PM(10) exposure, gaseous pollutants, and daily mortality in Inchon, South Korea.

To evaluate the relative importance of various measures of particulate and gaseous air pollution as predictors of daily mortality in Inchon, South Korea, the association between total daily mortality and air pollution was investigated for a 20-month period (January 1995 through August 1996). Poisson regression was used to regress daily death counts on each air pollutant, controlling for time trends, season, and meteorologic influences such as temperature and relative humidity. Regression coefficients of a 5-day moving average of particulate matter less than or = to 10 microm in aerodynamic diameter (PM(10)) on total mortality were positively significant when considered separately and simultaneously with other pollutants in the model. PM(10) remained significant when the models were confined to cardiovascular or respiratory mortality. Sulfur dioxide (SO(2)) and carbon monoxide (CO) were significantly related to respiratory mortality in the single-pollutant model. Ozone exposure was not statistically significant with regard to mortality in the above models, and graphic analysis showed that the relationship was nonlinear. A combined index of PM(10), nitrogen dioxide, SO(2), and CO seemed to better explain the exposure-response relationship with total mortality than an individual air pollutant. Pollutants should be considered together in the risk assessment of air pollution, as opposed to measuring the risk of individual pollutants. ImagesFigure 1Figure 2Figure 3Figure 4

The most acute air pollution episode in history occurred in London in 1952, when 4,000 excess deaths were attributed to air inversion. Other acute episodes have been reported in Donora, Pennsylvania, in 1948 and Los Angeles, California, in 1963. Since these episodes, investigations into the relationship between air pollution and mortality have demonstrated a positive association between air pollution levels and mortality, even at lower air pollution levels. A number of daily time-series studies have demonstrated an association between short-term exposure and increased mortality in air pollution areas when confounders were controlled (1,2). Daily mortality has been associated with total suspended particulates in Steubenville, Ohio (3); Philadelphia, Pennsylvania (4); and Mexico City, Mexico (5); and with particulate matter of < 10 pm (PM1O) in Utah Valley, Utah (2); Birmingham, Alabama (6); St. Louis, Missouri, and Kingston, New York (74; and Amsterdam, The Netherlands (8). Most published U.S. studies show that particulate matter is the pollutant which exhibits this relationship most dearly (2,3,(9)(10)(11)(12)(13), even though most of these studies have not considered multiple pollutants. The relationship between particulate air pollutants and mortality seems to continue well below current ambient air quality standards, and the exposure-response relationship is linear with no evidence of a threshold (14)(15)(16). Investigations have been carried out in Europe, although they yielded more diverse results in terms of the pollutant with the strongest associations. In most of the European studies there were significant relationships between mortality and either sulfur dioxide (SO2) or particulate matter (17)(18)(19)(20)(21)(22)(23)(24). Different areas of the world have different mixtures and levels of pollutants. Thus, it is unclear whether the same pollutants and mortality relationships found in the United States or Europe would apply to Inchon, an industrial city in South Korea. Particulate matter levels in Inchon are much higher than in most developed countries; however, SO2 levels have recently been decreasing rapidly. The combined effects of dust particles and other pollutants may play more important roles in the damaging process than dust partides alone. Combination indices of PMIO and gaseous pollutants can be useful tools to evaluate the relationship of pollutants with mortality; single or multiple pollutant models cannot analyze pollutant effects accurately when the pollutants are highly correlated with each other. Few studies have described the association between present levels of air pollution in East Asia and mortality. The objective of our study was to report on the association between daily mortality and ambient air pollution in Inchon, South Korea.

Materials and Methods
Daily mortality, weather, and air poliution data. Daily deaths for the Inchon area were read from the annual detail mortality tapes of the National Statistical Office (Taejon, Korea) for the period from 1 January 1995 to 31 August 1996. Cause-specific categories included in our analysis were total deaths, respiratory deaths, and cardiovascular deaths. The deaths due to accidents or violence were excluded from the total counts. Codes from the International Classification of Diseases, 10th revision (World Health Organization, Geneva) were used to define these categories.
The city of Inchon is located on the western side of the Korean peninsula, with a sea to the west and metropolitan Seoul to the east. It is the third largest city in the country. The 1996 census indicated that the population of Inchon was 2.4 million. Inchon has a four-season climate and lowlevel temperature inversions are common during the winter months. Data on 24-hr mean temperature and relative humidity were obtained from a centrally located Inchon weather station.
Air pollution data were obtained from the Department of the Environment (Seoul, Korea). Two monitoring sites for 24-hr measurements of PMIO and gaseous pollutants-SO2, nitrogen dioxide (NO2), ozone (03), and carbon monoxide (CO)-were established in residential areas. Concentrations of these pollutants were measured hourly, and 24-hr averages were constructed between measurement sites. For 0 a daytime 8-hr average was used instead o?a 24-hr average. Statistical analysis. The associations between daily mortality and air pollutants, as well as weather variables, were analyzed with a generalized additive model (GAM) estimating Poisson distribution. The relationship between pollution and mortality was complicated by the fact that periods with the highest pollution levels occurred during the winter months (with the exception of 03), and the higher incidence of death may be at least partially due to cold winter weather. Seasonal change or time trends also could confound the pollutant effects. Daily mortality was fit to the GAM, which included loess function of temperature, relative humidity, time trends, and indicator variables of the season. Once these models were established, the association of each air pollutant and multiple pollutants in predicting daily mortality was evaluated separately. Air pollutants may affect mortality with some time lags, and the appropriate averaging time for exposure may exceed 24 hr. Therefore, single-day exposure and multiple-day moving averages of up to 6 days were examined. To consider the correlation between the pollutants, and to evaluate the combined effects of pollutants, we introduced indices of combination of primary gaseous pollutants with PM1O using a 5-day moving average of pollutant levels.

03
-0.13** -0.17** -0.17** real environments, and thus better predict the mortality risk. To graphically analyze the dose-response relationships, the relative risks of mortality for pollutants and the combined indices were plotted using the GAM. Table 1 shows the distribution of mortality, meteorologic measurements, and air pollution between 1 January 1995 and 31 August 1996 for Inchon. An average of 22 persons died in the city each day. The Pearson correlations between the variables and total mortality are presented in Table 2. Correlations between mortality and primary pollutants were small but significantly positive: They ranged from 0.05 to 0.19. A negative correlation between the daily mortality and 03 was noticeable. There was also a negative correlation between the daily mortality and temperature or relative humidity. Correlations between exposure variables showed that there was a problem of collinearity among pollutants. PM1O was correlated significantly with other air pollutants, positively with NO2, SO2, and CO, and negatively with 03.

Results
A greater association with mortality was seen with the 5-day moving averages and the previous day's exposure than with other exposure variables. Tables 3-5 show the Poisson regression model, which indudes time trends, season, and weather variables for temperature and relative humidity. In the models that included a 5-day moving average of one pollutant or multiple pollutants, PM1O was a significant predictor of total mortality, whereas gaseous pollutants remained insignificant. An increase in the 5-day moving average of PM equal to 10 pg/m3, was associated with an increase in the relative risk of mortality to 0.8%. When the previous day's concentrations were considered, PM1O and NO2 were significant pollutants in the single-pollutant models. PM10 remained significant when the models were confined to cardiovascular or respiratory mortality. The association with PM1O, as measured by regression coefficients, was greatest for respiratory mortality. SO2 and CO were significantly related to respiratory mortality in the single-pollutant model using the previous day's concentrations. However, 03 was not statistically significant Table 3. Regression coefficients and standard errors when 5-day moving averages and the previous day's concentrations of one pollutant and multiple pollutants are considered in the Poisson regression for total mortal1ity.   when related to mortality except for the previous day's concentrations in the single-pollutant model oftotal mortality. The direction of the coefficient was negative in its relationship with total and cardiovascular mortality and positive with respiratory mortality. A loess function of 5-day moving averages of PM1O, NO2, SO2, and CO levels was used in the GAM to graphically analyze the dose-response relationship between pollutants and daily mortality. Figures 1 and 2 show the relative risks of total mortality by 5-day moving averages of pollutants and the combination indices of PM1O and primary gaseous pollutants. Excess mortality risk is dearly evident in the higher range of PM1O levels and the increase is dose responsive. When using combined indices of the pollutants, the exposure-response relationship for the index of overall pollutants was better than that of single pollutants or other combination indices. The index of PM10 + SO2 also showed a near-linear increase of relative risk, which suggests that the joint effect model of PM1O and S02 is a better predictor of mortality risk than effect models of individual pollutants. Contour analysis illustrated this joint effect well, showing that when concentrations of the two pollutants were high, the mortality rate peaked (Figure 3). Outputs from GAMs using a 5-day moving average and the previous day's concentrations were analyzed graphically to clarify the contradictory relationship between 03 and daily mortality (Figure 4). There was a clear change of direction in the relationship at approximately 23 ppb 03 concentration, suggesting that there is a threshold for the effects of 03 on mortality.

Discussion
This study documents a strong association between mortality and particulate air pollution in Inchon during the period of January 1995 to August 1996. This association was statistically significant in a model controlling other variables, and exhibited an exposure-response relationship. The increase in mortality risk with a 10-pg/m3 increase in PM1O was 0.8%. Results of most of the previous studies have suggested that a 10-pg/m3 increase in PM1O is associated with an increase in daily mortality equal to 0.5-1.5% (1). The mortality studies in Utah Valley (2) and Philadelphia (4) show that deaths due to respiratory disease were most strongly associated with particulate pollution levels, and statistical associations were also observed for deaths due to cardiovascular disease. In our study the highest regression coefficients were in the model for respiratory mortality. However, they did not reach statistical significance except when using multiple models with a 5-day moving average because of the low frequency of nonmalignant respiratory death. We also found PM1O to be significantly related to cardiovascular mortality, but the regression coefficients were lower than those for respiratory mortality.
The sources and levels of ambient air pollution, as well as population characteristics and habits, vary widely among the    United States, Europe, and Korea. However, the results of this study suggest that these differences do not exhibit a large influence on the relationship with mortality. It is highly unlikely that concordant results across so many locations could have occurred because of confounding or by chance. The highest particulate concentrations in Inchon occur during cold winter weather. There are substantial differences in the coincident weather patterns that accompany particulate exposure in different locations in the world. However, observed associations between particulate pollution and mortality are similar in many locations around the world. It is also unlikely that confounding weather effects are the source of the apparent pollution effects.
The most likely confounder that could be responsible for the effects observed is another pollutant or combination of pollutants which are highly correlated with particulate pollution (1). The high degree of correlation between air pollutants makes it difficult to analyze the single or multiple pollutant models appropriately. The fact that S02 and Volume 107, Number 1   NO2 are converted to sulfates and nitrates, thus contributing to the fine partides, makes the individual effects of the gaseous pollutants more difficult to explain (23). Therefore, indices of pollutants can more accurately represent the effects of complex air pollutants in spite of the different modes of action of the pollutants and can also be better predictors of mortality risk. When using combined indices of pollutants, the dose-response relationship of the index of overall pollutants is better than that of a single pollutant alone or of other combination indices. This index of overall pollutants can be useful to predict mortality risk and to manage air pollution policies. Our analysis shows no specific particulate pollution threshold. Mortality risk increased with PMIO levels in an exposure-response fashion. However, there seems to be a threshold for mortality for 03. The negative direction of the 0 relationship in the Poisson regression model was due to low 03 concentrations; therefore, results from higher 03 areas may differ.
The mechanism of the relationship between daily mortality and particulate air pollution is not obvious. Some studies have provided evidence that PMIO has free radical activity and causes lung inflammation (24,25). Heavy metals, transition metals, and polycyclic aromatic hydrocarbons in particulate pollutants may also cause harmful effects (26,27). However, specific mechanisms remain unclear. This epidemiologic study cannot explain the toxicologic mechanism for particulate pollution-induced mortality. However, the relationship of the combined indices suggests that there is a mutual interaction between partides and gaseous pollutants. This study is limited by its use of environmental monitoring data alone to represent ambient concentrations, which do not necessarily represent individual exposures. This can result in nondirectional misclassification of exposure and bias toward the null. Therefore, it is likely that our positive results underestimate the magnitude of the association. Some studies have found that PM 1 levels are highly correlated between indoor and outdoor air (28). Therefore, outdoor measurements of pollutants can also represent indoor environments, and thus be used as a proxy in the analysis.
In summary, the association of particulate air pollution with daily mortality in Inchon is similar in magnitude to the results in other communities in the United States and in Europe. These associations are still observed when weather variables and other pollutants are accounted for. The association between particulate air pollution and mortality was largest for respiratory deaths. Increased cardiovascular mortality associated with particulate air pollution was also observed. Further, there is no evidence to support a threshold for particulate air pollution. However, there appears to be an 03 threshold level for adverse health effects and mortality. Additional studies will be needed to confirm this threshold in communities with varying concentrations of 03. A combined index of overall pollutants, a variable more accurately reflecting real environments, can be a better predictor of acute mortality. Future studies should be directed at evaluating the mutual interaction of pollutants and their chemical composition.