Environmental Research 140 (2015) 684–690
Contents lists available at ScienceDirect
Environmental Research journal homepage: www.elsevier.com/locate/envres
Effects of ambient air particles on mortality in Seoul: Have the effects changed over time? Honghyok Kim a, Hyomi Kim a, Jong-Tae Lee a,b,n a b
Department of Public Health Sciences, Graduate School, Korea University, 136-703 Seoul, Republic of Korea Department of Environmental Health, College of Health Science, Korea University, 136-703 Seoul, Republic of Korea
art ic l e i nf o
a b s t r a c t
Article history: Received 11 December 2014 Received in revised form 27 May 2015 Accepted 31 May 2015 Available online 14 June 2015
Background: Several studies have shown that there may be temporal variation in PM short-term effect on mortality. This temporal pattern may play an important role in evaluating air quality policies. Objectives: We investigated temporal variation in the association between PM and mortality in Seoul, Korea, 1998–2011. Methods: We adopted a generalized additive model and a series of time windows of five years to analyze temporal variation in associations between PM and all-cause, cardiovascular, and respiratory mortality. This time-window approach offers not only a comparison between one and the other half period but also successive variation. Time-varying associations were estimated only for days without Asian dust (dust storm blown from the Gobi desert) intrusion. Results: Annual average PM10 and PM2.5 total mass decreased from 70.0 to 46.9 mg/m3 and 44.4 to 23.4 mg/m3, respectively, during 2001–2011. A 10 mg/m3 increase in PM10 was associated with 0.16% (95% CI¼ 0.03% to 0.35%) additional all-cause deaths in 2002–2006 and it increased to 0.26% (95% CI ¼0.05– 0.48%) in 2007–2011. For PM2.5, the association increased from 0.35% (95% CI ¼ 0.02% to 0.71%) to 0.48% (95% CI ¼0.08–0.88%). For cardiovascular and respiratory mortality, increasing trends with stronger estimates were found. Conclusions: The present study showed temporally increasing trends in associations between PM and mortality. Current policies may not be as effective to reducing health risks attributable to PM as expected. Air quality interventions should be encouraged in terms of causal factors for time-varying association between PM and mortality. & 2015 Elsevier Inc. All rights reserved.
Keywords: Particulate matter Fine particle Mortality Temporal variation Time-varying effect
1. Introduction Epidemiological research has demonstrated that acute exposure to particulate matter with an aerodynamic diameter o10 mm (“PM10”) is associated with diverse health response (Pope III and Dockery, 2006). Incorporating advanced measurement technologies, researchers have since then concentrated on smaller particles, with aerodynamic diameters o2.5 mm (“PM2.5”) and provided consistent evidence of a stronger association with health outcomes than that of PM10 or PM10–2.5 (Dong et al., 2013; Schwartz and Neas, 2000). In Seoul, the capital of South Korea, many studies have reported an association between short-term exposure to PM and an increase in risk of health outcomes (Heo et al., 2014; Lee et al., 2007; Son et al., 2012). n Corresponding author at: Department of Environmental Health, College of Health Science, Korea University, Seoul, Republic of Korea. Fax: þ 82 2 2298 0248. E-mail address: [email protected] (J.-T. Lee).
http://dx.doi.org/10.1016/j.envres.2015.05.029 0013-9351/& 2015 Elsevier Inc. All rights reserved.
“Asian dust,” also known as yellow dust, which originates in the Gobi Desert spread from Southern Mongolia to Northern China, is sporadically blown to Korean peninsula by westerly winds and aggravates air quality in the region. It leads to increase particulate concentrations, but especially tends to increase coarse particles with crustal components (Kim et al., 2003). In fact, such particles may be less deleterious to human health than fine particles (Pope III and Dockery, 2006). Thus, an association of particulates with mortality regardless of excluding days with Asian dust intrusion in a statistical model would possibly be underestimated in terms of PM effects for dominant non-Asian dust days (Lee et al., 2007). Accordingly, it is necessary to take this concept into account when analyzing PM effects on mortality in Korea. According to recent studies, there may be temporal variation in short-term effects of air pollution; the relative risk is not constant over time. Studies in the U.S. and Europe found decreasing patterns in PM effects over time (Breitner et al., 2009; Dominici et al., 2007). They suggested that air quality control policies may have
H. Kim et al. / Environmental Research 140 (2015) 684–690
had an impact on not only reducing PM mass concentration but also on desirable changes in the constituents of PM. If there are similar trends in Korea, a time-varying effect of PM may also be expected. Thus, we sought to analyze an association between PM and mortality in Seoul, considering the intrusion of Asian dust. Second, we investigated temporal variation in the association, and suggested possible reasons for a time-varying effect of PM on mortality. Finally, we made suggestions on future studies and ambient air quality policies.
2. Methods and materials 2.1. The scope and data The study location was Seoul, Korea, with a population of about 10 million (Statistics Korea, 2010). Data for mortality, PM10, PM2.5, and meteorological factors were obtained from Statistics Korea, the National Institute of Environmental Research, Seoul Metropolitan Government Research Institute of Public Health and Environment, and the Korea Metrological Administration, respectively. PM concentration in Seoul has been measured in national monitoring stations since 1995 for PM10 and 2001 for PM2.5. PM values were computed as daily (24 h) mean concentrations from hourly measurements. Valid daily mean concentration was calculated with data from at least 75% of stations (18 þ out of 25) and of hourly measurements (16 þ out of 24). Otherwise, we imputed an annual average from valid daily estimates. Missing values were also imputed with annual averages. Exceptions to spatial criteria were the year of 1998 for PM10 because of only 17 monitoring stations and the period of 2001–2004 for PM2.5 due to almost 10 of unevenly operating stations. The number of invalid daily PM10 was 270 (5.3%) from 1998 to 2011 and that of invalid daily PM2.5 was 563 (14.0%) from 2001 to 2011. Temperature and relative humidity were measured at a single fixed station in the Jongno district, a center of Seoul city. Daily mean temperature and relative humidity were calculated and there were no missing values for them. We classified cause-specific mortality into all-cause except for accidental causes (International Classification of Disease, Tenth Revision [ICD-10]: A00-R99 and Ninth Revision [ICD-9]: 1-799), cardiovascular (ICD-10: I00-I99, ICD-9: 390-459), and respiratory (ICD-10: J00-J99, ICD-9: 460-519) mortality. 2.2. Statistical analysis A generalized additive model (GAM) was used to analyze associations between PM10 and PM2.5 and mortality. Poisson distribution with over-dispersion was applied. To control potential confounders, we adjusted daily mean temperature, daily mean relative humidity, time-trend, day of the week, and holidays. Temperature, relative humidity, and time-trends were adjusted as a cubic regression spline with 5 knots and 7 knots/year. Knots were equally dispersed throughout values. In preliminary analysis, we did not detect partial autocorrelation which exceeded |0.1| in the main model with aforementioned knots. Day of the week and holidays were included in the model as dummy variables. We compared estimates of days without Asian dust intrusion (NAD, non-Asian dust days) to those from all days including days with Asian dust intrusion (NAD þAD, non-Asian dust Days plus Asian dust days). Then we conducted subsequent analyses for NAD. To explore temporal variation in associations between PM and cause-specific mortality, we used a series of time windows, five years long. This approach has been used in our previous work (Kim
685
et al., 2015). It enables to compare one and the other half period as many studies on temporal variation in relative risk did. It also allows to see consecutive variation with sufficient statistical power and avoids inter-annual harvesting effect (Kim et al., 2015). For statistical analyses, the SAS 9.4 and R 3.0.2 softwares were used. 2.3. Sensitivity analysis We changed the number of knots in spline function for temperature (4, 5 and 6), relative humidity (4, 5 and 6) and time (4, 5, 6 and 8) to see whether estimated association of PM with mortality was stable. As the size of population changed over the study period, annual variation of population was also adjusted as an offset in a gam model across time-windows, even though we expected that spline function for time-trend may cover the population change sufficiently and change in the size of population in a single city is relatively minimal. The annual number of population was obtained from the statistics of registered population in Korean Statistical Information Service. We also conducted sensitivity analyses using 1-year (annual variation) and 3-years of windows. Bayesian varying coefficient regression (VCR) was used to explore time-varying association between PM and mortality: βPM = f (YEAR )PM with ‘BayesX’ package in R software (Kneib et al., 2011). Similar approach was adapted elsewhere (Breitner et al., 2009). In addition, multi-pollutant model needs to be taken into consideration if temporal variation in PM effect is (partially) attributed to other pollutants. We added SO2, NO2, or O3 into the main model (so that two-pollutant model) to see if it shows different temporal patterns of PM effects.
3. Results 3.1. Descriptive statistics Regardless of Asian dust intrusion, the average daily PM10 concentration for the study period, 1998–2011, was 60.7 mg/m3. For PM2.5, it was 32.1 mg/m3 in 2001–2011. There were a total of 159 Asian dust days. Average concentrations of PM10 and PM2.5 were higher during Asian dust intrusion in the corresponding periods (Table 1). Annual ambient PM concentrations have declined consistently (Fig. 1) The average daily all-cause mortality was 93.3. Death count in those aged 65 years or older (65 þ ) represented two-thirds of all ages (Table 1). For cause-specific deaths, the averages were 24.4 and 5.7 in cardiovascular and respiratory-related death, respectively. Daily death count increased slightly over time as the elderly population grew (Supplemental materials, Fig. S1). 3.2. Quantification of PM effect on mortality Table 2 shows associations between PM10, PM2.5, and mortality. An increment of 10 mg/m3 in PM10 was associated with a 0.12% (95% CI ¼0.12–0.21%) increase in all-cause mortality in 1998–2011. For PM2.5, it was associated with a 0.20 (95% CI ¼ 0.01% to 0.41%) increase in risk of all-cause mortality in 2001–2011. When we excluded days with Asian dust intruded, both associations of PM10 and PM2.5 were strengthened to 0.22% (95% CI ¼ 0.10% to 0.35%) and 0.23% (95% CI ¼ 0.01% to 0.47%), respectively, for the corresponding periods. Looking into cause-specific mortality, stronger associations of PM were found compared to risk in all-cause mortality. As with all-cause mortality, estimates for NAD were higher. To confirm this effect modification, we checked statistical significance of an interaction between PM and the occurrence of
686
H. Kim et al. / Environmental Research 140 (2015) 684–690
Table 1 Descriptive statistics, Seoul, 1998–2011.
b
25th 50th
75th
100th
Periods
Exposure
Asian dust
% Increase (95% CI)
5.0 37.4 3.4 14.4 15.7
36.9 76.7 19.7 29.0 4.1
54.0 110.8 29.7 39.6 14.4
72.1 162.8 37.9 61.1 22.0
248.2 869.1 161.0 362.9 30.4
All-cause
1998–2011
PM10
2001–2011
PM10
NAD þ ADa NADb NAD þ ADa NADb NAD þ ADa NADb
0.12 0.22 0.10 0.19 0.20 0.23
(0.02, 0.21) (0.10, 0.35) (0, 0.20) (0.05, 0.33) ( 0.01, 0.41) ( 0.01, 0.47)
NAD þ ADa NADb NAD þ ADa NADb NAD þ ADa NADb
0.19 0.39 0.23 0.44 0.47 0.69
(0.02, 0.37) (0.16, 0.62) (0.03, 0.42) (0.18, 0.70) (0.07, 0.86) (0.24, 1.15)
NAD þ ADa NADb NAD þ ADa NADb NAD þ ADa NADb
0.33 0.75 0.31 0.77 0.92 1.34
( 0.03, 0.7) (0.27, 1.23) ( 0.09, 0.71) (0.22, 1.31) (0.10, 1.75) (0.39, 2.30)
Days Mean 0th
PM10 (mg/m3) (1998–2011) PM2.5 (mg/m3) (2001–2011) Daily mean temperature (°C) Daily mean humidity (%) All-cause deaths Z65 Years of age Cardiovascular deaths Z65 Years of age Respiratory deaths Z65 Years of age a
Table 2 Association between PM10 and PM2.5 with mortality, Seoul, 1998–2011.
NADa 4954 58.3 ADb 159 135.7 NADa 3887 31.4 b AD 130 52.0 5113 12.9
5113
62.0
19.4
51.2
62.5
72.5
96.3
5113 5113 5113
93.3 61.7 24.4
55 28 8
85 55 20
93 61 24
101 68 28
148 103 49
5113 5113 5113
18.0 5.7 4.8
5 0 0
15 4 3
18 5 5
21 7 6
37 21 18
PM2.5
CVD (I00-99)
1998–2011
PM10
2001–2011
PM10 PM2.5
RES (J00-99)
1998–2011
PM10
2001–2011
PM10
Days without Asian dust intrusion (NAD, non-Asian dust days). Days with Asian dust intrusion (AD, Asian dust days).
Asian dust in models. As a result, an interaction between PM10 and Asian dust intrusion was statistically significant (po 0.001). For PM2.5, it was significant in risk of cardiovascular mortality only (p o0.05). Therefore, subsequent analyses were conducted only for NAD. Using a series of time-windows, we sought to explore whether there was temporal variation in associations between PM and mortality. An association of PM10 with mortality increased from 0.16% (95% CI ¼ 0.03% to 0.35%) in 2002–2006 to 0.26% (95% CI ¼0.05–0.48%) in 2007–2011. For PM2.5, it increased from 0.35% (95% CI ¼ 0.02% to 0.71%) to 0.48% (95% CI ¼0.08–0.88%). The strongest associations of PM10 and PM2.5 and all-cause mortality were 0.37% (95% CI ¼0.15% to 0.58%) in 2005–2009, and 0.63% (95% CI ¼0.18–1.09%) in 2004–2008, respectively. Similar patterns were also seen in association with cardiovascular mortality while increasing trends were markedly for respiratory mortality. In most strata, associations in the most recent period were stronger than those in the oldest (Table 3). 3.3. Sensitivity analysis We repeated the main analysis changing the number of knots of cubic splines for temperature (4, and 6), relative humidity (4, and 6), and time (4, 6, and 8). These showed similar results (results not shown). The adjustment for annual population variation did not change temporal variation in associations of PM with mortality (results not shown). We also changed the length of the time
PM2.5
a
The study period including days with Asian dust intrusion (NAD þ AD, non-Asian dust days þ Asian
dust days). b Days without Asian dust intrusion (NAD, non-Asian dust days).
windows to one year (annual variation) and three years (Supplemental materials, Table S1-2). However, due to insufficient power, we found few significant associations across time-windows. As an alternative method, using Bayesian VCR, annual variations in associations of PM10 and PM2.5 and mortality were analyzed (Fig. 2). Even though this method was also not away from insufficient power, results were similar. In contrast to trends for all-cause and cardiovascular mortality, the association of PM with respiratory mortality increased last three (PM2.5) and four (PM10) years. The same analysis for all days including Asian dust days showed that the temporal patterns did not change (results not shown). In two-pollutant modeling, we found association between PM and mortality with wider confidence intervals. The two-pollutant models showed temporal patterns as the one-pollutant (PM) models did, except for the model with PM2.5 and SO2 for respiratory-related mortality; an estimate decreased from 1.79 (95% CI: 0.02 to 3.63) in 2002–2006 to 1.01 (95% CI: 0.98 to 3.05) in 2007–2011 (Supplemental materials Table S3-5). For annual variation from Bayesian VCR, trends in the two-pollutant models with wider posterior intervals were similar to those in the one-
Fig. 1. Annual trends in ambient concentrations of PM10 and PM2.5, Seoul, 1998–2011.
H. Kim et al. / Environmental Research 140 (2015) 684–690
687
Table 3 Temporal variation in associations between PM10 and PM2.5 and mortality for days without Asian dust intrusion using a series of 5-year time windows, Seoul, 1998–2011. % Increase (95% CI) All-cause
1998–2002 1999–2003 2000–2004 2001–2005 2002–2006 2003–2007 2004–2008 2005–2009 2006–2010 2007–2011
Cardiovascular
Respiratory
PM10
PM2.5
PM10
PM2.5
PM10
0.29 0.17 0.06 0.08 0.16 0.18 0.31 0.37 0.29 0.26
– – – 0.10 ( 0.22, 0.41) 0.35 ( 0.02, 0.71) 0.29 ( 0.08, 0.67) 0.63 (0.18, 1.09) 0.56 (0.10, 1.02) 0.43 (0.01, 0.85) 0.48 (0.08, 0.88)
0.41 0.24 0.27 0.35 0.43 0.39 0.68 0.75 0.64 0.60
– – – 0.58 (0.01, 1.15) 0.81 (0.16, 1.47) 0.53 ( 0.14, 1.22) 1.26 (0.43, 2.1) 1.13 (0.27, 2.01) 1.10 (0.28, 1.92) 1.12 (0.35, 1.9)
0.46 0.56 0.25 0.22 0.44 0.91 1.05 1.22 1.43 1.51
(0.09, 0.50) ( 0.01, 0.35) ( 0.12, 0.24) ( 0.11, 0.27) ( 0.03, 0.35) ( 0.01, 0.36) (0.09, 0.52) (0.15, 0.58) (0.07, 0.50) (0.05, 0.48)
(0.04, 0.78) (-0.09, 0.57) (-0.07, 0.60) (0.01, 0.70) (0.08, 0.77) (0.05, 0.73) (0.30, 1.07) (0.34, 1.15) (0.23, 1.05) (0.19, 1.02)
pollutant models (results not shown). The exception was not found.
4. Discussion In this study, we found that association of PM with mortality was enlarged when excluding days with Asian dust intrusion. These results were consistent with the previous study suggesting that increased PM concentration attributable to Asian dust intrusion contributed relatively less toxic particles fraction, so that it may cause underestimated risks of PM (Lee et al., 2007). Also, it was pointed out that recognition of an Asian dust event may change individual behavior to minimize exposure to ambient air particles (Lee et al., 2007). We emphasize that those two are related with bias to underestimating risks by using less representative population averages of ambient PM concentration as an indicator to the average of individual exposure to adverse particles in time-series analysis. Therefore, the price of using population average of ambient PM level in time-series analysis should be taken into account for impact assessment of PM for dominant non-Asian dust days. On the other hand, indicators which represent relatively less toxic fraction (in this case, PM10) are useful to assess impact of PM during Asian dust intrusion. Our efforts to explore temporal variation in an association of PM with mortality originated in curiosity about achievement from receptor-oriented air pollution reduction policies in Korea. To prevent deleterious PM effects, the Korean Ministry of Environment (KMOE) promulgated the first regulation on ambient PM10 concentration in 1993, and initiated the PM10 quality standard in
( 0.32, 1.25) ( 0.16, 1.29) ( 0.48, 0.99) ( 0.54, 1.00) ( 0.35, 1.24) (0.11, 1.71) (0.23, 1.87) (0.41, 2.04) (0.65, 2.21) (0.75, 2.28)
PM2.5 – – – 0.88 ( 0.37, 2.14) 1.48 ( 0.03, 3.02) 2.01 (0.42, 3.63) 2.15 (0.34, 4.00) 2.18 (0.40, 3.99) 2.20 (0.64, 3.79) 2.33 (0.92, 3.76)
1995 under the Clean Air Conservation Act (KMOE, 2013). The standard concentrations were 80 mg/m3 for annual average and 150 mg/m3 for daily average. The limit of annual average was intensified to 70 mg/m3 in 2001. In 2007, this standard changed to 50 mg/m3 and the existing daily average standard was tightened upto 100 mg/m3. The new standard for outdoor PM2.5 was announced in 2012 and will come into effect after 2015. It follows the recommended levels of interim target-2 in World Health Organization air quality guidelines: yearly 25 mg/m3 and daily 50 mg/m3. Furthermore, as targeting the capital and its neighboring areas where nearly half the total population of South Korea lives, the ‘Special Act on Metropolitan Air Quality Improvement’ was established in 2003 in order to improve air quality by cutting emissions of PM, NOx, and SOx markedly (KMOE, 2014). This Act, which has been in effect since 2005, involves reductions in automobile emissions, encouraging eco-friendly energy, air pollutant total mass emission management system, and aR tradable permit program. These two endeavors have led to a consistent decrease in ambient PM concentrations in metro Seoul. In fact, such policies are expected to encourage improvements in air quality in terms of not only ambient PM mass concentration, but also its composition. However, due to inadequate awareness and inconsistent measurements of its constituents, only long-term trends in PM mass, but not composition (i.e., whether it is less harmful/toxic), have so far been evaluated in Korea. Therefore, from an epidemiological point of view, exploration on a timevarying (i.e., a temporally decreasing trend in) association of PM and mortality can be an alternative if PM constituents vary over time. Our results showed that a relative risk of acute exposure to
Fig. 2. Annual variations in association of PM with mortality using Bayesian varying coefficient regression (VCR), Seoul, 1998–2011; estimates and 95% posterior intervals.
688
H. Kim et al. / Environmental Research 140 (2015) 684–690
ambient PM in mortality increased over time-windows. We did not find similar results elsewhere. Rather, Breitner et al. (2009) suggested that an association between short-term exposure to air pollution and mortality in Erfurt, Germany decreased due to air pollution controls. A study using data of the National Morbidity Mortality Air pollution Study in the US showed that PM effects on mortality seemed to decrease over time (Dominici et al., 2007). This study also suggested that it was probably a result of national and regional air quality policies, such as reducing allowable emissions of diesel particulate matter, nitrogen, and hydrocarbons. The present study indicates a significantly decreasing trend in ambient PM concentration like other regions, probably because of strengthened air quality standards and implemented policies. However, these interventions seemed not to consistently improve air quality in view point of human health risks. Breitner et al. (2009) questioned that temporal variation in effects of PM may depend on a non-linear dose–response curve between PM and mortality. This study conducted a likelihood-ratio test for two models: one linear and the other non-linear using a regression spline. We also tested likelihood ratio between the two models, and found a linear relationship between PM and mortality as they did. This indicates that time-varying effect of PM may not depend on shape of dose–response function. Therefore, we suggest that two possible reasons for the observed temporal variation in the present study should be highlighted: (1) the increased toxicity of PM and (2) reduction in bias induced by markedly decreasing ambient PM mass concentration. Unlike other ambient risk factors, such as temperature, SO2, NO2, O3,etc., “PM” is not a single concept but a complex mixture of diverse physiochemical (even biological) compounds. That is why even the same mass of PM has different toxicity depending on its composition. ‘PM10’ is commonly divided into PM2.5 and PM10–2.5. Furthermore, PM2.5 is also a mixture of different sized particles, including ultrafine particles (UPs). UPs consist of not only single but also aggregated particles with aerodynamic diameters o100 nm (Ibald-Mulli et al., 2002). Indeed, it has been hypothesized that UPs are responsible for PM effects at ambient concentrations (Seaton et al., 1995). Several epidemiological studies reported that number concentration (NC) of UP was more strongly (Stölzel et al., 2006) or only associated (Penttinen et al., 2001) with adverse health responses. Recently, associations of health outcomes with NC of Aitken mode and nucleation mode particles were discussed by a systematic review (Atkinson et al., 2014). Thus, one possibility for the increasing trend in the present study is that fraction of UP-scale particles (or even smaller particles) may have increased over time. Although there is no reported data or study regarding its temporal variation in Korea, to our knowledge, several studies conducted in Germany support the idea. One study found that in constituents of PM2.5 from 1993 to 1999 in East Germany, the number concentration of UPs was steady while NC of relatively larger particles (PM0.1–2.5) decreased and, consequently, PM2.5 mass concentration decreased (Pitz et al., 2001). This study mentioned that improved engine technologies (i.e., lower-emission diesels) were successful in reducing PM2.5 mass concentration but may have paradoxically resulted in emission of nanoparticles o50 nm. Other study which measured such long-term variation of PM emphasized that an improvement in ambient air quality in terms of mass concentration may not reduce, or may even increase, health risks (Kreyling et al., 2003). Along with consideration of size of PM, toxicity of chemical constituents in PM2.5 has been discussed as well. Indeed, recent multi-city and meta-analysis studies have reported associations of specific components with health end points (Atkinson et al., 2014; Levy et al., 2012). Albeit of issues on heterogeneity across individual studies, organic carbon, elemental carbon, nitrate, sulfate, nickel, and lead were generally associated with health outcomes.
In Korea, two studies investigated associations of chemical species with mortality in Seoul using different measurements. One study reported that nitrate, sulfate, and ammonium, which contributed a large proportion of PM2.5 total mass, were marginally associated with cardiovascular mortality (Son et al., 2012). The other showed that organic carbon, elemental carbon, and lead were associated with mortality and suggested that deleterious effects of PM2.5 may be mainly ascribed to local combustion sources (Heo et al., 2014). Given that the distribution of composition of PM2.5 depends on emission sources controlled by air quality policies and interventions, it will be necessary to measure the constituents of PM2.5 over the long-term, and to analyze association of the smallest unit of noxious particles as possible with mortality. Only change in ambient PM mass itself was not related to temporal variations in an effect of ambient PM on mortality. However, the extent of representativeness of population average ambient concentration to average of individual exposure in a timeseries analysis may depend markedly on change in PM mass. A biased estimate in Poisson time-series regression is directly related to the representativeness. This issue was scrutinized elsewhere (Zeger et al., 2000). On the assumption that a relative risk is constant over time, either an infiltration factor, a fraction of particles penetrating into indoors and remaining suspended (Diapouli et al., 2013) or of outdoor exposure to ambient-originated pollution (Zeger et al., 2000), is assumed to be less movable or stable. However, in context of a time-varying effect of ambient PM on mortality, it may be expected to change so much as it cannot be ignored. An infiltration factor of fine particle cannot easily be calculated because of indoor sources of PM such as environmental tobacco smoke (Strand et al., 2005). Alternatively, some researchers have used sulfate, originating mostly in outdoor environment, as a tracer of PM2.5 (Sarnat et al., 2002). The result of the study conducted from Spengler et al. (1981) showed that the difference in sulfate concentration between indoor and outdoor was smaller in areas with a lower concentration of outdoor sulfate. Godish (1989) noted that for air pollutants which do not derive from indoor sources, indoor/outdoor ratio is near to unity, and for SO2, particularly, it approaches relatively to unity in environment with low SO2 concentration. Provided that particles with diameter less than 1 mm are more highly infiltrated into indoors than larger particles (Sarnat et al., 2006) and reduction in bigger particles may be a main contributor to lowering total mass concentration, proportion of infiltrated particles may have increased over time. In addition, dramatic reduction of air pollution concentration may also encourage people to spend more time outside. Thus, it can be expected that decreasing ambient mass of PM may accompany an increasing either infiltration factor or fraction of exposure to outdoor-originated particles, so that bias from time-series analysis may shrink. Consequently, attenuating underestimation on relative risk may result in an increasing pattern of ambient PM effect. Bias from representativeness of population average in timeseries analysis should be also discussed in temporal variation in heat effect on mortality. An increasing rate of air conditioner use, a major factor, can cause average of ambient temperature to be less representative. It can magnify the degree of underestimation. A change in degree of underestimation is cumbersome to estimate a true risk in a statistical analysis. On the other hand, in context of policy efficacy, it can be viewed as a result of interventions (Zeger et al., 2000). Thus, our results imply that intensified air quality regulation might not have had desirable impact on personal risk due to ambient-origin PM in Seoul. Epidemiologists have researched effect modification by temperature in association of PM with mortality (Dong et al., 2013; Stafoggia et al., 2008). Two major aspects are of interest; the
H. Kim et al. / Environmental Research 140 (2015) 684–690
possibility of increasing exposure to ambient PM by keeping windows open during the summer season (Stafoggia et al., 2008) and, of a more toxic PM composition (Peng et al., 2005). These two issues are involved in our suggested hypotheses. The first is related with representativeness of outdoor PM. Regarding the increased toxicity of PM, this may be related to the formation of secondary sulfate and nitrate in the atmosphere. Sulfate and nitrate are generated secondarily in atmospheric chemical reactions (Khoder, 2002). Those tropospheric chemical reactions are facilitated by solar energy (photochemical reaction) to which ambient temperature may be a proxy. Thus, temporal variation in ambient temperature may contribute to time-varying effect of PM. According to Korean Peninsula Climate Change Prospect Report (KMA, 2012), the annual mean temperature increased by 0.36 °C/ 10 years during 1981–2010 in Korea. However, even though increasing ambient temperature is one common phenomenon in climate change, it may not be completely explicative itself. To address time-varying PM effects, we suggest that future researches should concentrate on two tracks: in-depth relationship between solar energy (or whatever indicates it) and the generation of secondary particles in terms of fraction of secondary particles in total PM mass, and relationship between exposure to ambient PM controlled by indoor sources and individual behaviors. In time-series analysis for short-term effect of PM on mortality, multi-pollutant modeling has been performed for importance of individual toxicity adjusted for other pollutants. We repeated the same analyses in two-pollutant model including SO2, NO2 and O3. We found similar results with wider confidence intervals. There was only one exception to association of PM2.5 and respiratory mortality in SO2 included model; it showed declining pattern over time. We think that including SO2 may have led to block PM2.5 – sulfate pathway under SO2 – sulfate relationship. Thus, decreasing trends may correspond to increasing fraction of sulfate within PM2.5. However, provided that recent systematic review reported that sulfate was associated with all-cause, cardiovascular and respiratory mortality (Atkinson et al., 2014), we could not rule out chance. We conducted the same analyses using time-windows of one year (annual variation) and three years. We also performed a Bayesian VCR model to assess annual variation in PM effects on mortality. As a result, we found highly fluctuated variation with wide confidence intervals (for Bayesian VCR, posterior intervals) due to insufficient power. This fluctuation may result from mixture of random variability, inter-annual harvesting and change in characteristics of population and so on. Therefore, it is desirable to make inference from results of time windows approach which avoids relatively aforementioned problems (including insufficient power) and observed annual variations worth only supportive. In the present study, we adjusted annual variation in the size of population as an offset and found similar results. To give further discussion on population change, proportion of different sub-populations at a greater/weaker risk would be intuitive in timevarying risks. For example, temporal variation in elderly population who are likely to be at greater risk is arguable. The proportion of 65 þ age group in Seoul increased from 5.4% in 2000 to 9.6% in 2010 (Statistics Korea, 2010). However, we found that association of PM with cardiovascular mortality in 65 þ age group was weaker than that for those aged less than 65 while association of PM with all-cause and respiratory mortality in 65 þ age group was higher in our preliminary research (unpublished). Thus, the change in elderly population is not sufficient explication on time-varying PM effects which was consistently observed in all of the three disease categories. Kim et al. (2015) discussed that possible shifts (e.g. change in prevalence of underlying diseases) in context of population characteristics may explain a time-varying relative risk. The present study has several limitations. First, the number of
689
PM monitoring stations was not constant over the study period. It may be related with time-varying measurement error which biases estimates inconsistently across time-windows. However, PM10 monitoring stations operating in at least 75% of the districts have been available since 1999. Furthermore, temporal variations in association of PM2.5 with mortality were similar to those of PM10 even though there were less PM2.5 monitoring stations in early 2000 s. That is why bias due to measurement error may be stable across time windows. Second, we did not further look into specific causes in cardiovascular or respiratory mortality. Focusing on specific disease may be more relevant because reasons of timevarying effect can vary by causes of mortality. For example, Kim et al. (2015) reported difference in time-varying effects of temperature on cardiovascular (ICD-10: I20-52) and on cerebrovascular deaths (ICD-10: I60-69) which we treated as cardiovascular category. Therefore, our study is a preliminary research for future studies to cope with more specific causes of mortality. Given that (1) associations of both PM10 and PM2.5 with mortality increased over years, peaked during the mid-2000s and stronger association stood still in recent years, and (2) the ‘Special Act on Metropolitan Air Quality Improvement’ was initiated in 2005, concurrent control measures for air quality in Seoul may not play as favorable a role in reducing personal exposure to adverse particles as expected, even though intervention succeeded in decreasing PM mass markedly. Therefore, we suggest that management of outdoor air pollution should focus not only on decreasing PM mass but also on controlling its chemical composition and its number concentration of UP-sized particles. Furthermore, routine measurements on chemical composition and different sizes of PM2.5 should be encouraged to evaluate accomplishment from air quality policies. Temporal variation in an association between PM and mortality should be reflected in the field of health impact assessments (HIA), such as environmental burden of disease or cost-benefit analyses. Researchers apply relative risks estimated in epidemiological studies, so that results of HIA are inevitably affected by a timevarying relative risk. To our knowledge, no HIA study on PM has addressed its time-varying risk. That is, under the assumption that relative risk of PM is constant in Seoul, Korea, health impact attributable to PM would be underestimated (or the efficacy of air pollution intervention may be overestimated). To do valid impact assessment on ambient particles, therefore, further studies are needed to analyze time-varying association as well as to seek for causal factors related to temporal variation in PM effects. Our suggestions would provide insight on exploring true relationship between PM and mortality.
5. Conclusion In this study, we found that adverse effects of PM on mortality may increase over time in Seoul, Korea. In addition, unawareness of Asian dust intrusion in a statistical analysis may cause underestimation of PM effect on mortality in non-Asian dust period. To elucidate temporal variation in PM effect, we generated hypotheses on increase in fraction of toxic constituents within PM and decrease in representativeness of population average ambient PM to individual exposure to ambient-origin PM in time-series analyses. If those cannot be ruled out, management of air quality in Seoul may not actually minimize the health risks of PM as expected. Orientation and objectives of future air pollution policies should be established in consideration of causal factors of temporal variation in association between PM and mortality.
690
H. Kim et al. / Environmental Research 140 (2015) 684–690
Acknowledgments This work was financially supported by the National Research Foundation of Korea Grant (2014R1A2A1A11052556) funded by the Korea government (MSIP).
Appendix A. Supplementary material Supplementary data associated with this article can be found in the online version at: http://dx.doi.org/10.1016/j.envres.2015.05. 029.
References Atkinson, R.W., et al., 2014. Fine particle components and health—a systematic review and meta-analysis of epidemiological time series studies of daily mortality and hospital admissions. J. Exposure Sci. Environ. Epidemiol. Breitner, S., et al., 2009. Short-term mortality rates during a decade of improved air quality in Erfurt, Germany. Environ. Health Perspect. 117, 448–454. Diapouli, E., et al., 2013. Estimating the concentration of indoor particles of outdoor origin: a review. J. Air Waste Manag. Assoc. 63, 1113–1129. Dominici, F., et al., 2007. Particulate air pollution and mortality in the United States: did the risks change from 1987 to 2000? Am. J. Epidemiol. 166, 880–888. Dong, F., et al., 2013. Association between ambient PM10/PM2.5 levels and population mortality of circulatory diseases: a case-crossover study in Beijing. Beijing da xue xue bao. Yi xue ban ¼ J. Peking Univ. Health Sci. 45, 398–404. Godish, T., 1989. Indoor Air Pollution Control. CRC Press, United States of America. Heo, J., et al., 2014. Fine particle air pollution and mortality: importance of specific sources and chemical species. Epidemiology 25, 379–388. Ibald-Mulli, A., et al., 2002. Epidemiological evidence on health effects of ultrafine particles. J. Aerosol Med. 15, 189–201. Khoder, M., 2002. Atmospheric conversion of sulfur dioxide to particulate sulfate and nitrogen dioxide to particulate nitrate and gaseous nitric acid in an urban area. Chemosphere 49, 675–684. Kim, H., et al., 2015. Has the impact of temperature on mortality really decreased over time? Sci. Total Environ. 512, 74–81. Kim, K.-H., et al., 2003. The chemical composition of fine and coarse particles in relation with the Asian Dust events. Atmos. Environ. 37, 753–765. Kneib, T. et al., 2011. Baye sX: R utilities Accompanying the Software Package BayesX R Package Version 0.2–5. Korean Meteorological Administration (KMA), 2012. Korean Peninsula Climate Change Prospect Report, Available at: 〈http://www.climate.go.kr/home/cc_data/ 2013/korea_climate/13-0311_korea_climate.pdf〉. Korean Ministry of Environment (KMOE), 2013. Annual Report of Ambient Air
Quality in Korea 2012, Available at: 〈http://webbook.me.go.kr/DLi-File/091/018/ 011/5563269.pdf〉. Korean Ministry of Environment (KMOE), 2014. White Paper of Environment 2013, Available at: 〈http://webbook.me.go.kr/DLi-File/091/019/001/5566014.pdf〉. Kreyling, W.G., et al., 2003. Diverging long-term trends in ambient urban particle mass and number concentrations associated with emission changes caused by the German unification. Atmos. Environ. 37, 3841–3848. Lee, J.-T., et al., 2007. A comparison of mortality related to urban air particles between periods with Asian dust days and without Asian dust days in Seoul, Korea, 2000–2004. Environ. Res. 105, 409–413. Levy, J.I., et al., 2012. A meta-analysis and multisite time-series analysis of the differential toxicity of major fine particulate matter constituents. Am. J. Epidemiol. 175, 1091–1099. Peng, R.D., et al., 2005. Seasonal analyses of air pollution and mortality in 100 US cities. Am. J. Epidemiol. 161, 585–594. Penttinen, P., et al., 2001. Ultrafine particles in urban air and respiratory health among adult asthmatics. Eur. Respir. J. 17, 428–435. Pitz, M., et al., 2001. Change of the ambient particle size distribution in East Germany between 1993 and 1999. Atmos. Environ. 35, 4357–4366. Pope III, C.A., Dockery, D.W., 2006. Health effects of fine particulate air pollution: lines that connect. J. Air Waste Manag. Assoc. 56, 709–742. Sarnat, J.A., et al., 2002. Using sulfur as a tracer of outdoor fine particulate matter. Environ. Sci. Technol. 36, 5305–5314. Sarnat, S.E., et al., 2006. The influences of ambient particle composition and size on particle infiltration in Los Angeles, CA, residences. J. Air Waste Manag. Assoc. 56, 186–196. Schwartz, J., Neas, L.M., 2000. Fine particles are more strongly associated than coarse particles with acute respiratory health effects in schoolchildren. Epidemiology 11, 6–10. Seaton, A., et al., 1995. Particulate air pollution and acute health effects. Lancet 345, 176–178. Son, J.-Y., et al., 2012. Characterization of fine particulate matter and associations between particulate chemical constituents and mortality in Seoul, Korea. Environ. Health Perspect. 120, 872. Spengler, J., et al., 1981. Long-term measurements of respirable sulfates and particles inside and outside homes. Atmos. Environ. 15, 23–30. Stölzel, M., et al., 2006. Daily mortality and particulate matter in different size classes in Erfurt, Germany. J. Exposure Sci. Environ. Epidemiol. 17, 458–467. Stafoggia, M., et al., 2008. Does temperature modify the association between air pollution and mortality? A multicity case-crossover analysis in Italy. Am. J. Epidemiol. 167, 1476–1485. Statistics Korea, 2010. Population and Housing Census. Available at: 〈http://kosis.kr/ eng/statisticsList/statisticsList_01List.jsp?vwcd ¼MT_ETITLE&parmTabId ¼M_ 01_01#SubCont〉. Strand, M., et al., 2005. Estimating effects of ambient PM2.5 exposure on health using PM2.5 component measurements and regression calibration. J. Exposure Sci. Environ. Epidemiol. 16, 30–38. Zeger, S.L., et al., 2000. Exposure measurement error in time-series studies of air pollution: concepts and consequences. Environ. Health Perspect. 108, 419.
Contents lists available at ScienceDirect
Environmental Research journal homepage: www.elsevier.com/locate/envres
Effects of ambient air particles on mortality in Seoul: Have the effects changed over time? Honghyok Kim a, Hyomi Kim a, Jong-Tae Lee a,b,n a b
Department of Public Health Sciences, Graduate School, Korea University, 136-703 Seoul, Republic of Korea Department of Environmental Health, College of Health Science, Korea University, 136-703 Seoul, Republic of Korea
art ic l e i nf o
a b s t r a c t
Article history: Received 11 December 2014 Received in revised form 27 May 2015 Accepted 31 May 2015 Available online 14 June 2015
Background: Several studies have shown that there may be temporal variation in PM short-term effect on mortality. This temporal pattern may play an important role in evaluating air quality policies. Objectives: We investigated temporal variation in the association between PM and mortality in Seoul, Korea, 1998–2011. Methods: We adopted a generalized additive model and a series of time windows of five years to analyze temporal variation in associations between PM and all-cause, cardiovascular, and respiratory mortality. This time-window approach offers not only a comparison between one and the other half period but also successive variation. Time-varying associations were estimated only for days without Asian dust (dust storm blown from the Gobi desert) intrusion. Results: Annual average PM10 and PM2.5 total mass decreased from 70.0 to 46.9 mg/m3 and 44.4 to 23.4 mg/m3, respectively, during 2001–2011. A 10 mg/m3 increase in PM10 was associated with 0.16% (95% CI¼ 0.03% to 0.35%) additional all-cause deaths in 2002–2006 and it increased to 0.26% (95% CI ¼0.05– 0.48%) in 2007–2011. For PM2.5, the association increased from 0.35% (95% CI ¼ 0.02% to 0.71%) to 0.48% (95% CI ¼0.08–0.88%). For cardiovascular and respiratory mortality, increasing trends with stronger estimates were found. Conclusions: The present study showed temporally increasing trends in associations between PM and mortality. Current policies may not be as effective to reducing health risks attributable to PM as expected. Air quality interventions should be encouraged in terms of causal factors for time-varying association between PM and mortality. & 2015 Elsevier Inc. All rights reserved.
Keywords: Particulate matter Fine particle Mortality Temporal variation Time-varying effect
1. Introduction Epidemiological research has demonstrated that acute exposure to particulate matter with an aerodynamic diameter o10 mm (“PM10”) is associated with diverse health response (Pope III and Dockery, 2006). Incorporating advanced measurement technologies, researchers have since then concentrated on smaller particles, with aerodynamic diameters o2.5 mm (“PM2.5”) and provided consistent evidence of a stronger association with health outcomes than that of PM10 or PM10–2.5 (Dong et al., 2013; Schwartz and Neas, 2000). In Seoul, the capital of South Korea, many studies have reported an association between short-term exposure to PM and an increase in risk of health outcomes (Heo et al., 2014; Lee et al., 2007; Son et al., 2012). n Corresponding author at: Department of Environmental Health, College of Health Science, Korea University, Seoul, Republic of Korea. Fax: þ 82 2 2298 0248. E-mail address: [email protected] (J.-T. Lee).
http://dx.doi.org/10.1016/j.envres.2015.05.029 0013-9351/& 2015 Elsevier Inc. All rights reserved.
“Asian dust,” also known as yellow dust, which originates in the Gobi Desert spread from Southern Mongolia to Northern China, is sporadically blown to Korean peninsula by westerly winds and aggravates air quality in the region. It leads to increase particulate concentrations, but especially tends to increase coarse particles with crustal components (Kim et al., 2003). In fact, such particles may be less deleterious to human health than fine particles (Pope III and Dockery, 2006). Thus, an association of particulates with mortality regardless of excluding days with Asian dust intrusion in a statistical model would possibly be underestimated in terms of PM effects for dominant non-Asian dust days (Lee et al., 2007). Accordingly, it is necessary to take this concept into account when analyzing PM effects on mortality in Korea. According to recent studies, there may be temporal variation in short-term effects of air pollution; the relative risk is not constant over time. Studies in the U.S. and Europe found decreasing patterns in PM effects over time (Breitner et al., 2009; Dominici et al., 2007). They suggested that air quality control policies may have
H. Kim et al. / Environmental Research 140 (2015) 684–690
had an impact on not only reducing PM mass concentration but also on desirable changes in the constituents of PM. If there are similar trends in Korea, a time-varying effect of PM may also be expected. Thus, we sought to analyze an association between PM and mortality in Seoul, considering the intrusion of Asian dust. Second, we investigated temporal variation in the association, and suggested possible reasons for a time-varying effect of PM on mortality. Finally, we made suggestions on future studies and ambient air quality policies.
2. Methods and materials 2.1. The scope and data The study location was Seoul, Korea, with a population of about 10 million (Statistics Korea, 2010). Data for mortality, PM10, PM2.5, and meteorological factors were obtained from Statistics Korea, the National Institute of Environmental Research, Seoul Metropolitan Government Research Institute of Public Health and Environment, and the Korea Metrological Administration, respectively. PM concentration in Seoul has been measured in national monitoring stations since 1995 for PM10 and 2001 for PM2.5. PM values were computed as daily (24 h) mean concentrations from hourly measurements. Valid daily mean concentration was calculated with data from at least 75% of stations (18 þ out of 25) and of hourly measurements (16 þ out of 24). Otherwise, we imputed an annual average from valid daily estimates. Missing values were also imputed with annual averages. Exceptions to spatial criteria were the year of 1998 for PM10 because of only 17 monitoring stations and the period of 2001–2004 for PM2.5 due to almost 10 of unevenly operating stations. The number of invalid daily PM10 was 270 (5.3%) from 1998 to 2011 and that of invalid daily PM2.5 was 563 (14.0%) from 2001 to 2011. Temperature and relative humidity were measured at a single fixed station in the Jongno district, a center of Seoul city. Daily mean temperature and relative humidity were calculated and there were no missing values for them. We classified cause-specific mortality into all-cause except for accidental causes (International Classification of Disease, Tenth Revision [ICD-10]: A00-R99 and Ninth Revision [ICD-9]: 1-799), cardiovascular (ICD-10: I00-I99, ICD-9: 390-459), and respiratory (ICD-10: J00-J99, ICD-9: 460-519) mortality. 2.2. Statistical analysis A generalized additive model (GAM) was used to analyze associations between PM10 and PM2.5 and mortality. Poisson distribution with over-dispersion was applied. To control potential confounders, we adjusted daily mean temperature, daily mean relative humidity, time-trend, day of the week, and holidays. Temperature, relative humidity, and time-trends were adjusted as a cubic regression spline with 5 knots and 7 knots/year. Knots were equally dispersed throughout values. In preliminary analysis, we did not detect partial autocorrelation which exceeded |0.1| in the main model with aforementioned knots. Day of the week and holidays were included in the model as dummy variables. We compared estimates of days without Asian dust intrusion (NAD, non-Asian dust days) to those from all days including days with Asian dust intrusion (NAD þAD, non-Asian dust Days plus Asian dust days). Then we conducted subsequent analyses for NAD. To explore temporal variation in associations between PM and cause-specific mortality, we used a series of time windows, five years long. This approach has been used in our previous work (Kim
685
et al., 2015). It enables to compare one and the other half period as many studies on temporal variation in relative risk did. It also allows to see consecutive variation with sufficient statistical power and avoids inter-annual harvesting effect (Kim et al., 2015). For statistical analyses, the SAS 9.4 and R 3.0.2 softwares were used. 2.3. Sensitivity analysis We changed the number of knots in spline function for temperature (4, 5 and 6), relative humidity (4, 5 and 6) and time (4, 5, 6 and 8) to see whether estimated association of PM with mortality was stable. As the size of population changed over the study period, annual variation of population was also adjusted as an offset in a gam model across time-windows, even though we expected that spline function for time-trend may cover the population change sufficiently and change in the size of population in a single city is relatively minimal. The annual number of population was obtained from the statistics of registered population in Korean Statistical Information Service. We also conducted sensitivity analyses using 1-year (annual variation) and 3-years of windows. Bayesian varying coefficient regression (VCR) was used to explore time-varying association between PM and mortality: βPM = f (YEAR )PM with ‘BayesX’ package in R software (Kneib et al., 2011). Similar approach was adapted elsewhere (Breitner et al., 2009). In addition, multi-pollutant model needs to be taken into consideration if temporal variation in PM effect is (partially) attributed to other pollutants. We added SO2, NO2, or O3 into the main model (so that two-pollutant model) to see if it shows different temporal patterns of PM effects.
3. Results 3.1. Descriptive statistics Regardless of Asian dust intrusion, the average daily PM10 concentration for the study period, 1998–2011, was 60.7 mg/m3. For PM2.5, it was 32.1 mg/m3 in 2001–2011. There were a total of 159 Asian dust days. Average concentrations of PM10 and PM2.5 were higher during Asian dust intrusion in the corresponding periods (Table 1). Annual ambient PM concentrations have declined consistently (Fig. 1) The average daily all-cause mortality was 93.3. Death count in those aged 65 years or older (65 þ ) represented two-thirds of all ages (Table 1). For cause-specific deaths, the averages were 24.4 and 5.7 in cardiovascular and respiratory-related death, respectively. Daily death count increased slightly over time as the elderly population grew (Supplemental materials, Fig. S1). 3.2. Quantification of PM effect on mortality Table 2 shows associations between PM10, PM2.5, and mortality. An increment of 10 mg/m3 in PM10 was associated with a 0.12% (95% CI ¼0.12–0.21%) increase in all-cause mortality in 1998–2011. For PM2.5, it was associated with a 0.20 (95% CI ¼ 0.01% to 0.41%) increase in risk of all-cause mortality in 2001–2011. When we excluded days with Asian dust intruded, both associations of PM10 and PM2.5 were strengthened to 0.22% (95% CI ¼ 0.10% to 0.35%) and 0.23% (95% CI ¼ 0.01% to 0.47%), respectively, for the corresponding periods. Looking into cause-specific mortality, stronger associations of PM were found compared to risk in all-cause mortality. As with all-cause mortality, estimates for NAD were higher. To confirm this effect modification, we checked statistical significance of an interaction between PM and the occurrence of
686
H. Kim et al. / Environmental Research 140 (2015) 684–690
Table 1 Descriptive statistics, Seoul, 1998–2011.
b
25th 50th
75th
100th
Periods
Exposure
Asian dust
% Increase (95% CI)
5.0 37.4 3.4 14.4 15.7
36.9 76.7 19.7 29.0 4.1
54.0 110.8 29.7 39.6 14.4
72.1 162.8 37.9 61.1 22.0
248.2 869.1 161.0 362.9 30.4
All-cause
1998–2011
PM10
2001–2011
PM10
NAD þ ADa NADb NAD þ ADa NADb NAD þ ADa NADb
0.12 0.22 0.10 0.19 0.20 0.23
(0.02, 0.21) (0.10, 0.35) (0, 0.20) (0.05, 0.33) ( 0.01, 0.41) ( 0.01, 0.47)
NAD þ ADa NADb NAD þ ADa NADb NAD þ ADa NADb
0.19 0.39 0.23 0.44 0.47 0.69
(0.02, 0.37) (0.16, 0.62) (0.03, 0.42) (0.18, 0.70) (0.07, 0.86) (0.24, 1.15)
NAD þ ADa NADb NAD þ ADa NADb NAD þ ADa NADb
0.33 0.75 0.31 0.77 0.92 1.34
( 0.03, 0.7) (0.27, 1.23) ( 0.09, 0.71) (0.22, 1.31) (0.10, 1.75) (0.39, 2.30)
Days Mean 0th
PM10 (mg/m3) (1998–2011) PM2.5 (mg/m3) (2001–2011) Daily mean temperature (°C) Daily mean humidity (%) All-cause deaths Z65 Years of age Cardiovascular deaths Z65 Years of age Respiratory deaths Z65 Years of age a
Table 2 Association between PM10 and PM2.5 with mortality, Seoul, 1998–2011.
NADa 4954 58.3 ADb 159 135.7 NADa 3887 31.4 b AD 130 52.0 5113 12.9
5113
62.0
19.4
51.2
62.5
72.5
96.3
5113 5113 5113
93.3 61.7 24.4
55 28 8
85 55 20
93 61 24
101 68 28
148 103 49
5113 5113 5113
18.0 5.7 4.8
5 0 0
15 4 3
18 5 5
21 7 6
37 21 18
PM2.5
CVD (I00-99)
1998–2011
PM10
2001–2011
PM10 PM2.5
RES (J00-99)
1998–2011
PM10
2001–2011
PM10
Days without Asian dust intrusion (NAD, non-Asian dust days). Days with Asian dust intrusion (AD, Asian dust days).
Asian dust in models. As a result, an interaction between PM10 and Asian dust intrusion was statistically significant (po 0.001). For PM2.5, it was significant in risk of cardiovascular mortality only (p o0.05). Therefore, subsequent analyses were conducted only for NAD. Using a series of time-windows, we sought to explore whether there was temporal variation in associations between PM and mortality. An association of PM10 with mortality increased from 0.16% (95% CI ¼ 0.03% to 0.35%) in 2002–2006 to 0.26% (95% CI ¼0.05–0.48%) in 2007–2011. For PM2.5, it increased from 0.35% (95% CI ¼ 0.02% to 0.71%) to 0.48% (95% CI ¼0.08–0.88%). The strongest associations of PM10 and PM2.5 and all-cause mortality were 0.37% (95% CI ¼0.15% to 0.58%) in 2005–2009, and 0.63% (95% CI ¼0.18–1.09%) in 2004–2008, respectively. Similar patterns were also seen in association with cardiovascular mortality while increasing trends were markedly for respiratory mortality. In most strata, associations in the most recent period were stronger than those in the oldest (Table 3). 3.3. Sensitivity analysis We repeated the main analysis changing the number of knots of cubic splines for temperature (4, and 6), relative humidity (4, and 6), and time (4, 6, and 8). These showed similar results (results not shown). The adjustment for annual population variation did not change temporal variation in associations of PM with mortality (results not shown). We also changed the length of the time
PM2.5
a
The study period including days with Asian dust intrusion (NAD þ AD, non-Asian dust days þ Asian
dust days). b Days without Asian dust intrusion (NAD, non-Asian dust days).
windows to one year (annual variation) and three years (Supplemental materials, Table S1-2). However, due to insufficient power, we found few significant associations across time-windows. As an alternative method, using Bayesian VCR, annual variations in associations of PM10 and PM2.5 and mortality were analyzed (Fig. 2). Even though this method was also not away from insufficient power, results were similar. In contrast to trends for all-cause and cardiovascular mortality, the association of PM with respiratory mortality increased last three (PM2.5) and four (PM10) years. The same analysis for all days including Asian dust days showed that the temporal patterns did not change (results not shown). In two-pollutant modeling, we found association between PM and mortality with wider confidence intervals. The two-pollutant models showed temporal patterns as the one-pollutant (PM) models did, except for the model with PM2.5 and SO2 for respiratory-related mortality; an estimate decreased from 1.79 (95% CI: 0.02 to 3.63) in 2002–2006 to 1.01 (95% CI: 0.98 to 3.05) in 2007–2011 (Supplemental materials Table S3-5). For annual variation from Bayesian VCR, trends in the two-pollutant models with wider posterior intervals were similar to those in the one-
Fig. 1. Annual trends in ambient concentrations of PM10 and PM2.5, Seoul, 1998–2011.
H. Kim et al. / Environmental Research 140 (2015) 684–690
687
Table 3 Temporal variation in associations between PM10 and PM2.5 and mortality for days without Asian dust intrusion using a series of 5-year time windows, Seoul, 1998–2011. % Increase (95% CI) All-cause
1998–2002 1999–2003 2000–2004 2001–2005 2002–2006 2003–2007 2004–2008 2005–2009 2006–2010 2007–2011
Cardiovascular
Respiratory
PM10
PM2.5
PM10
PM2.5
PM10
0.29 0.17 0.06 0.08 0.16 0.18 0.31 0.37 0.29 0.26
– – – 0.10 ( 0.22, 0.41) 0.35 ( 0.02, 0.71) 0.29 ( 0.08, 0.67) 0.63 (0.18, 1.09) 0.56 (0.10, 1.02) 0.43 (0.01, 0.85) 0.48 (0.08, 0.88)
0.41 0.24 0.27 0.35 0.43 0.39 0.68 0.75 0.64 0.60
– – – 0.58 (0.01, 1.15) 0.81 (0.16, 1.47) 0.53 ( 0.14, 1.22) 1.26 (0.43, 2.1) 1.13 (0.27, 2.01) 1.10 (0.28, 1.92) 1.12 (0.35, 1.9)
0.46 0.56 0.25 0.22 0.44 0.91 1.05 1.22 1.43 1.51
(0.09, 0.50) ( 0.01, 0.35) ( 0.12, 0.24) ( 0.11, 0.27) ( 0.03, 0.35) ( 0.01, 0.36) (0.09, 0.52) (0.15, 0.58) (0.07, 0.50) (0.05, 0.48)
(0.04, 0.78) (-0.09, 0.57) (-0.07, 0.60) (0.01, 0.70) (0.08, 0.77) (0.05, 0.73) (0.30, 1.07) (0.34, 1.15) (0.23, 1.05) (0.19, 1.02)
pollutant models (results not shown). The exception was not found.
4. Discussion In this study, we found that association of PM with mortality was enlarged when excluding days with Asian dust intrusion. These results were consistent with the previous study suggesting that increased PM concentration attributable to Asian dust intrusion contributed relatively less toxic particles fraction, so that it may cause underestimated risks of PM (Lee et al., 2007). Also, it was pointed out that recognition of an Asian dust event may change individual behavior to minimize exposure to ambient air particles (Lee et al., 2007). We emphasize that those two are related with bias to underestimating risks by using less representative population averages of ambient PM concentration as an indicator to the average of individual exposure to adverse particles in time-series analysis. Therefore, the price of using population average of ambient PM level in time-series analysis should be taken into account for impact assessment of PM for dominant non-Asian dust days. On the other hand, indicators which represent relatively less toxic fraction (in this case, PM10) are useful to assess impact of PM during Asian dust intrusion. Our efforts to explore temporal variation in an association of PM with mortality originated in curiosity about achievement from receptor-oriented air pollution reduction policies in Korea. To prevent deleterious PM effects, the Korean Ministry of Environment (KMOE) promulgated the first regulation on ambient PM10 concentration in 1993, and initiated the PM10 quality standard in
( 0.32, 1.25) ( 0.16, 1.29) ( 0.48, 0.99) ( 0.54, 1.00) ( 0.35, 1.24) (0.11, 1.71) (0.23, 1.87) (0.41, 2.04) (0.65, 2.21) (0.75, 2.28)
PM2.5 – – – 0.88 ( 0.37, 2.14) 1.48 ( 0.03, 3.02) 2.01 (0.42, 3.63) 2.15 (0.34, 4.00) 2.18 (0.40, 3.99) 2.20 (0.64, 3.79) 2.33 (0.92, 3.76)
1995 under the Clean Air Conservation Act (KMOE, 2013). The standard concentrations were 80 mg/m3 for annual average and 150 mg/m3 for daily average. The limit of annual average was intensified to 70 mg/m3 in 2001. In 2007, this standard changed to 50 mg/m3 and the existing daily average standard was tightened upto 100 mg/m3. The new standard for outdoor PM2.5 was announced in 2012 and will come into effect after 2015. It follows the recommended levels of interim target-2 in World Health Organization air quality guidelines: yearly 25 mg/m3 and daily 50 mg/m3. Furthermore, as targeting the capital and its neighboring areas where nearly half the total population of South Korea lives, the ‘Special Act on Metropolitan Air Quality Improvement’ was established in 2003 in order to improve air quality by cutting emissions of PM, NOx, and SOx markedly (KMOE, 2014). This Act, which has been in effect since 2005, involves reductions in automobile emissions, encouraging eco-friendly energy, air pollutant total mass emission management system, and aR tradable permit program. These two endeavors have led to a consistent decrease in ambient PM concentrations in metro Seoul. In fact, such policies are expected to encourage improvements in air quality in terms of not only ambient PM mass concentration, but also its composition. However, due to inadequate awareness and inconsistent measurements of its constituents, only long-term trends in PM mass, but not composition (i.e., whether it is less harmful/toxic), have so far been evaluated in Korea. Therefore, from an epidemiological point of view, exploration on a timevarying (i.e., a temporally decreasing trend in) association of PM and mortality can be an alternative if PM constituents vary over time. Our results showed that a relative risk of acute exposure to
Fig. 2. Annual variations in association of PM with mortality using Bayesian varying coefficient regression (VCR), Seoul, 1998–2011; estimates and 95% posterior intervals.
688
H. Kim et al. / Environmental Research 140 (2015) 684–690
ambient PM in mortality increased over time-windows. We did not find similar results elsewhere. Rather, Breitner et al. (2009) suggested that an association between short-term exposure to air pollution and mortality in Erfurt, Germany decreased due to air pollution controls. A study using data of the National Morbidity Mortality Air pollution Study in the US showed that PM effects on mortality seemed to decrease over time (Dominici et al., 2007). This study also suggested that it was probably a result of national and regional air quality policies, such as reducing allowable emissions of diesel particulate matter, nitrogen, and hydrocarbons. The present study indicates a significantly decreasing trend in ambient PM concentration like other regions, probably because of strengthened air quality standards and implemented policies. However, these interventions seemed not to consistently improve air quality in view point of human health risks. Breitner et al. (2009) questioned that temporal variation in effects of PM may depend on a non-linear dose–response curve between PM and mortality. This study conducted a likelihood-ratio test for two models: one linear and the other non-linear using a regression spline. We also tested likelihood ratio between the two models, and found a linear relationship between PM and mortality as they did. This indicates that time-varying effect of PM may not depend on shape of dose–response function. Therefore, we suggest that two possible reasons for the observed temporal variation in the present study should be highlighted: (1) the increased toxicity of PM and (2) reduction in bias induced by markedly decreasing ambient PM mass concentration. Unlike other ambient risk factors, such as temperature, SO2, NO2, O3,etc., “PM” is not a single concept but a complex mixture of diverse physiochemical (even biological) compounds. That is why even the same mass of PM has different toxicity depending on its composition. ‘PM10’ is commonly divided into PM2.5 and PM10–2.5. Furthermore, PM2.5 is also a mixture of different sized particles, including ultrafine particles (UPs). UPs consist of not only single but also aggregated particles with aerodynamic diameters o100 nm (Ibald-Mulli et al., 2002). Indeed, it has been hypothesized that UPs are responsible for PM effects at ambient concentrations (Seaton et al., 1995). Several epidemiological studies reported that number concentration (NC) of UP was more strongly (Stölzel et al., 2006) or only associated (Penttinen et al., 2001) with adverse health responses. Recently, associations of health outcomes with NC of Aitken mode and nucleation mode particles were discussed by a systematic review (Atkinson et al., 2014). Thus, one possibility for the increasing trend in the present study is that fraction of UP-scale particles (or even smaller particles) may have increased over time. Although there is no reported data or study regarding its temporal variation in Korea, to our knowledge, several studies conducted in Germany support the idea. One study found that in constituents of PM2.5 from 1993 to 1999 in East Germany, the number concentration of UPs was steady while NC of relatively larger particles (PM0.1–2.5) decreased and, consequently, PM2.5 mass concentration decreased (Pitz et al., 2001). This study mentioned that improved engine technologies (i.e., lower-emission diesels) were successful in reducing PM2.5 mass concentration but may have paradoxically resulted in emission of nanoparticles o50 nm. Other study which measured such long-term variation of PM emphasized that an improvement in ambient air quality in terms of mass concentration may not reduce, or may even increase, health risks (Kreyling et al., 2003). Along with consideration of size of PM, toxicity of chemical constituents in PM2.5 has been discussed as well. Indeed, recent multi-city and meta-analysis studies have reported associations of specific components with health end points (Atkinson et al., 2014; Levy et al., 2012). Albeit of issues on heterogeneity across individual studies, organic carbon, elemental carbon, nitrate, sulfate, nickel, and lead were generally associated with health outcomes.
In Korea, two studies investigated associations of chemical species with mortality in Seoul using different measurements. One study reported that nitrate, sulfate, and ammonium, which contributed a large proportion of PM2.5 total mass, were marginally associated with cardiovascular mortality (Son et al., 2012). The other showed that organic carbon, elemental carbon, and lead were associated with mortality and suggested that deleterious effects of PM2.5 may be mainly ascribed to local combustion sources (Heo et al., 2014). Given that the distribution of composition of PM2.5 depends on emission sources controlled by air quality policies and interventions, it will be necessary to measure the constituents of PM2.5 over the long-term, and to analyze association of the smallest unit of noxious particles as possible with mortality. Only change in ambient PM mass itself was not related to temporal variations in an effect of ambient PM on mortality. However, the extent of representativeness of population average ambient concentration to average of individual exposure in a timeseries analysis may depend markedly on change in PM mass. A biased estimate in Poisson time-series regression is directly related to the representativeness. This issue was scrutinized elsewhere (Zeger et al., 2000). On the assumption that a relative risk is constant over time, either an infiltration factor, a fraction of particles penetrating into indoors and remaining suspended (Diapouli et al., 2013) or of outdoor exposure to ambient-originated pollution (Zeger et al., 2000), is assumed to be less movable or stable. However, in context of a time-varying effect of ambient PM on mortality, it may be expected to change so much as it cannot be ignored. An infiltration factor of fine particle cannot easily be calculated because of indoor sources of PM such as environmental tobacco smoke (Strand et al., 2005). Alternatively, some researchers have used sulfate, originating mostly in outdoor environment, as a tracer of PM2.5 (Sarnat et al., 2002). The result of the study conducted from Spengler et al. (1981) showed that the difference in sulfate concentration between indoor and outdoor was smaller in areas with a lower concentration of outdoor sulfate. Godish (1989) noted that for air pollutants which do not derive from indoor sources, indoor/outdoor ratio is near to unity, and for SO2, particularly, it approaches relatively to unity in environment with low SO2 concentration. Provided that particles with diameter less than 1 mm are more highly infiltrated into indoors than larger particles (Sarnat et al., 2006) and reduction in bigger particles may be a main contributor to lowering total mass concentration, proportion of infiltrated particles may have increased over time. In addition, dramatic reduction of air pollution concentration may also encourage people to spend more time outside. Thus, it can be expected that decreasing ambient mass of PM may accompany an increasing either infiltration factor or fraction of exposure to outdoor-originated particles, so that bias from time-series analysis may shrink. Consequently, attenuating underestimation on relative risk may result in an increasing pattern of ambient PM effect. Bias from representativeness of population average in timeseries analysis should be also discussed in temporal variation in heat effect on mortality. An increasing rate of air conditioner use, a major factor, can cause average of ambient temperature to be less representative. It can magnify the degree of underestimation. A change in degree of underestimation is cumbersome to estimate a true risk in a statistical analysis. On the other hand, in context of policy efficacy, it can be viewed as a result of interventions (Zeger et al., 2000). Thus, our results imply that intensified air quality regulation might not have had desirable impact on personal risk due to ambient-origin PM in Seoul. Epidemiologists have researched effect modification by temperature in association of PM with mortality (Dong et al., 2013; Stafoggia et al., 2008). Two major aspects are of interest; the
H. Kim et al. / Environmental Research 140 (2015) 684–690
possibility of increasing exposure to ambient PM by keeping windows open during the summer season (Stafoggia et al., 2008) and, of a more toxic PM composition (Peng et al., 2005). These two issues are involved in our suggested hypotheses. The first is related with representativeness of outdoor PM. Regarding the increased toxicity of PM, this may be related to the formation of secondary sulfate and nitrate in the atmosphere. Sulfate and nitrate are generated secondarily in atmospheric chemical reactions (Khoder, 2002). Those tropospheric chemical reactions are facilitated by solar energy (photochemical reaction) to which ambient temperature may be a proxy. Thus, temporal variation in ambient temperature may contribute to time-varying effect of PM. According to Korean Peninsula Climate Change Prospect Report (KMA, 2012), the annual mean temperature increased by 0.36 °C/ 10 years during 1981–2010 in Korea. However, even though increasing ambient temperature is one common phenomenon in climate change, it may not be completely explicative itself. To address time-varying PM effects, we suggest that future researches should concentrate on two tracks: in-depth relationship between solar energy (or whatever indicates it) and the generation of secondary particles in terms of fraction of secondary particles in total PM mass, and relationship between exposure to ambient PM controlled by indoor sources and individual behaviors. In time-series analysis for short-term effect of PM on mortality, multi-pollutant modeling has been performed for importance of individual toxicity adjusted for other pollutants. We repeated the same analyses in two-pollutant model including SO2, NO2 and O3. We found similar results with wider confidence intervals. There was only one exception to association of PM2.5 and respiratory mortality in SO2 included model; it showed declining pattern over time. We think that including SO2 may have led to block PM2.5 – sulfate pathway under SO2 – sulfate relationship. Thus, decreasing trends may correspond to increasing fraction of sulfate within PM2.5. However, provided that recent systematic review reported that sulfate was associated with all-cause, cardiovascular and respiratory mortality (Atkinson et al., 2014), we could not rule out chance. We conducted the same analyses using time-windows of one year (annual variation) and three years. We also performed a Bayesian VCR model to assess annual variation in PM effects on mortality. As a result, we found highly fluctuated variation with wide confidence intervals (for Bayesian VCR, posterior intervals) due to insufficient power. This fluctuation may result from mixture of random variability, inter-annual harvesting and change in characteristics of population and so on. Therefore, it is desirable to make inference from results of time windows approach which avoids relatively aforementioned problems (including insufficient power) and observed annual variations worth only supportive. In the present study, we adjusted annual variation in the size of population as an offset and found similar results. To give further discussion on population change, proportion of different sub-populations at a greater/weaker risk would be intuitive in timevarying risks. For example, temporal variation in elderly population who are likely to be at greater risk is arguable. The proportion of 65 þ age group in Seoul increased from 5.4% in 2000 to 9.6% in 2010 (Statistics Korea, 2010). However, we found that association of PM with cardiovascular mortality in 65 þ age group was weaker than that for those aged less than 65 while association of PM with all-cause and respiratory mortality in 65 þ age group was higher in our preliminary research (unpublished). Thus, the change in elderly population is not sufficient explication on time-varying PM effects which was consistently observed in all of the three disease categories. Kim et al. (2015) discussed that possible shifts (e.g. change in prevalence of underlying diseases) in context of population characteristics may explain a time-varying relative risk. The present study has several limitations. First, the number of
689
PM monitoring stations was not constant over the study period. It may be related with time-varying measurement error which biases estimates inconsistently across time-windows. However, PM10 monitoring stations operating in at least 75% of the districts have been available since 1999. Furthermore, temporal variations in association of PM2.5 with mortality were similar to those of PM10 even though there were less PM2.5 monitoring stations in early 2000 s. That is why bias due to measurement error may be stable across time windows. Second, we did not further look into specific causes in cardiovascular or respiratory mortality. Focusing on specific disease may be more relevant because reasons of timevarying effect can vary by causes of mortality. For example, Kim et al. (2015) reported difference in time-varying effects of temperature on cardiovascular (ICD-10: I20-52) and on cerebrovascular deaths (ICD-10: I60-69) which we treated as cardiovascular category. Therefore, our study is a preliminary research for future studies to cope with more specific causes of mortality. Given that (1) associations of both PM10 and PM2.5 with mortality increased over years, peaked during the mid-2000s and stronger association stood still in recent years, and (2) the ‘Special Act on Metropolitan Air Quality Improvement’ was initiated in 2005, concurrent control measures for air quality in Seoul may not play as favorable a role in reducing personal exposure to adverse particles as expected, even though intervention succeeded in decreasing PM mass markedly. Therefore, we suggest that management of outdoor air pollution should focus not only on decreasing PM mass but also on controlling its chemical composition and its number concentration of UP-sized particles. Furthermore, routine measurements on chemical composition and different sizes of PM2.5 should be encouraged to evaluate accomplishment from air quality policies. Temporal variation in an association between PM and mortality should be reflected in the field of health impact assessments (HIA), such as environmental burden of disease or cost-benefit analyses. Researchers apply relative risks estimated in epidemiological studies, so that results of HIA are inevitably affected by a timevarying relative risk. To our knowledge, no HIA study on PM has addressed its time-varying risk. That is, under the assumption that relative risk of PM is constant in Seoul, Korea, health impact attributable to PM would be underestimated (or the efficacy of air pollution intervention may be overestimated). To do valid impact assessment on ambient particles, therefore, further studies are needed to analyze time-varying association as well as to seek for causal factors related to temporal variation in PM effects. Our suggestions would provide insight on exploring true relationship between PM and mortality.
5. Conclusion In this study, we found that adverse effects of PM on mortality may increase over time in Seoul, Korea. In addition, unawareness of Asian dust intrusion in a statistical analysis may cause underestimation of PM effect on mortality in non-Asian dust period. To elucidate temporal variation in PM effect, we generated hypotheses on increase in fraction of toxic constituents within PM and decrease in representativeness of population average ambient PM to individual exposure to ambient-origin PM in time-series analyses. If those cannot be ruled out, management of air quality in Seoul may not actually minimize the health risks of PM as expected. Orientation and objectives of future air pollution policies should be established in consideration of causal factors of temporal variation in association between PM and mortality.
690
H. Kim et al. / Environmental Research 140 (2015) 684–690
Acknowledgments This work was financially supported by the National Research Foundation of Korea Grant (2014R1A2A1A11052556) funded by the Korea government (MSIP).
Appendix A. Supplementary material Supplementary data associated with this article can be found in the online version at: http://dx.doi.org/10.1016/j.envres.2015.05. 029.
References Atkinson, R.W., et al., 2014. Fine particle components and health—a systematic review and meta-analysis of epidemiological time series studies of daily mortality and hospital admissions. J. Exposure Sci. Environ. Epidemiol. Breitner, S., et al., 2009. Short-term mortality rates during a decade of improved air quality in Erfurt, Germany. Environ. Health Perspect. 117, 448–454. Diapouli, E., et al., 2013. Estimating the concentration of indoor particles of outdoor origin: a review. J. Air Waste Manag. Assoc. 63, 1113–1129. Dominici, F., et al., 2007. Particulate air pollution and mortality in the United States: did the risks change from 1987 to 2000? Am. J. Epidemiol. 166, 880–888. Dong, F., et al., 2013. Association between ambient PM10/PM2.5 levels and population mortality of circulatory diseases: a case-crossover study in Beijing. Beijing da xue xue bao. Yi xue ban ¼ J. Peking Univ. Health Sci. 45, 398–404. Godish, T., 1989. Indoor Air Pollution Control. CRC Press, United States of America. Heo, J., et al., 2014. Fine particle air pollution and mortality: importance of specific sources and chemical species. Epidemiology 25, 379–388. Ibald-Mulli, A., et al., 2002. Epidemiological evidence on health effects of ultrafine particles. J. Aerosol Med. 15, 189–201. Khoder, M., 2002. Atmospheric conversion of sulfur dioxide to particulate sulfate and nitrogen dioxide to particulate nitrate and gaseous nitric acid in an urban area. Chemosphere 49, 675–684. Kim, H., et al., 2015. Has the impact of temperature on mortality really decreased over time? Sci. Total Environ. 512, 74–81. Kim, K.-H., et al., 2003. The chemical composition of fine and coarse particles in relation with the Asian Dust events. Atmos. Environ. 37, 753–765. Kneib, T. et al., 2011. Baye sX: R utilities Accompanying the Software Package BayesX R Package Version 0.2–5. Korean Meteorological Administration (KMA), 2012. Korean Peninsula Climate Change Prospect Report, Available at: 〈http://www.climate.go.kr/home/cc_data/ 2013/korea_climate/13-0311_korea_climate.pdf〉. Korean Ministry of Environment (KMOE), 2013. Annual Report of Ambient Air
Quality in Korea 2012, Available at: 〈http://webbook.me.go.kr/DLi-File/091/018/ 011/5563269.pdf〉. Korean Ministry of Environment (KMOE), 2014. White Paper of Environment 2013, Available at: 〈http://webbook.me.go.kr/DLi-File/091/019/001/5566014.pdf〉. Kreyling, W.G., et al., 2003. Diverging long-term trends in ambient urban particle mass and number concentrations associated with emission changes caused by the German unification. Atmos. Environ. 37, 3841–3848. Lee, J.-T., et al., 2007. A comparison of mortality related to urban air particles between periods with Asian dust days and without Asian dust days in Seoul, Korea, 2000–2004. Environ. Res. 105, 409–413. Levy, J.I., et al., 2012. A meta-analysis and multisite time-series analysis of the differential toxicity of major fine particulate matter constituents. Am. J. Epidemiol. 175, 1091–1099. Peng, R.D., et al., 2005. Seasonal analyses of air pollution and mortality in 100 US cities. Am. J. Epidemiol. 161, 585–594. Penttinen, P., et al., 2001. Ultrafine particles in urban air and respiratory health among adult asthmatics. Eur. Respir. J. 17, 428–435. Pitz, M., et al., 2001. Change of the ambient particle size distribution in East Germany between 1993 and 1999. Atmos. Environ. 35, 4357–4366. Pope III, C.A., Dockery, D.W., 2006. Health effects of fine particulate air pollution: lines that connect. J. Air Waste Manag. Assoc. 56, 709–742. Sarnat, J.A., et al., 2002. Using sulfur as a tracer of outdoor fine particulate matter. Environ. Sci. Technol. 36, 5305–5314. Sarnat, S.E., et al., 2006. The influences of ambient particle composition and size on particle infiltration in Los Angeles, CA, residences. J. Air Waste Manag. Assoc. 56, 186–196. Schwartz, J., Neas, L.M., 2000. Fine particles are more strongly associated than coarse particles with acute respiratory health effects in schoolchildren. Epidemiology 11, 6–10. Seaton, A., et al., 1995. Particulate air pollution and acute health effects. Lancet 345, 176–178. Son, J.-Y., et al., 2012. Characterization of fine particulate matter and associations between particulate chemical constituents and mortality in Seoul, Korea. Environ. Health Perspect. 120, 872. Spengler, J., et al., 1981. Long-term measurements of respirable sulfates and particles inside and outside homes. Atmos. Environ. 15, 23–30. Stölzel, M., et al., 2006. Daily mortality and particulate matter in different size classes in Erfurt, Germany. J. Exposure Sci. Environ. Epidemiol. 17, 458–467. Stafoggia, M., et al., 2008. Does temperature modify the association between air pollution and mortality? A multicity case-crossover analysis in Italy. Am. J. Epidemiol. 167, 1476–1485. Statistics Korea, 2010. Population and Housing Census. Available at: 〈http://kosis.kr/ eng/statisticsList/statisticsList_01List.jsp?vwcd ¼MT_ETITLE&parmTabId ¼M_ 01_01#SubCont〉. Strand, M., et al., 2005. Estimating effects of ambient PM2.5 exposure on health using PM2.5 component measurements and regression calibration. J. Exposure Sci. Environ. Epidemiol. 16, 30–38. Zeger, S.L., et al., 2000. Exposure measurement error in time-series studies of air pollution: concepts and consequences. Environ. Health Perspect. 108, 419.
