZ, BJ HQ, TJ JC, TJ HG, QHD CC, HD QD, ZJK Temperature Humidity Dew Point Wind Speed
95% CI
0.02 6.12 0.45 6.97 3.87 4.38 8.13** 1.12 0.18 9.32 16.32** 4.22 8.22 6.41 Linear function Linear function Linear function Linear function
Smog 6.56 8.46 12.84 15.62 10.68 29.84 25.15
0.18 0.05 2.17 7.07* 1.09 17.30** 9.41 Natural Natural Natural Natural
95% CI
spline spline spline spline
6.36 7.40 6.17 0.02 8.94 4.86 5.90 with 2 df with 2 df with 2 df with 2 df
Smog 6.41 8.09 11.25 14.66 12.21 31.22 27.22
0.57 0.74 2.53 6.86* 1.46 17.89** 8.76 Natural Natural Natural Natural
95% CI
spline spline spline spline
6.79 6.86 5.92 0.27 8.66 5.47 6.51 with 3 df with 3 df with 3 df with 3 df
6.06 8.95 11.74 14.50 12.70 31.78 26.53
M. Zhou et al. / Environmental Research 136 (2015) 396–404
403
Table C1 The association between mortality and current day, 1-day lag and 2-day lag PM2.5 concentration. Panel A: without control Current
95% CI
DC, BJ TZ, BJ HQ, TJ JC, TJ HG, QHD CC, HD QD, ZJK Weather
0.08 0.19 0.21 0.40 0.26 0.71 0.85 No
0.42** 0.60** 0.66** 0.90** 0.53 1.12** 1.16 No
0.76 1.00 1.11 1.41 1.32 1.54 3.17 No
1-Day lag
95% CI
0.15 0.58** 0.21 1.10** 0.54 1.00** 1.29
0.20 0.17 0.26 0.60 0.25 0.57 0.71
1-Day lag
95% CI
0.01 0.26 0.13 0.77** 0.43 0.35 0.89
0.39 0.18 0.39 0.31 0.38 0.16 1.50
0.51 0.98 0.68 1.60 1.33 1.44 3.30
2-Day lag
95% CI
0.03 0.27 0.10 1.08** 0.05 0.67** 1.19
0.33 0.14 0.57 0.58 0.87 0.20 0.76
2-Day lag
95% CI
0.18 0.13 0.27 0.60** 0.41 0.19 0.64
0.53 0.54 0.77 0.17 1.25 0.72 1.43
0.38 0.68 0.38 1.58 0.77 1.14 3.15
Panel B: with control
DC, BJ TZ, BJ HQ, TJ JC, TJ HG, QHD CC, HD QD, ZJK Weather
Current
95% CI
0.03 0.42 0.72** 0.54** 0.52 0.60** 1.03 Yes
0.42 0.11 0.18 0.03 0.37 0.08 1.37 Yes
0.47 0.95 1.26 1.04 1.41 1.12 3.44 Yes
the smog episodes kill more outdoor laborers than elderly people who spend most or all of their time indoors. Death of a young person in a rural family could be disastrous if the person is the major income earner. Therefore, alerting the rural residents about the danger of smog and educating them how to take preventive measures might have huge positive social impacts in rural areas. Overall, findings in this study suggest that the smog episodes and fine particulates air pollution have greater and more detrimental impacts on rural residents, especially those living close to big polluted cities. Compared with urban residents who are constantly alerted by the media about the severe air pollution, the rural residents are much less aware of and prepared for air pollution and don't usually take avoidance measures. As a consequence, the estimates in rural areas are closer to the true effects of air pollution. Previous studies only focus on urban areas where both air pollution data and mortality data are available (see Shang et al., 2013, for a recent review). Given the fact that more than half of Chinese people live in rural areas, understanding the health risks air pollution imposed on rural people is long overdue. We hope this preliminary study can motivate more academic research in this underexplored area. We conducted a variety of sensitivity checks and the results are reported as online Appendices. For the smog model, we checked whether the results are robust to the inclusion of control variables and the choices of different degrees of freedom in the spline function. We started by analyzing a linear function of each meteorological factor, and then extended the analysis by changing the choice degrees of freedom. The regression results, as summarized in Appendix B, are quite robust to the choices of different degrees of freedom in the spline functions, as long as some minimal representation is included. For the PM2.5 model, we experimented with different lag structures of PM2.5 for each location. The results are reported in Appendix C. In most cases mortality is either statistically significantly associated with current day PM2.5 or with one-day lagged PM2.5. So using a 2-day moving average picks up the average effect of both. The estimates using single day PM2.5 are either slightly smaller or slightly larger than those using the twoday moving averages, depending on the underlying dynamics of air pollution effects. For districts/counties showing no statistically significant associations between two-day moving average PM2.5 concentrations and mortality, a different lag structure made no difference.
0.36 0.70 0.66 1.24 1.24 0.85 3.28
0.17 0.28 0.22 1.04 0.43 0.34 2.71
5. Conclusion We find a significant association between the smog episodes (or PM2.5 concentration) and mortality in rural areas of China. The associations are robust to inclusion of meteorological factors and model specifications. Young rural residents ( o60) are also more likely to die during the smog period than young urban residents. The smog episodes and fine particulates air pollution had bigger and more detrimental impacts on rural residents, especially those living close to big polluted cities.
Acknowledgments This study is partially funded by a seed grant from the China Medical Board, United States (Grant No. 13-137) to Peking Union School of Public Health, and the National Basic Research Program of China (“973 Program”) (No. 2012CB955500).
Appendix A See appendix Fig. A1.
Appendix B See appendix Table B1.
Appendix C See appendix Table C1.
References Aunan, K., Pan, X.-C., 2004. Exposure-response functions for health effects of ambient air pollution applicable for China – a meta-analysis. Sci. Total Environ. 329, 3–16. Chan, C.K., Yao, X., 2008. Air pollution in mega cities in China. Atmos. Environ. 42, 1–42.
404
M. Zhou et al. / Environmental Research 136 (2015) 396–404
Chen, R., et al., 2012a. Short-term exposure to sulfur dioxide and daily mortality in 17 Chinese cities: the China air pollution and health effects study (CAPES). Environ. Res. 118, 101–106. Chen, R., et al., 2012b. Association of particulate air pollution with daily mortality: the China air pollution and health effects study. Am. J. Epidemiol. 175, 1173–1181. Dominici, F., et al., 2004. Improved semiparametric time series models of air pollution and mortality. J. Am. Stat. Assoc. 99, 938–948. Gu, C., 2013. Smoothing Spline ANOVA Models, vol. 297, Springer. Guo, Y., et al., 2009. The association between fine particulate air pollution and hospital emergency room visits for cardiovascular diseases in Beijing, China. Sci. Total Environ. 407, 4826–4830. Guo, Y., et al., 2013. The burden of air pollution on years of life lost in Beijing, China, 2004–08: retrospective regression analysis of daily deaths. Br. Med. J. 347, f7139. Guo, Y., et al., 2010. The relationship between particulate air pollution and emergency hospital visits for hypertension in Beijing, China. Sci. Total Environ. 408, 4446–4450. He, K., et al., 2002. Urban air pollution in china: current status, characteristics, and progress. Ann. Rev. Energy Environ. 27, 397–431. Kan, H., et al., 2008. Season, sex, age, and education as modifiers of the effects of outdoor air pollution on daily mortality in Shanghai, China: The Public Health and Air Pollution in Asia (PAPA) Study. Environ. Health Perspect. 116, 1183–1188. Nemery, B., et al., 2001. The Meuse Valley fog of 1930: an air pollution disaster. Lancet 357, 704–708.
Sarnat, J.A., et al., 2005. Ambient gas concentrations and personal particulate matter exposures: implications for studying the health effects of particles. Epidemiology 16, 385–395 (10.1097/01.ede.0000155505.04775.33). Schreniz, H., et al., 1949. Air Pollution in Donora, PA., Epidemiology of the Unusual Smog Episode of October 1948; Preliminary Report. Public Health Service Bulletin, PHS. Shang, Y., et al., 2013. Systematic review of Chinese studies of short-term exposure to air pollution and daily mortality. Environ. Int. 54, 100–111. Tao, Y., et al., 2012. Estimated acute effects of ambient ozone and nitrogen dioxide on mortality in the Pearl River Delta of southern China. Environ. Health Perspect. 120, 393–398. U.K. Ministry of Health, 1954. Mortality and Morbidity during the London Fog of December 1952. Reports on Public Health and Medical Subjects no. 95. Ministry of Health, London. Venners, S.A., et al., 2003. Particulate matter, sulfur dioxide, and daily mortality in Chongqing, China. Environ. Health Perspect. 111, 562–567. Wong, C.-M., et al., 2008. Public Health and Air Pollution in Asia (PAPA): a multicity study of short-term effects of air pollution on mortality. Environ. Health Perspect. 116, 1195–1202. Wong, T.W., et al., 2002. Associations between daily mortalities from respiratory and cardiovascular diseases and air pollution in Hong Kong, China. Occup. Environ. Med. 59, 30–35.
