- Email: [email protected]
A vulnerability study of the low-income elderly in the context of high temperature and mortality in Seoul, Korea
Science of the Total Environment 371 (2006) 82 – 88 www.elsevier.com/locate/scitotenv
A vulnerability study of the low-income elderly in the context of high temperature and mortality in Seoul, Korea Youngmin Kim a,⁎, Seunghun Joh b,1 a
Graduate School of Environmental Studies Seoul National University, San 56-1, Shillim-dong, Kwanak-Ku, Seoul 151-742, South Korea b Environmental Institute for Life and Peace, 128-11 Hyojadong Deogyanggu Goyangshi Kyonggido, 412-140, South Korea Received 24 March 2006; received in revised form 21 July 2006; accepted 9 August 2006 Available online 26 September 2006
Abstract Introduction: We investigated the impact of environmental high temperature on mortality in Seoul, Korea, and the consequences of high temperature-induced mortality with a focus on the low-income elderly. Methods: Changes in the risk of death by age and income were estimated by a 1 °C increase in temperature using a generalized additive model adjusting for non-temperature related factors: time trends, seasonality, and air pollution. The study covered the years of 2000, 2001, and 2002. Results: We found that income and age were potential factors in high-temperature-induced excess mortality. Evidences to support these results are as follows: first, regarding the effect of an economic factor in the association between mortality and high temperature, the study shows that the mortality rate of the low-income group is higher, by as much as 1.3- to 1.7-fold, than that of the general population. Second, taking age into consideration, the mortality of low-income elderly people is 1.5-fold higher than that of the whole low-income group. The combined effect of income and age on mortality is estimated as 2.3-fold higher than that of the general population. But the results of the low-income and elderly group were not statistically significant due to wide standard deviation. Conclusions: The relationship between high-temperature-induced excess mortality, income, and age suggests the need for a public health message, yet many results were not statistically significant: preventive and health care interventions need to be administered to the elderly and low-income group during periods of high temperature. © 2006 Elsevier B.V. All rights reserved. Keywords: High temperature; Mortality; Low-income group; Elderly population; Seoul
1. Introduction There have been many studies of socioeconomic status in the context of health, especially mortality and ⁎ Corresponding author. Tel.: +80 2 880 8829; fax: +80 2 883 8620. E-mail addresses: [email protected] (Y. Kim), [email protected] (S. Joh). 1 Tel.: +80 2 358 7657; fax: +80 2 887 6905. 0048-9697/$ - see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.scitotenv.2006.08.014
morbidity. These studies have shown an inverse relationship of socioeconomic factors on mortality (Huisman et al., 2004; Fukuda et al., 1973). Low-income groups appear to be highly vulnerable to the physical environment; for instance, high temperature and natural disaster. The greatest burden of health risks is very often borne by the disadvantaged in our societies. The vast majority of threats to health are found more commonly among poor people, in people with
Y. Kim, S. Joh / Science of the Total Environment 371 (2006) 82–88
83
2. Materials and methods
total population. There were 34,185 people aged 65 years or over among the beneficiaries. From the total mortality of the beneficiaries, deaths attributable to external causes such as traffic accidents, suicide, poisoning, and homicide ([ICD-10] codes V01–Y89) were excluded, since they were not temperature-related. The complete data from 13 districts were available from January 2000 to December 2002. The daily death counts for the general population (non-specific population as a control group) were provided by the Korean National Statistical Office. These records include date, place, cause of death, and demographic factors such as age and sex. Deaths attributable to external causes ([ICD-10] codes V01–Y89) were excluded from the total mortality of the general population. The final mortality series was structured to examine summer-only mortality/high-temperature associations so that days before May 1 or after September 30 and the days over 26.8 °C, 27.9 °C, 29.2 °C, 30.5 °C, and 31.7 °C which are the values counted for 75%, 80%, 85%, 90%, 95% of the daily maximum temperature, respectively, were assigned for modeling process. Daily meteorological parameters, including daily maximum, mean and minimum temperature and relative humidity were obtained from the Korea Meteorological Administration. Air pollution data were obtained from the National Institute of Environmental Research, which monitors the ambient air pollution at 27 sites in Seoul. Daily average concentrations of PM10, SO2, CO, and NO2 from each monitoring site were calculated, and the means of these daily averages were used for Seoul's concentration values. For ozone, the means of the daily maximum concentration from each monitoring site were used as Seoul's concentration value.
2.1. Data source
2.2. Statistical methods
We use two types of data: records of death for the lowincome group and the meteorological records of temperature and air pollution for the Seoul area. The low-income group in this study is defined by the Korean government as Beneficiaries of the National Basic Livelihood Security System. The Korean government sets minimum living costs every year, according to which eligibility for government assistance is determined. The monthly minimum living cost, for instance in 2001, was set at US$330 for a single household, US$550 for a two-person household, US$760 for a three-person household, and US$960 for four-person household, respectively. In Seoul, the beneficiaries in 2001 amount to 78,724 households with 154,601 people, which is 1.6% of the
We estimated a high temperature–mortality association by using generalized additive models (GAMs) with non-parametric smoothing functions to describe nonlinear relations. This analytical approach is one of the standard time-series methods that have been developed for air pollution studies. The approach takes into account any smooth variations in the denominator and rates over time (Gouveia et al., 2003). Possible confounding factors that might influence the relationship between temperature and death counts, such as long-term trend fluctuation, seasonality, relative humidity, day-of-week effect, and air pollutants were controlled. Locally weighted regression (LOESS) was used as a smoothing function. We used Akaike's information criterion (AIC), to guide the determination of the
little formal education, and those with lowly occupations. These risks cluster and they accumulate over time (WHO, 2002). Many studies have reported that days of low and high ambient temperature are associated with increases in mortality. Episode analysis and time-series analysis have been used to determine the acute effects of very hot weather on populations (Gouveia et al., 2003). Many of the studies examining the relationship between temperature and mortality have not, however, addressed the importance of income-relevant factors. Naughton et al. (1999) reported that air-conditioners are the strongest protective factor against heat-related death, which means that low-income groups can be very vulnerable to the effects of high temperature. Poverty is a risk factor for heat-related illnesses and deaths because the poor are more likely to live in heat-island urban areas (Lee, 1980) and are less likely to be able to afford airconditioning (Jones et al., 1980; Greenberg et al., 1950; Patz et al., 2000; Curriero et al., 2002; Diaz et al., 1986). This study was undertaken to investigate low-income conditions in the context of high temperature-induced mortality in Seoul, the capital of the Republic of Korea. The analytic framework of this study is based on intermittent stressor–vulnerability–health consequence, which has been described by Galea et al. (2005). In the framework, the intermittent stressor is equivalent to high temperature. The vulnerability indicates the level of ability to adapt to high temperature; in this case the variables are income and age. The health consequence is pertinent to mortality of a low-income elderly class exposed to high temperature.
84
Y. Kim, S. Joh / Science of the Total Environment 371 (2006) 82–88
Table 1 Statistics for temperature, humidity, death counts, and air pollutant concentrations during study period (Jan. 1 2000–Dec.31 2002)
Daily mean temperature (°C) Daily maximum temperature (°C) Daily minimum temperature (°C) Humidity (%) Daily death counts for beneficiaries (N ) Daily death counts for general population (N ) O3 (ppb) CO(ppm) NO2 (ppb) PM10 (μg/m3) SO2 (ppb)
Min.
25th a
75th b
Max.
Mean
Total N c
S.D.
− 15.5 − 12.4 − 18.6 26.1 0 30.0 2.8 0.23 7.9 14.2 2.2
3.9 8.1 0.1 51.9 3 49.0 21.7 0.6 26.4 43.9 3.5
22.1 26.7 18.5 72.2 6 61.0 44.8 1.0 43.7 74.1 6.7
30.4 35.3 25.8 96.0 13.0 92.0 127.8 2.4 87.6 723.1 14.9
12.8 17.3 8.9 62.2 4.5 55.4 35.0 0.8 35.8 68.2 5.3
1096 1096 1096 1096 1096 1096 1096 1096 1096 1096 1096
10.4 10.7 10.4 13.9 2.1 8.9 18.5 3.3 12.0 51.7 2.2
S.D.: standard deviation. a The value counts for 25%. b The value counts for 75%. c Total days of the data.
smoothness of the confounder (Rainham et al., 2003). The model specifications we used are as follows: lnðEðY ÞÞ ¼ a þ bðXt Þ þ RSi ðXi Þ þ gDðdowÞ
ð1Þ
where E(Y ) is the expected daily death counts; β is the coefficients (slop) for high temperature (Xt); Si(Xi) are the smooth function of date, relative humidity and air pollutants, and D(dow) are indicator variables for days of the week. As the regression coefficient is β, relative risk (RR) and 95% confidence interval (95% CI) are as follows: RR ¼ expðbÞ
ð2Þ
95% CI ¼ expðbF1:96 standard errorÞ
ð3Þ
Here, RR indicates change rate for expected death due to a 1 °C increase in temperature. SAS and S-plus were utilized in the analysis for managing and fitting the data to GAMs. It has been recently noted elsewhere that using the default function parameters, especially when fitting models with nonparametric terms, can underestimate the variance of fitted model parameters, leading to type 1 error (Ramsey et al., 2003). Related research has also shown that use of the default convergence criteria in SPlus can result in biased fitted linear parameters (Dominici et al., 2002). To guard against these problems related to concurvity (the nonparametric analogue of multicollinearity) and bias in the regression estimates, we adopted more stringent convergence. After creating the base model, we fitted two groups (general population and low-income group) then
compared RR of the low-income group with the general population. In order to analyze delayed effects of daily maximum temperature, lagged predictor variables were employed: unlagged (LAG0) and lagged by 1–3 days (LAG1, LAG2, and LAG3). We also compared a model setting air pollutants as confounding factors with a model without air pollutants to assess modification of the effect of thermal stress. 3. Results and discussion 3.1. Statistical results The statistics for temperature, humidity, the concentrations of air pollutants, and daily death counts for the study period (Jan. 2000–Dec. 2002) are given in Table 1. Table 2 Statistics for daily death rates of elderly group (from May 2000 to September 2002) Daily maximum temperature (°C) Min. 12.0 25.0 25th a Median 27.7 75th b 30.4 Maximum 35.3 Mean 27.4 Total N c 459 S.D. 3.87 S.D.: standard deviation. a The value counts for 25%. b The value counts for 75%. c Total days of the data.
Beneficiary group (N )
General population (N )
0.0 1.0 2.0 3.0 8.0 2.3 459 1.54
16.0 27.0 32.6 35.0 53.0 31.6 459 6.23
Y. Kim, S. Joh / Science of the Total Environment 371 (2006) 82–88 Table 3 The relative risks (RR) due to 1 °C temperature increase RR of general 95% CI population Over 26.8 °C Over 27.9 °C Over 29.2 °C Over 30.5 °C Over 31.7 °C
RR of beneficiary group
95% CI
1.011 (LAG0) 1.001–1.020 1.018 (LAG2) 0.990–1.047 1.026 (LAG0) 1.015–1.038 1.033 (LAG2) 0.996–1.072 1.019 (LAG0) 1.004–1.036 1.033 (LAG2) 0.983–1.085 1.027 (LAG0) 1.005–1.051 1.047 (LAG3) 0.985–1.114 1.025 (LAG0) 0.982–1.071 1.042 (LAG3) 0.934–1.163
LAG0, LAG1, LAG2, and LAG3 indicate 0, 1, 2 and 3 days of lag effects, respectively. CI: confidence interval.
During the study period, the average daily maximum temperature is 17.3 °C and the average relative humidity is 62.2%. The daily death counts for the general population in 13 districts are 55.4 and the daily death counts for beneficiaries are 4.5. The daily death counts of people over 65 years during May 1 through September 30 (2000–2002) are given in Table 2. The mean of the daily death counts for the elderly from the beneficiary group is 2.3 and for the elderly in the general population the value is 31.6. 3.2. Relative risks The excess mortality rates for total population at a temperature over 26.8 °C, 27.9 °C, 29.2 °C, 30.5 °C, and 31.7 °C were calculated. The results are given in Table 3. The excess mortality rates for the general population are 1.1% (95% CI: 0.1–2.0%), 2.6% (95% CI: 1.5–3.8%), 1.9% (95% CI: 0.4–3.6%), 2.7% (95% CI: 0.5–5.1%), and 2.5% (95% CI: − 1.8–7.1%), respectively. The excess mortality rates for the beneficiary group are 1.8% (95% CI: − 1.0–4.7%), 3.3% (95% CI: − 0.4–7.2%), 3.3% (95% CI: − 1.7–8.5%), 4.7% (95% CI: − 1.5– 11.4%), and 4.2% (95% CI: − 6.6–16.3%), respectively. In all cases for the total population, the excess mortality rates for the beneficiary group have about 1.3- to 1.7-fold higher than those for the general population. As a minimum value of 1.3-fold, in Table 3, the excess mortality rate for the general population is 2.6%/°C (95% CI: 1.5–3.8%) and for the beneficiary group it is 3.3%/°C (95% CI: − 0.4–7.2%) in the case of temperatures over 27.9 °C. The maximum value of 1.7-fold is based on the excess mortality rate for temperatures over
85
30.5 °C. For the general population the excess mortality rate is 2.7%/°C (95% CI: 0.5–5.1%) and for the beneficiary group it is 4.7%/°C(95% CI: − 1.5– 11.4%). In both groups, the excess mortality was highest at temperatures over 30.5 °C. In many relative risks for total population, the standard errors include 1.00. Due to the low death count, especially, for all those in the beneficiary group which included 1.00, the results are not statistically significant. Limited data were applied to the over 65 years age group, from May 2000 to September 2002 to investigate the relationship between the daily maximum temperature and changes of excess mortality. The excess mortality rates for the elderly from the beneficiary group has about 2.3-fold higher than those from the general population. As shown in Table 4, the excess mortality rate for the elderly from the general population is 0.4%/°C (95% CI: − 0.1–0.9%) and for the beneficiary group it is 0.9%/°C (95% CI: − 0.7–2.5%). In all cases of relative risks for elderly group, the results are not statistically significant. The lag effects of 0–3 days were applied to each model. No lag effect was found for the general population, while a 2-day lag effect is clear for the beneficiary group (Tables 3 and 4). 3.3. The difference between the general population and low-income group By comparing the excess mortality rates due to a 1 °C increase of daily maximum temperature, it was shown that the excess mortality rates of the beneficiaries are 1.3-fold higher at temperatures over 27.9 °C, 1.7-fold higher at temperatures over 29.2 °C and 30.5 °C than that of the general population (Table 3). The differences in excess mortality due to 1 °C increase between the two groups tend to get broader as the temperature increases, as shown in Fig. 1.
Table 4 Relative risk of 1 °C increase for the elders from general population and the elders from the beneficiary group, from May 2000 to September 2002
LAG0 LAG1 LAG2 LAG3
The elders from general population (95% CI)
The elders from the beneficiary group (95% CI)
1.004 (0.999–1.009) 0.999 (0.994–1.005) 0.999 (0.995–1.004) 0.994 (0.990–0.999)
0.992 (0.974–1.010) 0.996 (0.981–1.011) 1.009 (0.993–1.025) 1.007 (0.991–1.023)
LAG0, LAG1, LAG2, and LAG3 indicate lag effects of 0, 1, 2, and 3 days, respectively. CI: confidence interval.
86
Y. Kim, S. Joh / Science of the Total Environment 371 (2006) 82–88
sistent with other research using same method (Gouveia et al., 2003; Rainham et al., 2003). The literature on the association between health and weather according to income is scarce. However, there are several social factors that contribute to potential increased health risk from lower individual and household income (Carty et al., 2002).
Fig. 1. Percent change in mortality due to 1 °C increase in temperature.
Specific to age, the elderly group within the beneficiaries shows 1.5-fold greater excess death rates than the whole beneficiary group. Excess mortality rate (LAG2) for the whole beneficiary group is 0.6% (95% CI: −0.6–1.8%) as shown in Table 5. For people over 65 years of age within the same group it is 0.9% (95% CI: −0.7–2.5%) as shown in Table 4. In the elderly group from the beneficiaries(the lowincome elderly group), the excess mortality due to a 1 °C increase is 2.3-fold higher than that of the elderly group of the general population. Excess mortality rate (LAG0) for people over 65 years of age in the general population is 0.4% (95% CI: − 0.1–0.9%) and for those over 65 years of age in the beneficiary group it is 0.9% (95% CI: − 0.7–2.5%) as shown in Table 5. 4. Discussion This study found that high temperature was associated with daily mortality in Seoul and the relationship could be higher in low-income group, yet our results were not statistically significant in many cases. For an increase of 1 °C in temperature we observed between 1.1% (above 26.8 °C) and 2.7% (above 30.5 °C) increases in mortality of whole the general population. These results are con-
• residences tend to be crowded and constructed with poor quality materials; • residences tend to be located in close proximity to industrial areas; • insufficient and inadequate clothing • lower access to climate control (e.g. air conditioning to cope with high temperatures) due to cost, etc.; and, • incomplete nutritious diets and often higher exposures to smoking Our results reflect the effect of these potential factors; the adverse relationship between mortality due to high temperature and income. Regarding the economic factor in the association between mortality and high temperature, our study shows that the mortality rate due to temperature of the low-income group is higher than that of the general population by as much as 1.3- to 1.7-fold. Especially elders (N65 years old) from low-income group were observed very vulnerable to high temperature. Considering age, the mortality of low-income elderly is 1.5-fold higher than that of the whole lowincome group. The combined effects of income and age are estimated as a 2.3-fold increase in mortality compared with that of the general population. Many epidemiological studies have shown the association between daily mortality and air pollution. The studies for Seoul has also showed that sulfur dioxide, PM10, and ozone are associated with the increase of
Table 5 Relative risk of 1 °C increase for general population and the beneficiary group, from May 2000 to September 2002
LAG0 LAG1 LAG2 LAG3
General population (95% CI)
Beneficiary group (95% CI)
1.004 (1.000–1.008) 1.000 (0.996–1.003) 1.002 (0.998–1.005) 0.997 (0.993–1.000)
1.002 (0.989–1.015) 0.993 (0.982–1.004) 1.006 (0.994–1.018) 1.004 (0.992–1.016)
LAG0, LAG1, LAG2, and LAG3 indicate lag effects of 0, 1, 2, and 3 days, respectively. CI: confidence interval.
Fig. 2. Relative risk of 1 °C increase in temperature for general population.
Y. Kim, S. Joh / Science of the Total Environment 371 (2006) 82–88
Fig. 3. Relative risk of 1 °C increase in temperature for beneficiary group.
daily mortality (Lee et al., 1999; Kim et al., 2004). We compared the relative risk rates with control for air pollutants with those without control. The excess mortality for the general population does not reveal significant differences, as illustrated in Fig. 2, whether or not adjustment is made for air pollutants. In contrast, the control for air pollutants reveals a clear distinction for the beneficiary group. The excess mortality rates for 26.8 °C, 27.9 °C, and 29.2 °C show no significant difference regardless of the control for air pollutants. However, at temperatures over 30 °C, the mortality without control is significantly higher than that with control, as shown in Fig. 3. The result implies that air pollution has an influence on the excess mortality of the beneficiaries at temperatures over 30 °C as a confounding factor. It indicates indirectly that the low-income group is more vulnerable to the effects of air pollution at high temperature than the general population. Therefore, it is necessary to control air pollutants to avoid overestimation of the effect of high temperature in the low-income group, in particular. Previous studies have found weak lag effect in summer, suggesting that direct exposure to heat may be responsible for a proportion of heat-related deaths (Kunst et al., 1993; Ballester et al., 1997; Whitman and Good, 1995; El-Zein et al., 2004). For the general population no lag effect was found, indicating very acute effect of mortality due to high temperature. For the low-income group, on the other hand, 2 or 3 days' lag effect was clear. We could not find the cause of this difference. Further studies (e.g. study distinguishing cause of death) need to be made to investigate the reason. Our study examined excess mortality due to high temperature with only one variable: daily maximum temperatures. Daily mean temperature and minimum
87
temperature showed negative correlation with daily death counts, whereas maximum temperature showed positive. When the minimum temperature and mortality from May 2000 to September 2002 were fitted to our model, the parameters of GLM also were negative. Unfortunately, we did not perform a sex-specific study to compare the differences for each age group (e.g. children and adults) due to a low daily death count for the beneficiary group. The reason for the limited amount of data is that the government of Seoul has not managed the beneficiaries systematically for a very long time. Once these data have started to accumulate in the future, further studies will be possible. Generally, the standard errors can be set around the risk ratio, and if the 95% confidence intervals exclude the value of 1.0, there is a statistically significant difference between the risks (Jekel et al., 2001). In this study, however, the standard error includes 1.00 in many cases, which is speculated due to the low death count in the beneficiary group and elderly group. This might suggest that the relative risk, i.e. the regression coefficient, is not statistically significant. However, this study did not consider this point when comparing the general population and the beneficiary group. This is a critical shortcoming of our study. Our results suggest that low income condition and old age have potential effects on high-temperature-induced mortality. Low-income and the elderly group could be vulnerable to high temperature, reflecting inequalities in health impacts related to climate change. Preventive measures and health care interventions, therefore, need to be administered to the elderly and low-income group during periods of high temperature and more detailed analysis must be undertaken. Acknowledgements The primary impetus for this study stemmed from a work of master's thesis for Youngmin Kim at Graduate School of Environmental Studies (GSES), Seoul National University. We would like to thank Profs. Jungwk Kim and Jeongjeon Rhee at GSES for the thesis committee, Prof. Jongtae Lee at Hanyang University for helpful information, Dr. Sunyoung Kim and Boeun Lee, for help with data analysis. References Ballester F, Corella D, Perez-Hoyoz S, Saez M, Hervas A. Mortality as a function of temperature: a study in Valencia, Spain, 1991–1993. Int J Epidemiol 1997;26(3):551–61.
88
Y. Kim, S. Joh / Science of the Total Environment 371 (2006) 82–88
Carty P, Crabbé P, Craig L, Krewski D, Franklin M, King M, et al. Health effects of climate change and health co-benefits resulting from potential Kyoto-driven policies: a Canadian perspective. Health Canada in support of the Expert Panel Workshop on Climate Change and Health and Well-being, April 15–16, 2002, Ottawa, Canada, 2002; 2002. 13 pp. Curriero FC, Heiner KS, Samet JM, Zeger SL, Strug L, Patz JA. Temperature and mortality in 11 cities of the eastern United States. Am J Epidemiol 2002;155(1):80–7. Diaz J, Garcia R, Castro V, Hernandez E, Lopez C, Otero A. Effects of extremely hot days on people older than 65 years in Seville (Spain) from 1986 to 1997. Int J Biometeorol 2002;46:145–9. Dominici F, McDermott A, Zeger SL, Samet JM. On the use of generalized additive models in time-series studies of air pollution and health. Am J Epidemiol 2002;156:193–203. El-Zein A, Tewtel-Salemb M, Nehmec G. A time-series analysis of mortality and air temperature in Greater Beirut. Sci Total Environ 2004;330:71–80. Fukuda Y, Nakamura K, Takano T. Municipal socioeconomic status and mortality in Japan: sex and age differences, and trends in 1973–1998. Soc Sci Med 2004;59:2435–45. Galea S, Ahern J, Karpati A. A model of underlying socioeconomic vulnerability in human populations: evidence from variability in population health and implications for public health. Soc Sci Med 2005;60:2417–30. Gouveia N, Hajat S, Armstrong B. Scioeconomic differentials in the temperature–mortality relationship in São Paulo, Brazil. Int J Epidemiol 2003;32:390–7. Greenberg JH, Broomberg J, Reed CM, Gustafson TL, Beauchamp RA. The epidemiology of heat-related deaths, Texas 1950, 1970–79, and 1980. Am J Public Health 1983;73:805–7. Huisman M, Kunst AE, Mackenbach JP. Socioeconomic inequalities in morbidity among the elderly: a European overview. Soc Sci Med 2004;57:861–73.
Jekel JF, Katz DL, Elmore JG. Epidemiology, biostatistics, and preventive medicine. Philadelphia: W.B. Saunders Company; 2001. 184 pp. Jones TS, Liang AP, Kilbourne EM, Griffin MR, Patriarca PA, Wassilak SG, et al. Morbidity and mortality associated with the July 1980 heat wave in St. Louis and Kansas City. JAMA 1982;247:3327–31. Kim H, Lee JT, Hong YC, Yi SM, Kim Y. Evaluating the effect of daily PM10 variation on mortality. Inhal Toxicol 2004;16(Suppl. 1):1–4. Kunst AE, Looman CWN, Mackenbach JP. Outdoor air temperature and mortality in the Netherlands: a time-series analysis. Am J Epidemiol 1993;137(3):331–41. Lee DH. Seventy-five years of searching for a heat index. Environ Res 1980;22:331–56. Lee JT, Shin DC, Chung Y. Air pollution and daily mortality in Seoul and Ulsan, Korea. Environ Health Perspect 1999;107:149–54. Naughton MP, Henderson A, Mirabelli AC, Kaiser R, Wilhelm JL, Kieszak SM, et al. Heat-related mortality during a 1999 heat wave in Chicago. Am J Prev Med 2002;22:221–7. Patz JA, Mcgeehin MA, Bernard SM, Ebi KL, Epstein PR, Grambsch A, et al. The potential health impacts of climate variability and change for the United States: executive summary of the report of the health sector of the U.S. National Assessment. Environ Health Perspect 2000;108(4). Rainham DGC, Smoyer-Tomic KE. The role of air pollution in the relationship between a heat stress index and human mortality in Toronto. Environ Res 2003;93:9-19. Ramsey T, Burnett R, Krewski D. Underestimation of standard errors in generalized additive models linking mortality to ambient particulate matter. Epidemiology 2003;14:18–23. Whitman S, Good G. Mortality in Chicago attributed to the July 1995 heat wave. Am J Public Health 1997;87:1515–9. WHO. The world health report 2002: reducing risk and promoting healthy life; 2002.
A vulnerability study of the low-income elderly in the context of high temperature and mortality in Seoul, Korea Youngmin Kim a,⁎, Seunghun Joh b,1 a
Graduate School of Environmental Studies Seoul National University, San 56-1, Shillim-dong, Kwanak-Ku, Seoul 151-742, South Korea b Environmental Institute for Life and Peace, 128-11 Hyojadong Deogyanggu Goyangshi Kyonggido, 412-140, South Korea Received 24 March 2006; received in revised form 21 July 2006; accepted 9 August 2006 Available online 26 September 2006
Abstract Introduction: We investigated the impact of environmental high temperature on mortality in Seoul, Korea, and the consequences of high temperature-induced mortality with a focus on the low-income elderly. Methods: Changes in the risk of death by age and income were estimated by a 1 °C increase in temperature using a generalized additive model adjusting for non-temperature related factors: time trends, seasonality, and air pollution. The study covered the years of 2000, 2001, and 2002. Results: We found that income and age were potential factors in high-temperature-induced excess mortality. Evidences to support these results are as follows: first, regarding the effect of an economic factor in the association between mortality and high temperature, the study shows that the mortality rate of the low-income group is higher, by as much as 1.3- to 1.7-fold, than that of the general population. Second, taking age into consideration, the mortality of low-income elderly people is 1.5-fold higher than that of the whole low-income group. The combined effect of income and age on mortality is estimated as 2.3-fold higher than that of the general population. But the results of the low-income and elderly group were not statistically significant due to wide standard deviation. Conclusions: The relationship between high-temperature-induced excess mortality, income, and age suggests the need for a public health message, yet many results were not statistically significant: preventive and health care interventions need to be administered to the elderly and low-income group during periods of high temperature. © 2006 Elsevier B.V. All rights reserved. Keywords: High temperature; Mortality; Low-income group; Elderly population; Seoul
1. Introduction There have been many studies of socioeconomic status in the context of health, especially mortality and ⁎ Corresponding author. Tel.: +80 2 880 8829; fax: +80 2 883 8620. E-mail addresses: [email protected] (Y. Kim), [email protected] (S. Joh). 1 Tel.: +80 2 358 7657; fax: +80 2 887 6905. 0048-9697/$ - see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.scitotenv.2006.08.014
morbidity. These studies have shown an inverse relationship of socioeconomic factors on mortality (Huisman et al., 2004; Fukuda et al., 1973). Low-income groups appear to be highly vulnerable to the physical environment; for instance, high temperature and natural disaster. The greatest burden of health risks is very often borne by the disadvantaged in our societies. The vast majority of threats to health are found more commonly among poor people, in people with
Y. Kim, S. Joh / Science of the Total Environment 371 (2006) 82–88
83
2. Materials and methods
total population. There were 34,185 people aged 65 years or over among the beneficiaries. From the total mortality of the beneficiaries, deaths attributable to external causes such as traffic accidents, suicide, poisoning, and homicide ([ICD-10] codes V01–Y89) were excluded, since they were not temperature-related. The complete data from 13 districts were available from January 2000 to December 2002. The daily death counts for the general population (non-specific population as a control group) were provided by the Korean National Statistical Office. These records include date, place, cause of death, and demographic factors such as age and sex. Deaths attributable to external causes ([ICD-10] codes V01–Y89) were excluded from the total mortality of the general population. The final mortality series was structured to examine summer-only mortality/high-temperature associations so that days before May 1 or after September 30 and the days over 26.8 °C, 27.9 °C, 29.2 °C, 30.5 °C, and 31.7 °C which are the values counted for 75%, 80%, 85%, 90%, 95% of the daily maximum temperature, respectively, were assigned for modeling process. Daily meteorological parameters, including daily maximum, mean and minimum temperature and relative humidity were obtained from the Korea Meteorological Administration. Air pollution data were obtained from the National Institute of Environmental Research, which monitors the ambient air pollution at 27 sites in Seoul. Daily average concentrations of PM10, SO2, CO, and NO2 from each monitoring site were calculated, and the means of these daily averages were used for Seoul's concentration values. For ozone, the means of the daily maximum concentration from each monitoring site were used as Seoul's concentration value.
2.1. Data source
2.2. Statistical methods
We use two types of data: records of death for the lowincome group and the meteorological records of temperature and air pollution for the Seoul area. The low-income group in this study is defined by the Korean government as Beneficiaries of the National Basic Livelihood Security System. The Korean government sets minimum living costs every year, according to which eligibility for government assistance is determined. The monthly minimum living cost, for instance in 2001, was set at US$330 for a single household, US$550 for a two-person household, US$760 for a three-person household, and US$960 for four-person household, respectively. In Seoul, the beneficiaries in 2001 amount to 78,724 households with 154,601 people, which is 1.6% of the
We estimated a high temperature–mortality association by using generalized additive models (GAMs) with non-parametric smoothing functions to describe nonlinear relations. This analytical approach is one of the standard time-series methods that have been developed for air pollution studies. The approach takes into account any smooth variations in the denominator and rates over time (Gouveia et al., 2003). Possible confounding factors that might influence the relationship between temperature and death counts, such as long-term trend fluctuation, seasonality, relative humidity, day-of-week effect, and air pollutants were controlled. Locally weighted regression (LOESS) was used as a smoothing function. We used Akaike's information criterion (AIC), to guide the determination of the
little formal education, and those with lowly occupations. These risks cluster and they accumulate over time (WHO, 2002). Many studies have reported that days of low and high ambient temperature are associated with increases in mortality. Episode analysis and time-series analysis have been used to determine the acute effects of very hot weather on populations (Gouveia et al., 2003). Many of the studies examining the relationship between temperature and mortality have not, however, addressed the importance of income-relevant factors. Naughton et al. (1999) reported that air-conditioners are the strongest protective factor against heat-related death, which means that low-income groups can be very vulnerable to the effects of high temperature. Poverty is a risk factor for heat-related illnesses and deaths because the poor are more likely to live in heat-island urban areas (Lee, 1980) and are less likely to be able to afford airconditioning (Jones et al., 1980; Greenberg et al., 1950; Patz et al., 2000; Curriero et al., 2002; Diaz et al., 1986). This study was undertaken to investigate low-income conditions in the context of high temperature-induced mortality in Seoul, the capital of the Republic of Korea. The analytic framework of this study is based on intermittent stressor–vulnerability–health consequence, which has been described by Galea et al. (2005). In the framework, the intermittent stressor is equivalent to high temperature. The vulnerability indicates the level of ability to adapt to high temperature; in this case the variables are income and age. The health consequence is pertinent to mortality of a low-income elderly class exposed to high temperature.
84
Y. Kim, S. Joh / Science of the Total Environment 371 (2006) 82–88
Table 1 Statistics for temperature, humidity, death counts, and air pollutant concentrations during study period (Jan. 1 2000–Dec.31 2002)
Daily mean temperature (°C) Daily maximum temperature (°C) Daily minimum temperature (°C) Humidity (%) Daily death counts for beneficiaries (N ) Daily death counts for general population (N ) O3 (ppb) CO(ppm) NO2 (ppb) PM10 (μg/m3) SO2 (ppb)
Min.
25th a
75th b
Max.
Mean
Total N c
S.D.
− 15.5 − 12.4 − 18.6 26.1 0 30.0 2.8 0.23 7.9 14.2 2.2
3.9 8.1 0.1 51.9 3 49.0 21.7 0.6 26.4 43.9 3.5
22.1 26.7 18.5 72.2 6 61.0 44.8 1.0 43.7 74.1 6.7
30.4 35.3 25.8 96.0 13.0 92.0 127.8 2.4 87.6 723.1 14.9
12.8 17.3 8.9 62.2 4.5 55.4 35.0 0.8 35.8 68.2 5.3
1096 1096 1096 1096 1096 1096 1096 1096 1096 1096 1096
10.4 10.7 10.4 13.9 2.1 8.9 18.5 3.3 12.0 51.7 2.2
S.D.: standard deviation. a The value counts for 25%. b The value counts for 75%. c Total days of the data.
smoothness of the confounder (Rainham et al., 2003). The model specifications we used are as follows: lnðEðY ÞÞ ¼ a þ bðXt Þ þ RSi ðXi Þ þ gDðdowÞ
ð1Þ
where E(Y ) is the expected daily death counts; β is the coefficients (slop) for high temperature (Xt); Si(Xi) are the smooth function of date, relative humidity and air pollutants, and D(dow) are indicator variables for days of the week. As the regression coefficient is β, relative risk (RR) and 95% confidence interval (95% CI) are as follows: RR ¼ expðbÞ
ð2Þ
95% CI ¼ expðbF1:96 standard errorÞ
ð3Þ
Here, RR indicates change rate for expected death due to a 1 °C increase in temperature. SAS and S-plus were utilized in the analysis for managing and fitting the data to GAMs. It has been recently noted elsewhere that using the default function parameters, especially when fitting models with nonparametric terms, can underestimate the variance of fitted model parameters, leading to type 1 error (Ramsey et al., 2003). Related research has also shown that use of the default convergence criteria in SPlus can result in biased fitted linear parameters (Dominici et al., 2002). To guard against these problems related to concurvity (the nonparametric analogue of multicollinearity) and bias in the regression estimates, we adopted more stringent convergence. After creating the base model, we fitted two groups (general population and low-income group) then
compared RR of the low-income group with the general population. In order to analyze delayed effects of daily maximum temperature, lagged predictor variables were employed: unlagged (LAG0) and lagged by 1–3 days (LAG1, LAG2, and LAG3). We also compared a model setting air pollutants as confounding factors with a model without air pollutants to assess modification of the effect of thermal stress. 3. Results and discussion 3.1. Statistical results The statistics for temperature, humidity, the concentrations of air pollutants, and daily death counts for the study period (Jan. 2000–Dec. 2002) are given in Table 1. Table 2 Statistics for daily death rates of elderly group (from May 2000 to September 2002) Daily maximum temperature (°C) Min. 12.0 25.0 25th a Median 27.7 75th b 30.4 Maximum 35.3 Mean 27.4 Total N c 459 S.D. 3.87 S.D.: standard deviation. a The value counts for 25%. b The value counts for 75%. c Total days of the data.
Beneficiary group (N )
General population (N )
0.0 1.0 2.0 3.0 8.0 2.3 459 1.54
16.0 27.0 32.6 35.0 53.0 31.6 459 6.23
Y. Kim, S. Joh / Science of the Total Environment 371 (2006) 82–88 Table 3 The relative risks (RR) due to 1 °C temperature increase RR of general 95% CI population Over 26.8 °C Over 27.9 °C Over 29.2 °C Over 30.5 °C Over 31.7 °C
RR of beneficiary group
95% CI
1.011 (LAG0) 1.001–1.020 1.018 (LAG2) 0.990–1.047 1.026 (LAG0) 1.015–1.038 1.033 (LAG2) 0.996–1.072 1.019 (LAG0) 1.004–1.036 1.033 (LAG2) 0.983–1.085 1.027 (LAG0) 1.005–1.051 1.047 (LAG3) 0.985–1.114 1.025 (LAG0) 0.982–1.071 1.042 (LAG3) 0.934–1.163
LAG0, LAG1, LAG2, and LAG3 indicate 0, 1, 2 and 3 days of lag effects, respectively. CI: confidence interval.
During the study period, the average daily maximum temperature is 17.3 °C and the average relative humidity is 62.2%. The daily death counts for the general population in 13 districts are 55.4 and the daily death counts for beneficiaries are 4.5. The daily death counts of people over 65 years during May 1 through September 30 (2000–2002) are given in Table 2. The mean of the daily death counts for the elderly from the beneficiary group is 2.3 and for the elderly in the general population the value is 31.6. 3.2. Relative risks The excess mortality rates for total population at a temperature over 26.8 °C, 27.9 °C, 29.2 °C, 30.5 °C, and 31.7 °C were calculated. The results are given in Table 3. The excess mortality rates for the general population are 1.1% (95% CI: 0.1–2.0%), 2.6% (95% CI: 1.5–3.8%), 1.9% (95% CI: 0.4–3.6%), 2.7% (95% CI: 0.5–5.1%), and 2.5% (95% CI: − 1.8–7.1%), respectively. The excess mortality rates for the beneficiary group are 1.8% (95% CI: − 1.0–4.7%), 3.3% (95% CI: − 0.4–7.2%), 3.3% (95% CI: − 1.7–8.5%), 4.7% (95% CI: − 1.5– 11.4%), and 4.2% (95% CI: − 6.6–16.3%), respectively. In all cases for the total population, the excess mortality rates for the beneficiary group have about 1.3- to 1.7-fold higher than those for the general population. As a minimum value of 1.3-fold, in Table 3, the excess mortality rate for the general population is 2.6%/°C (95% CI: 1.5–3.8%) and for the beneficiary group it is 3.3%/°C (95% CI: − 0.4–7.2%) in the case of temperatures over 27.9 °C. The maximum value of 1.7-fold is based on the excess mortality rate for temperatures over
85
30.5 °C. For the general population the excess mortality rate is 2.7%/°C (95% CI: 0.5–5.1%) and for the beneficiary group it is 4.7%/°C(95% CI: − 1.5– 11.4%). In both groups, the excess mortality was highest at temperatures over 30.5 °C. In many relative risks for total population, the standard errors include 1.00. Due to the low death count, especially, for all those in the beneficiary group which included 1.00, the results are not statistically significant. Limited data were applied to the over 65 years age group, from May 2000 to September 2002 to investigate the relationship between the daily maximum temperature and changes of excess mortality. The excess mortality rates for the elderly from the beneficiary group has about 2.3-fold higher than those from the general population. As shown in Table 4, the excess mortality rate for the elderly from the general population is 0.4%/°C (95% CI: − 0.1–0.9%) and for the beneficiary group it is 0.9%/°C (95% CI: − 0.7–2.5%). In all cases of relative risks for elderly group, the results are not statistically significant. The lag effects of 0–3 days were applied to each model. No lag effect was found for the general population, while a 2-day lag effect is clear for the beneficiary group (Tables 3 and 4). 3.3. The difference between the general population and low-income group By comparing the excess mortality rates due to a 1 °C increase of daily maximum temperature, it was shown that the excess mortality rates of the beneficiaries are 1.3-fold higher at temperatures over 27.9 °C, 1.7-fold higher at temperatures over 29.2 °C and 30.5 °C than that of the general population (Table 3). The differences in excess mortality due to 1 °C increase between the two groups tend to get broader as the temperature increases, as shown in Fig. 1.
Table 4 Relative risk of 1 °C increase for the elders from general population and the elders from the beneficiary group, from May 2000 to September 2002
LAG0 LAG1 LAG2 LAG3
The elders from general population (95% CI)
The elders from the beneficiary group (95% CI)
1.004 (0.999–1.009) 0.999 (0.994–1.005) 0.999 (0.995–1.004) 0.994 (0.990–0.999)
0.992 (0.974–1.010) 0.996 (0.981–1.011) 1.009 (0.993–1.025) 1.007 (0.991–1.023)
LAG0, LAG1, LAG2, and LAG3 indicate lag effects of 0, 1, 2, and 3 days, respectively. CI: confidence interval.
86
Y. Kim, S. Joh / Science of the Total Environment 371 (2006) 82–88
sistent with other research using same method (Gouveia et al., 2003; Rainham et al., 2003). The literature on the association between health and weather according to income is scarce. However, there are several social factors that contribute to potential increased health risk from lower individual and household income (Carty et al., 2002).
Fig. 1. Percent change in mortality due to 1 °C increase in temperature.
Specific to age, the elderly group within the beneficiaries shows 1.5-fold greater excess death rates than the whole beneficiary group. Excess mortality rate (LAG2) for the whole beneficiary group is 0.6% (95% CI: −0.6–1.8%) as shown in Table 5. For people over 65 years of age within the same group it is 0.9% (95% CI: −0.7–2.5%) as shown in Table 4. In the elderly group from the beneficiaries(the lowincome elderly group), the excess mortality due to a 1 °C increase is 2.3-fold higher than that of the elderly group of the general population. Excess mortality rate (LAG0) for people over 65 years of age in the general population is 0.4% (95% CI: − 0.1–0.9%) and for those over 65 years of age in the beneficiary group it is 0.9% (95% CI: − 0.7–2.5%) as shown in Table 5. 4. Discussion This study found that high temperature was associated with daily mortality in Seoul and the relationship could be higher in low-income group, yet our results were not statistically significant in many cases. For an increase of 1 °C in temperature we observed between 1.1% (above 26.8 °C) and 2.7% (above 30.5 °C) increases in mortality of whole the general population. These results are con-
• residences tend to be crowded and constructed with poor quality materials; • residences tend to be located in close proximity to industrial areas; • insufficient and inadequate clothing • lower access to climate control (e.g. air conditioning to cope with high temperatures) due to cost, etc.; and, • incomplete nutritious diets and often higher exposures to smoking Our results reflect the effect of these potential factors; the adverse relationship between mortality due to high temperature and income. Regarding the economic factor in the association between mortality and high temperature, our study shows that the mortality rate due to temperature of the low-income group is higher than that of the general population by as much as 1.3- to 1.7-fold. Especially elders (N65 years old) from low-income group were observed very vulnerable to high temperature. Considering age, the mortality of low-income elderly is 1.5-fold higher than that of the whole lowincome group. The combined effects of income and age are estimated as a 2.3-fold increase in mortality compared with that of the general population. Many epidemiological studies have shown the association between daily mortality and air pollution. The studies for Seoul has also showed that sulfur dioxide, PM10, and ozone are associated with the increase of
Table 5 Relative risk of 1 °C increase for general population and the beneficiary group, from May 2000 to September 2002
LAG0 LAG1 LAG2 LAG3
General population (95% CI)
Beneficiary group (95% CI)
1.004 (1.000–1.008) 1.000 (0.996–1.003) 1.002 (0.998–1.005) 0.997 (0.993–1.000)
1.002 (0.989–1.015) 0.993 (0.982–1.004) 1.006 (0.994–1.018) 1.004 (0.992–1.016)
LAG0, LAG1, LAG2, and LAG3 indicate lag effects of 0, 1, 2, and 3 days, respectively. CI: confidence interval.
Fig. 2. Relative risk of 1 °C increase in temperature for general population.
Y. Kim, S. Joh / Science of the Total Environment 371 (2006) 82–88
Fig. 3. Relative risk of 1 °C increase in temperature for beneficiary group.
daily mortality (Lee et al., 1999; Kim et al., 2004). We compared the relative risk rates with control for air pollutants with those without control. The excess mortality for the general population does not reveal significant differences, as illustrated in Fig. 2, whether or not adjustment is made for air pollutants. In contrast, the control for air pollutants reveals a clear distinction for the beneficiary group. The excess mortality rates for 26.8 °C, 27.9 °C, and 29.2 °C show no significant difference regardless of the control for air pollutants. However, at temperatures over 30 °C, the mortality without control is significantly higher than that with control, as shown in Fig. 3. The result implies that air pollution has an influence on the excess mortality of the beneficiaries at temperatures over 30 °C as a confounding factor. It indicates indirectly that the low-income group is more vulnerable to the effects of air pollution at high temperature than the general population. Therefore, it is necessary to control air pollutants to avoid overestimation of the effect of high temperature in the low-income group, in particular. Previous studies have found weak lag effect in summer, suggesting that direct exposure to heat may be responsible for a proportion of heat-related deaths (Kunst et al., 1993; Ballester et al., 1997; Whitman and Good, 1995; El-Zein et al., 2004). For the general population no lag effect was found, indicating very acute effect of mortality due to high temperature. For the low-income group, on the other hand, 2 or 3 days' lag effect was clear. We could not find the cause of this difference. Further studies (e.g. study distinguishing cause of death) need to be made to investigate the reason. Our study examined excess mortality due to high temperature with only one variable: daily maximum temperatures. Daily mean temperature and minimum
87
temperature showed negative correlation with daily death counts, whereas maximum temperature showed positive. When the minimum temperature and mortality from May 2000 to September 2002 were fitted to our model, the parameters of GLM also were negative. Unfortunately, we did not perform a sex-specific study to compare the differences for each age group (e.g. children and adults) due to a low daily death count for the beneficiary group. The reason for the limited amount of data is that the government of Seoul has not managed the beneficiaries systematically for a very long time. Once these data have started to accumulate in the future, further studies will be possible. Generally, the standard errors can be set around the risk ratio, and if the 95% confidence intervals exclude the value of 1.0, there is a statistically significant difference between the risks (Jekel et al., 2001). In this study, however, the standard error includes 1.00 in many cases, which is speculated due to the low death count in the beneficiary group and elderly group. This might suggest that the relative risk, i.e. the regression coefficient, is not statistically significant. However, this study did not consider this point when comparing the general population and the beneficiary group. This is a critical shortcoming of our study. Our results suggest that low income condition and old age have potential effects on high-temperature-induced mortality. Low-income and the elderly group could be vulnerable to high temperature, reflecting inequalities in health impacts related to climate change. Preventive measures and health care interventions, therefore, need to be administered to the elderly and low-income group during periods of high temperature and more detailed analysis must be undertaken. Acknowledgements The primary impetus for this study stemmed from a work of master's thesis for Youngmin Kim at Graduate School of Environmental Studies (GSES), Seoul National University. We would like to thank Profs. Jungwk Kim and Jeongjeon Rhee at GSES for the thesis committee, Prof. Jongtae Lee at Hanyang University for helpful information, Dr. Sunyoung Kim and Boeun Lee, for help with data analysis. References Ballester F, Corella D, Perez-Hoyoz S, Saez M, Hervas A. Mortality as a function of temperature: a study in Valencia, Spain, 1991–1993. Int J Epidemiol 1997;26(3):551–61.
88
Y. Kim, S. Joh / Science of the Total Environment 371 (2006) 82–88
Carty P, Crabbé P, Craig L, Krewski D, Franklin M, King M, et al. Health effects of climate change and health co-benefits resulting from potential Kyoto-driven policies: a Canadian perspective. Health Canada in support of the Expert Panel Workshop on Climate Change and Health and Well-being, April 15–16, 2002, Ottawa, Canada, 2002; 2002. 13 pp. Curriero FC, Heiner KS, Samet JM, Zeger SL, Strug L, Patz JA. Temperature and mortality in 11 cities of the eastern United States. Am J Epidemiol 2002;155(1):80–7. Diaz J, Garcia R, Castro V, Hernandez E, Lopez C, Otero A. Effects of extremely hot days on people older than 65 years in Seville (Spain) from 1986 to 1997. Int J Biometeorol 2002;46:145–9. Dominici F, McDermott A, Zeger SL, Samet JM. On the use of generalized additive models in time-series studies of air pollution and health. Am J Epidemiol 2002;156:193–203. El-Zein A, Tewtel-Salemb M, Nehmec G. A time-series analysis of mortality and air temperature in Greater Beirut. Sci Total Environ 2004;330:71–80. Fukuda Y, Nakamura K, Takano T. Municipal socioeconomic status and mortality in Japan: sex and age differences, and trends in 1973–1998. Soc Sci Med 2004;59:2435–45. Galea S, Ahern J, Karpati A. A model of underlying socioeconomic vulnerability in human populations: evidence from variability in population health and implications for public health. Soc Sci Med 2005;60:2417–30. Gouveia N, Hajat S, Armstrong B. Scioeconomic differentials in the temperature–mortality relationship in São Paulo, Brazil. Int J Epidemiol 2003;32:390–7. Greenberg JH, Broomberg J, Reed CM, Gustafson TL, Beauchamp RA. The epidemiology of heat-related deaths, Texas 1950, 1970–79, and 1980. Am J Public Health 1983;73:805–7. Huisman M, Kunst AE, Mackenbach JP. Socioeconomic inequalities in morbidity among the elderly: a European overview. Soc Sci Med 2004;57:861–73.
Jekel JF, Katz DL, Elmore JG. Epidemiology, biostatistics, and preventive medicine. Philadelphia: W.B. Saunders Company; 2001. 184 pp. Jones TS, Liang AP, Kilbourne EM, Griffin MR, Patriarca PA, Wassilak SG, et al. Morbidity and mortality associated with the July 1980 heat wave in St. Louis and Kansas City. JAMA 1982;247:3327–31. Kim H, Lee JT, Hong YC, Yi SM, Kim Y. Evaluating the effect of daily PM10 variation on mortality. Inhal Toxicol 2004;16(Suppl. 1):1–4. Kunst AE, Looman CWN, Mackenbach JP. Outdoor air temperature and mortality in the Netherlands: a time-series analysis. Am J Epidemiol 1993;137(3):331–41. Lee DH. Seventy-five years of searching for a heat index. Environ Res 1980;22:331–56. Lee JT, Shin DC, Chung Y. Air pollution and daily mortality in Seoul and Ulsan, Korea. Environ Health Perspect 1999;107:149–54. Naughton MP, Henderson A, Mirabelli AC, Kaiser R, Wilhelm JL, Kieszak SM, et al. Heat-related mortality during a 1999 heat wave in Chicago. Am J Prev Med 2002;22:221–7. Patz JA, Mcgeehin MA, Bernard SM, Ebi KL, Epstein PR, Grambsch A, et al. The potential health impacts of climate variability and change for the United States: executive summary of the report of the health sector of the U.S. National Assessment. Environ Health Perspect 2000;108(4). Rainham DGC, Smoyer-Tomic KE. The role of air pollution in the relationship between a heat stress index and human mortality in Toronto. Environ Res 2003;93:9-19. Ramsey T, Burnett R, Krewski D. Underestimation of standard errors in generalized additive models linking mortality to ambient particulate matter. Epidemiology 2003;14:18–23. Whitman S, Good G. Mortality in Chicago attributed to the July 1995 heat wave. Am J Public Health 1997;87:1515–9. WHO. The world health report 2002: reducing risk and promoting healthy life; 2002.