- Email: [email protected]
Relationship between winter temperature and mortality in Seoul, South Korea, from 1994 to 2006
S CI EN CE OF T H E T OTAL EN V I RO N M EN T 4 0 7 ( 2 0 09 ) 21 5 8–2 16 4
a v a i l a b l e a t w w w. s c i e n c e d i r e c t . c o m
w w w. e l s e v i e r. c o m / l o c a t e / s c i t o t e n v
Relationship between winter temperature and mortality in Seoul, South Korea, from 1994 to 2006 Jongsik Haa , Joungho Yoona , Ho Kimb,⁎ a
Korea Environment Institute, 613-2 Bulgwang-dong Eunpyeong-gu, Seoul 122-706, South Korea Department of Biostatistics and Epidemiology, School of Public Health and the Institute of Health and Environment, Seoul National University, 28 Yunkeon-dong Chongro-gu Seoul 110-799, South Korea b
AR TIC LE D ATA
ABSTR ACT
Article history:
In South Korea, the mortality increases with temperature above a certain threshold during
Received 4 July 2008
the warm season. In contrast, despite the common burden of cold weather, little is known
Received in revised form
about the effects of low temperatures on mortality. We evaluated the relationship between
26 November 2008
low temperatures and mortality in winter (December–February) in Seoul, South Korea, from
Accepted 5 December 2008
1994 to 2006. Data were obtained from government agencies. After adjusting for trends in
Available online 20 January 2009
time, day of the week, holidays, and relative humidity, we explored the associations between mortality and various cold indicators of winter in Seoul, South Korea. First, we
Keywords:
fitted nonparametric smoothing regression models to check the shape of associations and
Cold wave
then fitted threshold models (including two different slopes in a model) to estimate the
Low-temperature mortality
thresholds and the effects of low temperatures using the Akaike Information Criterion (AIC).
Threshold
The graphical associations between cardiorespiratory, cardiovascular, and all causes of mortality and the cold wave index (CWI = Tmin, previous day − Tmin, current day) were observed. We confirmed the threshold according to a lag structure and after that, estimated the effects of CWI below the threshold, respectively. The effects of the daytime CWI lagged by 0–2 days were the strongest among those of the daytime CWI lagged by 0–6 days. The most significant mortality outcomes were cardiovascular-related. Although we could not consider respiratory-related mortality, the effect of CWI on cardiovascular-related mortality below a certain threshold was greater than cardiorespiratory-related or all cause-related mortality. In addition, the association between mortality and CWI was more immediate and vulnerable in an elderly subgroup (≥ 65 years old) than in a younger subgroup (0–64 years old). Our results suggest that public health programs should be considered to alleviate not only the effect of sudden change in winter temperature on mortality. © 2008 Elsevier B.V. All rights reserved.
1.
Introduction
It is now widely accepted in the scientific community that the Earth's climate is changing as a result of human activities. Climate change causes various environmental and ecological changes, and some of these changes will affect human health. In 2007, the Intergovernmental Panel on Climate Change (IPCC) reported that climate change is likely to lead to more
intense and frequent extreme weather events and that human exposure to such changing weather patterns could result in increased death, disability, and suffering (IPCC, 2007). The dangers of hot weather were demonstrated dramatically by the 2003 heat wave that accounted for more than 30,000 deaths throughout Western Europe and more than 2000 deaths in England and Wales alone (Johnson et al., 2005; Kosatsky, 2005). Detrimental effects of cold weather on health
⁎ Corresponding author. Tel.: +82 2 740 8874; fax: +82 2 745 9104. E-mail address: [email protected] (H. Kim). 0048-9697/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.scitotenv.2008.12.029
S CI EN C E OF TH E T OTAL EN V I RO N M EN T 4 0 7 ( 2 0 09 ) 21 5 8–2 16 4
have also been observed. These have been assessed using a variety of health endpoints, since the early 20th century (Russell, 1928; Holland et al., 1961; Loudon, 1964; Pattenden et al., 2003). In South Korea, the mortality increases with temperature above a certain threshold during the warm season (Kim et al., 2006). In contrast, despite the common burden of exposure to cold weather, little is known about the links between low temperatures and mortality. This has resulted in an apparent lack of concern by government agencies to proactively reduce risk in South Korea. Mortality rates for cardiovascular and respiratory disease typically exhibit distinct seasonal variation, with the highest rates occurring in the winter months (McKee, 1989; Eng and Mercer, 1998; Bowie and Jackson, 2002). The main factor driving this seasonal pattern is thought to be temperature (Basu and Samet, 2002). Braga et al. (2001) reported that low temperatures are strongly associated with increased cardiorespiratory-caused deaths. Keatinge et al. (1997) reported that Mortality increased to a greater extent with given fall of temperature in regions with warm winters. We evaluated the relationship between low temperatures and mortality during winter (December–February) in Seoul, South Korea, from 1994 to 2006. We used the time-series generalized additive model (GAM) and included nonparametric smoothing functions of a cold wave index (CWI) for our initial analysis (Basu and Samet, 2002). We then fit the data to time-series piecewise log-linear models, with one slope representing values below the CWI threshold and one representing those above the threshold (Armstrong, 2006).
2.
Materials and methods
2.1.
Study scope
This study focused on the relation between cold weather and mortality in Seoul, South Korea. We examined the effect of sudden change in temperature on winter mortality, particularly for the elderly. We postulated that delay and threshold between exposure and response might be heterogeneous among population groups. Therefore, we performed the analysis for the entire population, as well as for two age subgroups: 0–64 and ≥ 65 years old. For each group, we performed separate model-fitting analyses according to lags of 0, 1, and 2 days after the cold indicator to compare the magnitude of the intensity of influence by age group.
2.2.
Weather and mortality data
Weather data, including measurements of relative humidity taken every 3 h and hourly measurements of ambient temperature, were provided by the Korean Meteorological Administration. We began with substantial preliminary analyses of GAMs with natural cubic splines in which several types of representative values served as daily cold indicators, e.g., the daily CWI, minimum temperature, and mean temperature. We selected the CWI as the indicator based on the best graphical association by visual assessment in the model fittings. The
2159
CWI was defined as CWI = T_mp − T_mc, where T_mc is the minimum temperature of the current day, and T_mp is the minimum temperature of the previous day. For example, a CWI of 5 °C indicates that the current day's minimum temperature is 5 °C lower than the previous day's minimum temperature. In general, a cold wave is defined as an unusually large and rapid drop in temperature over a short period of time, e.g., 24 h. Therefore, the CWI is not the absolute cold, but rather the relative cold in human-perceived equivalent temperature (i.e., how cold it feels, termed the “felt air temperature by a rapid fall in temperature within a 24 hour period”). The Korean Meteorological Administration calculates the criterion for announcing a cold wave warning using the CWI formula (KMA, 2008). The Korean National Statistical Office provided the mortality data. These data excluded deaths of individuals in the study area who did not reside within the area, as well as deaths of individuals caused by accidents (ICD10 codes V00-Y99). In this study, we considered the following deaths: all causes of mortality (ICD10 codes A00-U99), cardiorespiratory-related mortality (ICD10 codes I00–I99, J00–J99), and cardiovascular-related mortality (ICD10 codes I00–I99). We excluded respiratory-related mortality because of limited daily counts (mean = 6.23, SD = 3.05 in winter in Seoul from 1994 to 2006). The relationship between the CWI and mortality may be confounded by trends in time, day of the week, holidays, and humidity. For controlling trends in time, we used indicator variables such that the first winter was January and February 1994, the second winter was December 1994 and January and February 1995, up to the last winter, which was December 2006. In addition, we included smoothing function of date on our model to control the possible confounding effect of time. For controlling day of the week and holidays, we used indicator variables to categorize days into holidays, including Sunday; the day after a holiday; Saturday; and other days. In the end, we controlled daily mean humidity using a natural cubic spline of four degrees of freedom.
2.3.
Modeling approach
The statistical analysis was conducted in two steps. First, to investigate the shape of the CWI–mortality curve, we fitted GAMs with natural cubic splines to the CWI data, using four degrees of freedom as a compromise between providing adequate control for unmeasured confounders and leaving sufficient information from which to estimate the CWI effect. Second, if the shape of the CWI–mortality curve was a reverse J shape (i.e., a negative association with CWI below the threshold and no association or a small positive association above the threshold), the following piecewise log-linear model was fit to the data: logEðY Þ = b0 + S1 ðx1 Þ + N + Sp xp + b1 ðCWI nÞ + + b2 CWI; ð1Þ
where x1, x2, …, xp are covariates, and (CWI − ξ)+ = min{CWI − ξ,0}. If the CWI is less than the threshold value (ξ): logEðYÞ = b0 b2 n + S1 ðx1 Þ + N + Sp xp + ðb1 + b2 Þ CWI:
ð2Þ
2160
S CI EN CE OF T H E T OTAL EN V I RO N M EN T 4 0 7 ( 2 0 09 ) 21 5 8–2 16 4
If the CWI is greater than the threshold value (ξ):
logEðY Þ = b0 + S1 ðx1 Þ + N + Sp xp + b2 CWI:
ð3Þ
These two lines are constrained to meet at the threshold value (ξ). The model is also known as a B-mode splined linear regression model (Akaike, 1973; Armstrong, 2006; Kim et al., 2004). We identified the threshold values that produced the model with the lowest AIC. However, when the relationship was linear, we fitted the data using the following linear model: logEðY Þ = b0 + S1 ðx1 Þ + N + Sp xp + b1 CWI
ð4Þ
where x1, x2, …, xp are covariates. The Eq. (1) does not assign any death to CWI levels above the threshold whereas the Eq. (4) does. So, the RR (negative association) may be higher in the Eq. (1). If threshold effect of CWI exists in the mortality risk level, CWI risk estimates based on an assumption of linearity may represent an underestimation of the true risk. Relative risk (RR) was calculated using exp(β), where β is the related model coefficient. The percentage change was derived from RR using the formula (RR − 1) × 100. All analyses were performed using SPlus v2000 (Mathsoft Inc., Cambridge, MA). The convergence tolerances of the GAM functions were set to 10− 9, with a limit of 1000 iterations to avoid biased estimates of regression coefficient and standard error (Dominici et al., 2002; Pattenden et al., 2003).
3.
Results
3.1.
Description
In the time-series of the daily mean temperature and daily CWI from 1994 to 2006, there is a clear yearly pattern in CWI, with more fluctuations occurring during the winter months (Fig. 1). The standard deviation of the CWI was 3.21 °C during
the cold season (December–February), 1.41 °C during the hot season (June–August), and 2.45 °C during the other seasons (Table 1). The average daily mortality due to all causes was 12, 5, and 5% greater in the cold season than in the hot season, other seasons, and year-round, respectively. A similar relationship was observed for cardiorespiratory- and cardiovascular-related deaths.
3.2.
Graphical results
We plotted the estimated CWI–mortality functions derived from the GAM with the CWI in the smoothing function (Fig. 2). The graphical analysis using 1-day lagged CWI resulted in a reverse J-shaped curve for every population group for cardiovascular-related mortality, but only for the 0–64 year old age group for cardiorespiratory-related mortality. The strength of the graphical relationship seemed to depend on the cause of death; it was greater for cardiovascular-related than cardiorespiratory-related mortality, which were both greater than for all causes of mortality.
3.3.
Effects and thresholds
Once the threshold was determined in the CWI–mortality associations, we fitted piece-wise log-linear models to estimate the threshold and the slope for the range of CWI below the threshold. For cardiorespiratory-related mortality, a 1 °C decrease in the 1-day lagged CWI below −2.9 °C was significantly associated with increased mortality, and the effect of a CWI below the threshold was a 2.379% (95% confidence interval [CI]: 0.374–4.343%) increase in the 0–64 year old age group (Table 2). A 1 °C decrease in the 2-day lagged CWI was linearly associated and the effect was a 0.754% (95% CI: 0.261–1.299%) increase in the 0–64 year old age group. For cardiovascular-related mortality, the thresholds in the 1-day lagged CWI were −2.8 °C, − 3.3 °C, and − 0.2 °C, and the effects of the CWI were a 1.799% (95% CI: 0.631–2.950%), 2.938% (95% CI: 0.557–5.262%), and 0.839% (95% CI: 0.089–1.582%) increase in all
Fig. 1 – Daily mean temperature and daily cold wave index in Seoul from 1994 to 2006.
2161
S CI EN C E OF TH E T OTAL EN V I RO N M EN T 4 0 7 ( 2 0 09 ) 21 5 8–2 16 4
Table 1 – Characteristics of the population and weather in Seoul, South Korea, from 1994 to 2006 Winter (N = 1173) (December– February)
Daily mortality counts
All cause
Cardiorespiratory (I00–I99, J00–J99) Cardiovascular (I00–I99) Daily cold wave index (°C) Daily mean temperature (°C) Daily mean relative humidity (%)
All ages ≥ 65 0–64 All ages ≥ 65 0–64 All ages ≥ 65 0–64
Summer (N = 1196) ( June–August)
Other seasons (N = 2379)
Year-round (N = 4748)
Mean
SD
Mean
SD
Mean
SD
Mean
SD
102.38 62.34 40.04 33.81 24.23 9.58 27.57 19.23 8.34 0.01 −0.18 56.84
12.35 10.08 7.25 7.03 5.79 3.48 5.85 4.78 3.11 3.21 4.31 13.36
91.35 54.31 37.04 28.14 20.12 8.02 23.33 16.30 7.03 −0.06 24.60 73.24
11.39 9.28 6.66 6.24 5.34 2.93 5.56 4.79 2.64 1.41 2.81 11.52
97.83 58.78 39.05 31.36 22.37 8.99 26.04 18.12 7.92 0.02 13.52 60.30
11.27 9.38 6.80 6.19 5.38 3.26 5.58 4.77 3.02 2.45 6.44 13.96
97.32 58.53 38.79 31.16 22.27 8.89 25.73 17.94 7.79 0.00 12.93 62.70
12.22 9.95 6.97 6.72 5.67 3.28 5.85 4.89 2.99 2.46 10.21 14.64
age groups, the 0–64 year old age group, and the ≥65 year old age group, respectively (Table 2). In addition, in the ≥65 year old age group, the threshold in the 0–day lagged CWI was −3.4 °C, and the effect was a 1.593% (95% CI: 0.108–3.055%) increase. A 1 °C decrease in the 2-day lagged CWI was linearly associated, and the effect was a 0.711% (95% CI: 0.130–1.288%) increase in the 0–64 year old age group.
4.
Discussion
Our study was designed to estimate the RR of the CWI on mortality by cause of death and age group. For the fitting with 0–2 days lagged CWI for cardiovascular-related death, we found that (1) model with 0–1 day lags was better than models
Fig. 2 – Shape of the cold wave index–mortality curve in winter.
2162
S CI EN CE OF T H E T OTAL EN V I RO N M EN T 4 0 7 ( 2 0 09 ) 21 5 8–2 16 4
Table 2 – Estimated threshold and percent increases Cause of death
Cardiorespiratory (I00–I99, J00–J99)
Age group (years)
All ages
≥ 65
0–64
Cardiovascular (I00–I99)
All ages
≥ 65
0–64
Cold wave index lag in days
0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2
Graphical association
None Threshold None Threshold None None None Threshold Linear Threshold Threshold None Threshold Threshold None None Threshold Linear
Threshold (°C)
Estimated percent increase in daily mortality in 1 °C temperature decreases below threshold
Observed
Sampled 95% CI
Percent increase
95% CI
NA − 1.5 NA − 2.3 NA NA NA − 2.9 NA − 3.2 − 2.8 NA − 3.4 − 0.2 NA NA − 3.3 NA
– – – – – – – – – – (− 3.25 to −2.12) – (−5.04 to − 2.99) (−3.53–1.25) – – (−4.63 to − 2.98) –
NA 0.793 NA 0.812 NA NA NA 2.379 0.754 0.541 1.799 NA 1.593 0.839 NA NA 2.938 0.711
NA (− 0.018–1.598) NA (− 0.369–1.979) NA NA NA (0.374–4.343) (0.261–1.299) (− 0.845–1.909) (0.631–2.950) NA (0.108–3.055) (0.089–1.582) NA NA (0.557–5.262) (0.130–1.288)
Note 1: The threshold is the temperature at which the risk of mortality begins to increase as the cold wave index decreases. Note 2: NA means ‘Not Available’. Note 3: The sampled 95% CI of threshold was estimated for each exposure–response curve having significance effect below a certain threshold in cardiovascular-related mortality by re-sampling technique. In the re-sampling technique, the same method to identify the threshold was repeated 1000 times by leaving 10% subject out each time, and then the empirical distribution of the threshold was identified. We calculated the sampled 95% confidence interval by using the mean and standard deviation of the distributions.
with any other lag for the ≥65 year old age group; (2) model with 1–2 day lags was better than models with any other lag for the 0–64 year old age group. For a 1-day lagged CWI for cardiovascular-related death, thresholds were −0.2 °C, −3.3 °C and −2.8 °C, for ≥65 year old age group, 0–64 year old age group, and all ages group, respectively. However, the degree of increase in RR below the threshold for the 0–64 year old age group (2.938%, 95% CI: 0.557–5.262%) was larger than those for the ≥65 year old age group (0.839%, 95% CI: 0.089–1.582%) and for the all ages group (1.799%, 95% CI: 0.631–2.950%). These results could be explained by the following two points. First, the strongest association for the 0–64 year old age group tended to be a 1-day lag, whereas that of the ≥65 year old age group was a 0-day lagged CWI, which might indicate that the current day's exposure has a more immediate effect than that of the previous day on the elderly. Second, the estimated threshold for the ≥65 year old age group with a 1-day lagged CWI (−0.2 °C vs. −3.3 °C for 0–64 year old age group) indicates that the elderly may be more vulnerable to a lower CWI than young people. On the other hand, the estimated percentage increases for cardiovascular-related death below the threshold of a 1-day lagged CWI were 2.938% for the 0–64 year old age group and 0.839% for the ≥65 year old age group. Because we separately estimated the effect of CWI on mortality below different thresholds for each age group, the slopes of CWI below the threshold can not be compared with any other slope without the threshold. The re-sampling technique was employed to assess the stability of the threshold for each exposure–response curve
having significance effect below a certain threshold in cardiovascular-related mortality. In the re-sampling technique, the same method to identify the threshold was repeated 1000 times by leaving 10% subject out each time. In Fig. 2, the sampled 95% confidence intervals from the simulations show the degree of stability. Although the sampled 95% CI for the ≥65 year old age group with a 1-day lagged CWI was larger than that of any other groups, all of them included the observed thresholds. The re-sampling result implies that the threshold is fairly stable. The results using cause-of-death data showed the same pattern, with cardiovascular-related death stronger than that of cardiorespiratory-related death, and statistically significant for each age group and lagged CWI. So, even though we did not conduct an analysis for respiratory-caused death, we may deduce that people with cardiovascular disease are more vulnerable to cold weather than those with respiratory disease. The effects of cold temperatures may occur over prolonged periods of time, in some cases up to a few weeks (Keatinge and Donaldson, 2001; Carder et al., 2005). However, the CWI that we used represented the current day's minimum temperature fluctuation relative to the previous day's minimum temperature. Therefore, our results could differ from studies that use only the temperature of the current day as the cold indicator. Many studies have suggested that particulate air pollution affects the daily mortality rate. For example, studies on the effects of air pollution on the daily death rate in Seoul, South Korea, in the 1990s report a 1.3–3.7% increase in the rate of all causes of mortality for an inter-quartile increase in the concentration of particulate Matter 10 micrometers or less
S CI EN C E OF TH E T OTAL EN V I RO N M EN T 4 0 7 ( 2 0 09 ) 21 5 8–2 16 4
(PM10) (Lee and Schwartz, 1999; Lee et al., 1999a,b). In spite of that, Braga et al. (2001) did not control for PM10 in temperature and mortality study. We could not consider the PM10 effect due to limited data period. In Korea, PM10 data is obtainable from 1997. So, if there is a substantial confounding or modifying effect, an inappropriately specified model may result in bias. We considered the city of Seoul; thus, our results are limited to this city only. Keatinge et al. (1997) reported that mortality increases to a greater extent for a given decrease in temperature in regions that have warm winters. Our results should not be generalized to other regions. It is necessary to determine whether the application of our methods to other regions would produce similar results. In Our study, December, January, and February were defined as winter. South Korea has four distinct seasons. Summer is hot and humid, and winter is cold and dry. In fact, the cold period from December to February is the coldest in South Korea and has largest fluctuations of CWI. We performed the same analyses using the new definition of cold period (NOV–FEB, for example) to check the consistent patterns of the results. For a 1-day lagged CWI for cardiovascular-related death, the thresholds were −2.7 °C, − 4.0 °C, and −2.0 °C, for all age groups, the 0–64 year old age group, and the ≥65 year old age group, respectively. Estimated threshold values from the 3 month analyses were −2.8 °C, −3.3 °C, and −0.2 °C, for all age groups, the 0–64 year old age group, and the ≥65 year old age group, respectively. And the effects of the CWI below the threshold were 1.783% (95% CI: 0.770–2.788%), 3.998% (95% CI: 1.455–6.474%), and 1.336% (95% CI: 0.316– 2.345%) for all age groups, the 0–64 year old age group, and the ≥65 year old age group, respectively. They were 1.799 (0.631–2.950), 2.938 (0.557–5.262), and 0.839 (0.089–1.582). The results from the new definition of cold period (NOV–FEB) were not very different to those from the cold period definition from December to February.
5.
Conclusions
The effects of the daytime CWI lagged by 0–2 days were the strongest among those of the daytime CWI lagged by 0–6 days. The overall shape of the CWI–mortality curve was nonlinear, and mortality tended to increase as the CWI decreased. In addition, this increase in mortality was steeper below the CWI threshold. The most significant mortality outcomes were cardiovascular-related. Although we could not consider respiratory-related mortality, the effect of CWI on cardiovascular-related mortality below a certain threshold was greater than cardiorespiratory-related or all cause-related mortality. In addition, the association between mortality and CWI was more immediate and vulnerable in an elderly subgroup (≥65 years old) than in a younger subgroup (0–64 years old). A recent publication from the World Health Organization encourages public health decision makers to act now to address climate hazards, especially those related to heat waves (Menne et al., 2005). Our results provide evidence that there is also a serious risk of cold-related mortality. Public health initiatives to combat the dangers of ambient temperatures should target people who are particularly vulnerable to not only hot weather, but also cold weather.
2163
Acknowledgment This study was supported by the Eco-Tecnopia-21, Ministry of the Environment (No. 091-071-057) and Korea Environment Institute (2008-032).
REFERENCES
Akaike H. Information theory and an extension of the maximum likelihood principal. In: Petrov BN, Caski F, editors. Second International Symposium on Information Theory. Budapest: Akademial Kiado; 1973. p. 267–81. Armstrong B. Models for the relationship between ambient temperature and daily mortality. Epidemiology 2006;17(6):624–31. Basu R, Samet JM. Relation between elevated ambient temperature and mortality: a review of the epidemiologic evidence. Epidemiol Rev 2002;24:190–202. Bowie N, Jackson G. The raised incidence of winter deaths. General Register Office for Scotland, occasional paper number 7; 2002. Braga ALF, Zanobetti A, Schwartz J. The time course of weather-related deaths. Epidemiology 2001;12(6):662–7. Carder M, McNamee R, Beverland I, Elton R, Cohen GR, Boyd J, et al. The lagged effect of cold temperature and wind-chill on cardiorespiratory mortality in Scotland. Occup Environ Med 2005;62:702–10. 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. Eng H, Mercer J. Seasonal variations in mortality caused by cardiovascular disease in Norway and Ireland. J Cardiovasc Risk 1998;5:89–95. Holland WW, Spicer CC, Wilson JMG. Influence of weather on respiratory and heart disease. Lancet 1961;2:334. IPCC. Fourth Assessment Report: Climate Change 2007 — impacts, adaptation and vulnerability; 2007. Johnson H, Kovats S, McGregor G, Stedman J, Gibbs M, Walton H. The impact of the 2003 heat wave on daily mortality in England and Wales and the use of rapid weekly mortality estimates. Euro Surveill 2005;10:161–5. Keatinge WR, Donaldson GC. Mortality related to cold and air pollution in London after allowance for effects of associated weather patterns. Environ Res Sect A 2001;86:209–16. Keatinge WR, Donaldson GC, Bucher K, Jendritsky G, Cordioli E, Martinelli M. Cold exposure and winter mortality from ischaemic heart disease, cerebrovascular disease, respiratory disease, and all causes in warm and cold regions of Europe. Lancet 1997;349:1341–6. Kim SY, Lee JT, Hong YC, Ahn KJ, Kim H. Determining the threshold effect of ozone on daily mortality: an analysis of ozone and mortality in Seoul, Korea, 1995–1999. Environ Res 2004;94:113–9. Kim H, Ha J, Park J. High temperature, heat index, and mortality in 6 major cities in South Korea. Arch Environ Occup Health 2006;61(6):265–70. KMA. Regulations of forecast services. Seoul: Korea Meteorological Administration, 2008; 24/03/2008. Downloaded in November 2008, website: http://web.kma.go.kr/open/pds/legal/legal02/ legal02-06.html. Kosatsky T. The 2003 European heat waves. Euro Surveill 2005;10:148–9. Lee JT, Schwartz J. Reanalysis of the effects of air pollution on daily mortality in Seoul, Korea: a case-crossover design. Environ Health Perspect 1999;107(8):633–6.
2164
S CI EN CE OF T H E T OTAL EN V I RO N M EN T 4 0 7 ( 2 0 09 ) 21 5 8–2 16 4
Lee JT, Dockery DW, Kim CB, Jee SH, Chung Y. A meta-analysis of ambient air pollution in relation to daily mortality in Seoul, 1991–1995. J Prev Med Pub Health 1999a;32(2):177–82. Lee JT, Shin DC, Chung Y. Air pollution and daily morality in Seoul and Ulsan, Korea. Environ Health Perspect 1999b;107(2):149–54. Loudon RG. Weather and cough. Am Rev Respir Dis 1964;89:352. McKee CM. Deaths in winter: can Britain learn from Europe? Eur J Epidemiol 1989;5:178–82.
Menne B, Ebi KL, editors. Climate change and adaptation strategies for human health. Heidelburg: Steinkopff Verlag, Darmstadt; 2005. Pattenden S, Nikiforov B, Armstrong BG. Mortality and temperature in Sofia and London. J Epidemiol Community Health 2003;57:628–33. Russell WT. The relative influence of fog and low temperature. Lancet 1928;2:1128–30.
a v a i l a b l e a t w w w. s c i e n c e d i r e c t . c o m
w w w. e l s e v i e r. c o m / l o c a t e / s c i t o t e n v
Relationship between winter temperature and mortality in Seoul, South Korea, from 1994 to 2006 Jongsik Haa , Joungho Yoona , Ho Kimb,⁎ a
Korea Environment Institute, 613-2 Bulgwang-dong Eunpyeong-gu, Seoul 122-706, South Korea Department of Biostatistics and Epidemiology, School of Public Health and the Institute of Health and Environment, Seoul National University, 28 Yunkeon-dong Chongro-gu Seoul 110-799, South Korea b
AR TIC LE D ATA
ABSTR ACT
Article history:
In South Korea, the mortality increases with temperature above a certain threshold during
Received 4 July 2008
the warm season. In contrast, despite the common burden of cold weather, little is known
Received in revised form
about the effects of low temperatures on mortality. We evaluated the relationship between
26 November 2008
low temperatures and mortality in winter (December–February) in Seoul, South Korea, from
Accepted 5 December 2008
1994 to 2006. Data were obtained from government agencies. After adjusting for trends in
Available online 20 January 2009
time, day of the week, holidays, and relative humidity, we explored the associations between mortality and various cold indicators of winter in Seoul, South Korea. First, we
Keywords:
fitted nonparametric smoothing regression models to check the shape of associations and
Cold wave
then fitted threshold models (including two different slopes in a model) to estimate the
Low-temperature mortality
thresholds and the effects of low temperatures using the Akaike Information Criterion (AIC).
Threshold
The graphical associations between cardiorespiratory, cardiovascular, and all causes of mortality and the cold wave index (CWI = Tmin, previous day − Tmin, current day) were observed. We confirmed the threshold according to a lag structure and after that, estimated the effects of CWI below the threshold, respectively. The effects of the daytime CWI lagged by 0–2 days were the strongest among those of the daytime CWI lagged by 0–6 days. The most significant mortality outcomes were cardiovascular-related. Although we could not consider respiratory-related mortality, the effect of CWI on cardiovascular-related mortality below a certain threshold was greater than cardiorespiratory-related or all cause-related mortality. In addition, the association between mortality and CWI was more immediate and vulnerable in an elderly subgroup (≥ 65 years old) than in a younger subgroup (0–64 years old). Our results suggest that public health programs should be considered to alleviate not only the effect of sudden change in winter temperature on mortality. © 2008 Elsevier B.V. All rights reserved.
1.
Introduction
It is now widely accepted in the scientific community that the Earth's climate is changing as a result of human activities. Climate change causes various environmental and ecological changes, and some of these changes will affect human health. In 2007, the Intergovernmental Panel on Climate Change (IPCC) reported that climate change is likely to lead to more
intense and frequent extreme weather events and that human exposure to such changing weather patterns could result in increased death, disability, and suffering (IPCC, 2007). The dangers of hot weather were demonstrated dramatically by the 2003 heat wave that accounted for more than 30,000 deaths throughout Western Europe and more than 2000 deaths in England and Wales alone (Johnson et al., 2005; Kosatsky, 2005). Detrimental effects of cold weather on health
⁎ Corresponding author. Tel.: +82 2 740 8874; fax: +82 2 745 9104. E-mail address: [email protected] (H. Kim). 0048-9697/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.scitotenv.2008.12.029
S CI EN C E OF TH E T OTAL EN V I RO N M EN T 4 0 7 ( 2 0 09 ) 21 5 8–2 16 4
have also been observed. These have been assessed using a variety of health endpoints, since the early 20th century (Russell, 1928; Holland et al., 1961; Loudon, 1964; Pattenden et al., 2003). In South Korea, the mortality increases with temperature above a certain threshold during the warm season (Kim et al., 2006). In contrast, despite the common burden of exposure to cold weather, little is known about the links between low temperatures and mortality. This has resulted in an apparent lack of concern by government agencies to proactively reduce risk in South Korea. Mortality rates for cardiovascular and respiratory disease typically exhibit distinct seasonal variation, with the highest rates occurring in the winter months (McKee, 1989; Eng and Mercer, 1998; Bowie and Jackson, 2002). The main factor driving this seasonal pattern is thought to be temperature (Basu and Samet, 2002). Braga et al. (2001) reported that low temperatures are strongly associated with increased cardiorespiratory-caused deaths. Keatinge et al. (1997) reported that Mortality increased to a greater extent with given fall of temperature in regions with warm winters. We evaluated the relationship between low temperatures and mortality during winter (December–February) in Seoul, South Korea, from 1994 to 2006. We used the time-series generalized additive model (GAM) and included nonparametric smoothing functions of a cold wave index (CWI) for our initial analysis (Basu and Samet, 2002). We then fit the data to time-series piecewise log-linear models, with one slope representing values below the CWI threshold and one representing those above the threshold (Armstrong, 2006).
2.
Materials and methods
2.1.
Study scope
This study focused on the relation between cold weather and mortality in Seoul, South Korea. We examined the effect of sudden change in temperature on winter mortality, particularly for the elderly. We postulated that delay and threshold between exposure and response might be heterogeneous among population groups. Therefore, we performed the analysis for the entire population, as well as for two age subgroups: 0–64 and ≥ 65 years old. For each group, we performed separate model-fitting analyses according to lags of 0, 1, and 2 days after the cold indicator to compare the magnitude of the intensity of influence by age group.
2.2.
Weather and mortality data
Weather data, including measurements of relative humidity taken every 3 h and hourly measurements of ambient temperature, were provided by the Korean Meteorological Administration. We began with substantial preliminary analyses of GAMs with natural cubic splines in which several types of representative values served as daily cold indicators, e.g., the daily CWI, minimum temperature, and mean temperature. We selected the CWI as the indicator based on the best graphical association by visual assessment in the model fittings. The
2159
CWI was defined as CWI = T_mp − T_mc, where T_mc is the minimum temperature of the current day, and T_mp is the minimum temperature of the previous day. For example, a CWI of 5 °C indicates that the current day's minimum temperature is 5 °C lower than the previous day's minimum temperature. In general, a cold wave is defined as an unusually large and rapid drop in temperature over a short period of time, e.g., 24 h. Therefore, the CWI is not the absolute cold, but rather the relative cold in human-perceived equivalent temperature (i.e., how cold it feels, termed the “felt air temperature by a rapid fall in temperature within a 24 hour period”). The Korean Meteorological Administration calculates the criterion for announcing a cold wave warning using the CWI formula (KMA, 2008). The Korean National Statistical Office provided the mortality data. These data excluded deaths of individuals in the study area who did not reside within the area, as well as deaths of individuals caused by accidents (ICD10 codes V00-Y99). In this study, we considered the following deaths: all causes of mortality (ICD10 codes A00-U99), cardiorespiratory-related mortality (ICD10 codes I00–I99, J00–J99), and cardiovascular-related mortality (ICD10 codes I00–I99). We excluded respiratory-related mortality because of limited daily counts (mean = 6.23, SD = 3.05 in winter in Seoul from 1994 to 2006). The relationship between the CWI and mortality may be confounded by trends in time, day of the week, holidays, and humidity. For controlling trends in time, we used indicator variables such that the first winter was January and February 1994, the second winter was December 1994 and January and February 1995, up to the last winter, which was December 2006. In addition, we included smoothing function of date on our model to control the possible confounding effect of time. For controlling day of the week and holidays, we used indicator variables to categorize days into holidays, including Sunday; the day after a holiday; Saturday; and other days. In the end, we controlled daily mean humidity using a natural cubic spline of four degrees of freedom.
2.3.
Modeling approach
The statistical analysis was conducted in two steps. First, to investigate the shape of the CWI–mortality curve, we fitted GAMs with natural cubic splines to the CWI data, using four degrees of freedom as a compromise between providing adequate control for unmeasured confounders and leaving sufficient information from which to estimate the CWI effect. Second, if the shape of the CWI–mortality curve was a reverse J shape (i.e., a negative association with CWI below the threshold and no association or a small positive association above the threshold), the following piecewise log-linear model was fit to the data: logEðY Þ = b0 + S1 ðx1 Þ + N + Sp xp + b1 ðCWI nÞ + + b2 CWI; ð1Þ
where x1, x2, …, xp are covariates, and (CWI − ξ)+ = min{CWI − ξ,0}. If the CWI is less than the threshold value (ξ): logEðYÞ = b0 b2 n + S1 ðx1 Þ + N + Sp xp + ðb1 + b2 Þ CWI:
ð2Þ
2160
S CI EN CE OF T H E T OTAL EN V I RO N M EN T 4 0 7 ( 2 0 09 ) 21 5 8–2 16 4
If the CWI is greater than the threshold value (ξ):
logEðY Þ = b0 + S1 ðx1 Þ + N + Sp xp + b2 CWI:
ð3Þ
These two lines are constrained to meet at the threshold value (ξ). The model is also known as a B-mode splined linear regression model (Akaike, 1973; Armstrong, 2006; Kim et al., 2004). We identified the threshold values that produced the model with the lowest AIC. However, when the relationship was linear, we fitted the data using the following linear model: logEðY Þ = b0 + S1 ðx1 Þ + N + Sp xp + b1 CWI
ð4Þ
where x1, x2, …, xp are covariates. The Eq. (1) does not assign any death to CWI levels above the threshold whereas the Eq. (4) does. So, the RR (negative association) may be higher in the Eq. (1). If threshold effect of CWI exists in the mortality risk level, CWI risk estimates based on an assumption of linearity may represent an underestimation of the true risk. Relative risk (RR) was calculated using exp(β), where β is the related model coefficient. The percentage change was derived from RR using the formula (RR − 1) × 100. All analyses were performed using SPlus v2000 (Mathsoft Inc., Cambridge, MA). The convergence tolerances of the GAM functions were set to 10− 9, with a limit of 1000 iterations to avoid biased estimates of regression coefficient and standard error (Dominici et al., 2002; Pattenden et al., 2003).
3.
Results
3.1.
Description
In the time-series of the daily mean temperature and daily CWI from 1994 to 2006, there is a clear yearly pattern in CWI, with more fluctuations occurring during the winter months (Fig. 1). The standard deviation of the CWI was 3.21 °C during
the cold season (December–February), 1.41 °C during the hot season (June–August), and 2.45 °C during the other seasons (Table 1). The average daily mortality due to all causes was 12, 5, and 5% greater in the cold season than in the hot season, other seasons, and year-round, respectively. A similar relationship was observed for cardiorespiratory- and cardiovascular-related deaths.
3.2.
Graphical results
We plotted the estimated CWI–mortality functions derived from the GAM with the CWI in the smoothing function (Fig. 2). The graphical analysis using 1-day lagged CWI resulted in a reverse J-shaped curve for every population group for cardiovascular-related mortality, but only for the 0–64 year old age group for cardiorespiratory-related mortality. The strength of the graphical relationship seemed to depend on the cause of death; it was greater for cardiovascular-related than cardiorespiratory-related mortality, which were both greater than for all causes of mortality.
3.3.
Effects and thresholds
Once the threshold was determined in the CWI–mortality associations, we fitted piece-wise log-linear models to estimate the threshold and the slope for the range of CWI below the threshold. For cardiorespiratory-related mortality, a 1 °C decrease in the 1-day lagged CWI below −2.9 °C was significantly associated with increased mortality, and the effect of a CWI below the threshold was a 2.379% (95% confidence interval [CI]: 0.374–4.343%) increase in the 0–64 year old age group (Table 2). A 1 °C decrease in the 2-day lagged CWI was linearly associated and the effect was a 0.754% (95% CI: 0.261–1.299%) increase in the 0–64 year old age group. For cardiovascular-related mortality, the thresholds in the 1-day lagged CWI were −2.8 °C, − 3.3 °C, and − 0.2 °C, and the effects of the CWI were a 1.799% (95% CI: 0.631–2.950%), 2.938% (95% CI: 0.557–5.262%), and 0.839% (95% CI: 0.089–1.582%) increase in all
Fig. 1 – Daily mean temperature and daily cold wave index in Seoul from 1994 to 2006.
2161
S CI EN C E OF TH E T OTAL EN V I RO N M EN T 4 0 7 ( 2 0 09 ) 21 5 8–2 16 4
Table 1 – Characteristics of the population and weather in Seoul, South Korea, from 1994 to 2006 Winter (N = 1173) (December– February)
Daily mortality counts
All cause
Cardiorespiratory (I00–I99, J00–J99) Cardiovascular (I00–I99) Daily cold wave index (°C) Daily mean temperature (°C) Daily mean relative humidity (%)
All ages ≥ 65 0–64 All ages ≥ 65 0–64 All ages ≥ 65 0–64
Summer (N = 1196) ( June–August)
Other seasons (N = 2379)
Year-round (N = 4748)
Mean
SD
Mean
SD
Mean
SD
Mean
SD
102.38 62.34 40.04 33.81 24.23 9.58 27.57 19.23 8.34 0.01 −0.18 56.84
12.35 10.08 7.25 7.03 5.79 3.48 5.85 4.78 3.11 3.21 4.31 13.36
91.35 54.31 37.04 28.14 20.12 8.02 23.33 16.30 7.03 −0.06 24.60 73.24
11.39 9.28 6.66 6.24 5.34 2.93 5.56 4.79 2.64 1.41 2.81 11.52
97.83 58.78 39.05 31.36 22.37 8.99 26.04 18.12 7.92 0.02 13.52 60.30
11.27 9.38 6.80 6.19 5.38 3.26 5.58 4.77 3.02 2.45 6.44 13.96
97.32 58.53 38.79 31.16 22.27 8.89 25.73 17.94 7.79 0.00 12.93 62.70
12.22 9.95 6.97 6.72 5.67 3.28 5.85 4.89 2.99 2.46 10.21 14.64
age groups, the 0–64 year old age group, and the ≥65 year old age group, respectively (Table 2). In addition, in the ≥65 year old age group, the threshold in the 0–day lagged CWI was −3.4 °C, and the effect was a 1.593% (95% CI: 0.108–3.055%) increase. A 1 °C decrease in the 2-day lagged CWI was linearly associated, and the effect was a 0.711% (95% CI: 0.130–1.288%) increase in the 0–64 year old age group.
4.
Discussion
Our study was designed to estimate the RR of the CWI on mortality by cause of death and age group. For the fitting with 0–2 days lagged CWI for cardiovascular-related death, we found that (1) model with 0–1 day lags was better than models
Fig. 2 – Shape of the cold wave index–mortality curve in winter.
2162
S CI EN CE OF T H E T OTAL EN V I RO N M EN T 4 0 7 ( 2 0 09 ) 21 5 8–2 16 4
Table 2 – Estimated threshold and percent increases Cause of death
Cardiorespiratory (I00–I99, J00–J99)
Age group (years)
All ages
≥ 65
0–64
Cardiovascular (I00–I99)
All ages
≥ 65
0–64
Cold wave index lag in days
0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2
Graphical association
None Threshold None Threshold None None None Threshold Linear Threshold Threshold None Threshold Threshold None None Threshold Linear
Threshold (°C)
Estimated percent increase in daily mortality in 1 °C temperature decreases below threshold
Observed
Sampled 95% CI
Percent increase
95% CI
NA − 1.5 NA − 2.3 NA NA NA − 2.9 NA − 3.2 − 2.8 NA − 3.4 − 0.2 NA NA − 3.3 NA
– – – – – – – – – – (− 3.25 to −2.12) – (−5.04 to − 2.99) (−3.53–1.25) – – (−4.63 to − 2.98) –
NA 0.793 NA 0.812 NA NA NA 2.379 0.754 0.541 1.799 NA 1.593 0.839 NA NA 2.938 0.711
NA (− 0.018–1.598) NA (− 0.369–1.979) NA NA NA (0.374–4.343) (0.261–1.299) (− 0.845–1.909) (0.631–2.950) NA (0.108–3.055) (0.089–1.582) NA NA (0.557–5.262) (0.130–1.288)
Note 1: The threshold is the temperature at which the risk of mortality begins to increase as the cold wave index decreases. Note 2: NA means ‘Not Available’. Note 3: The sampled 95% CI of threshold was estimated for each exposure–response curve having significance effect below a certain threshold in cardiovascular-related mortality by re-sampling technique. In the re-sampling technique, the same method to identify the threshold was repeated 1000 times by leaving 10% subject out each time, and then the empirical distribution of the threshold was identified. We calculated the sampled 95% confidence interval by using the mean and standard deviation of the distributions.
with any other lag for the ≥65 year old age group; (2) model with 1–2 day lags was better than models with any other lag for the 0–64 year old age group. For a 1-day lagged CWI for cardiovascular-related death, thresholds were −0.2 °C, −3.3 °C and −2.8 °C, for ≥65 year old age group, 0–64 year old age group, and all ages group, respectively. However, the degree of increase in RR below the threshold for the 0–64 year old age group (2.938%, 95% CI: 0.557–5.262%) was larger than those for the ≥65 year old age group (0.839%, 95% CI: 0.089–1.582%) and for the all ages group (1.799%, 95% CI: 0.631–2.950%). These results could be explained by the following two points. First, the strongest association for the 0–64 year old age group tended to be a 1-day lag, whereas that of the ≥65 year old age group was a 0-day lagged CWI, which might indicate that the current day's exposure has a more immediate effect than that of the previous day on the elderly. Second, the estimated threshold for the ≥65 year old age group with a 1-day lagged CWI (−0.2 °C vs. −3.3 °C for 0–64 year old age group) indicates that the elderly may be more vulnerable to a lower CWI than young people. On the other hand, the estimated percentage increases for cardiovascular-related death below the threshold of a 1-day lagged CWI were 2.938% for the 0–64 year old age group and 0.839% for the ≥65 year old age group. Because we separately estimated the effect of CWI on mortality below different thresholds for each age group, the slopes of CWI below the threshold can not be compared with any other slope without the threshold. The re-sampling technique was employed to assess the stability of the threshold for each exposure–response curve
having significance effect below a certain threshold in cardiovascular-related mortality. In the re-sampling technique, the same method to identify the threshold was repeated 1000 times by leaving 10% subject out each time. In Fig. 2, the sampled 95% confidence intervals from the simulations show the degree of stability. Although the sampled 95% CI for the ≥65 year old age group with a 1-day lagged CWI was larger than that of any other groups, all of them included the observed thresholds. The re-sampling result implies that the threshold is fairly stable. The results using cause-of-death data showed the same pattern, with cardiovascular-related death stronger than that of cardiorespiratory-related death, and statistically significant for each age group and lagged CWI. So, even though we did not conduct an analysis for respiratory-caused death, we may deduce that people with cardiovascular disease are more vulnerable to cold weather than those with respiratory disease. The effects of cold temperatures may occur over prolonged periods of time, in some cases up to a few weeks (Keatinge and Donaldson, 2001; Carder et al., 2005). However, the CWI that we used represented the current day's minimum temperature fluctuation relative to the previous day's minimum temperature. Therefore, our results could differ from studies that use only the temperature of the current day as the cold indicator. Many studies have suggested that particulate air pollution affects the daily mortality rate. For example, studies on the effects of air pollution on the daily death rate in Seoul, South Korea, in the 1990s report a 1.3–3.7% increase in the rate of all causes of mortality for an inter-quartile increase in the concentration of particulate Matter 10 micrometers or less
S CI EN C E OF TH E T OTAL EN V I RO N M EN T 4 0 7 ( 2 0 09 ) 21 5 8–2 16 4
(PM10) (Lee and Schwartz, 1999; Lee et al., 1999a,b). In spite of that, Braga et al. (2001) did not control for PM10 in temperature and mortality study. We could not consider the PM10 effect due to limited data period. In Korea, PM10 data is obtainable from 1997. So, if there is a substantial confounding or modifying effect, an inappropriately specified model may result in bias. We considered the city of Seoul; thus, our results are limited to this city only. Keatinge et al. (1997) reported that mortality increases to a greater extent for a given decrease in temperature in regions that have warm winters. Our results should not be generalized to other regions. It is necessary to determine whether the application of our methods to other regions would produce similar results. In Our study, December, January, and February were defined as winter. South Korea has four distinct seasons. Summer is hot and humid, and winter is cold and dry. In fact, the cold period from December to February is the coldest in South Korea and has largest fluctuations of CWI. We performed the same analyses using the new definition of cold period (NOV–FEB, for example) to check the consistent patterns of the results. For a 1-day lagged CWI for cardiovascular-related death, the thresholds were −2.7 °C, − 4.0 °C, and −2.0 °C, for all age groups, the 0–64 year old age group, and the ≥65 year old age group, respectively. Estimated threshold values from the 3 month analyses were −2.8 °C, −3.3 °C, and −0.2 °C, for all age groups, the 0–64 year old age group, and the ≥65 year old age group, respectively. And the effects of the CWI below the threshold were 1.783% (95% CI: 0.770–2.788%), 3.998% (95% CI: 1.455–6.474%), and 1.336% (95% CI: 0.316– 2.345%) for all age groups, the 0–64 year old age group, and the ≥65 year old age group, respectively. They were 1.799 (0.631–2.950), 2.938 (0.557–5.262), and 0.839 (0.089–1.582). The results from the new definition of cold period (NOV–FEB) were not very different to those from the cold period definition from December to February.
5.
Conclusions
The effects of the daytime CWI lagged by 0–2 days were the strongest among those of the daytime CWI lagged by 0–6 days. The overall shape of the CWI–mortality curve was nonlinear, and mortality tended to increase as the CWI decreased. In addition, this increase in mortality was steeper below the CWI threshold. The most significant mortality outcomes were cardiovascular-related. Although we could not consider respiratory-related mortality, the effect of CWI on cardiovascular-related mortality below a certain threshold was greater than cardiorespiratory-related or all cause-related mortality. In addition, the association between mortality and CWI was more immediate and vulnerable in an elderly subgroup (≥65 years old) than in a younger subgroup (0–64 years old). A recent publication from the World Health Organization encourages public health decision makers to act now to address climate hazards, especially those related to heat waves (Menne et al., 2005). Our results provide evidence that there is also a serious risk of cold-related mortality. Public health initiatives to combat the dangers of ambient temperatures should target people who are particularly vulnerable to not only hot weather, but also cold weather.
2163
Acknowledgment This study was supported by the Eco-Tecnopia-21, Ministry of the Environment (No. 091-071-057) and Korea Environment Institute (2008-032).
REFERENCES
Akaike H. Information theory and an extension of the maximum likelihood principal. In: Petrov BN, Caski F, editors. Second International Symposium on Information Theory. Budapest: Akademial Kiado; 1973. p. 267–81. Armstrong B. Models for the relationship between ambient temperature and daily mortality. Epidemiology 2006;17(6):624–31. Basu R, Samet JM. Relation between elevated ambient temperature and mortality: a review of the epidemiologic evidence. Epidemiol Rev 2002;24:190–202. Bowie N, Jackson G. The raised incidence of winter deaths. General Register Office for Scotland, occasional paper number 7; 2002. Braga ALF, Zanobetti A, Schwartz J. The time course of weather-related deaths. Epidemiology 2001;12(6):662–7. Carder M, McNamee R, Beverland I, Elton R, Cohen GR, Boyd J, et al. The lagged effect of cold temperature and wind-chill on cardiorespiratory mortality in Scotland. Occup Environ Med 2005;62:702–10. 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. Eng H, Mercer J. Seasonal variations in mortality caused by cardiovascular disease in Norway and Ireland. J Cardiovasc Risk 1998;5:89–95. Holland WW, Spicer CC, Wilson JMG. Influence of weather on respiratory and heart disease. Lancet 1961;2:334. IPCC. Fourth Assessment Report: Climate Change 2007 — impacts, adaptation and vulnerability; 2007. Johnson H, Kovats S, McGregor G, Stedman J, Gibbs M, Walton H. The impact of the 2003 heat wave on daily mortality in England and Wales and the use of rapid weekly mortality estimates. Euro Surveill 2005;10:161–5. Keatinge WR, Donaldson GC. Mortality related to cold and air pollution in London after allowance for effects of associated weather patterns. Environ Res Sect A 2001;86:209–16. Keatinge WR, Donaldson GC, Bucher K, Jendritsky G, Cordioli E, Martinelli M. Cold exposure and winter mortality from ischaemic heart disease, cerebrovascular disease, respiratory disease, and all causes in warm and cold regions of Europe. Lancet 1997;349:1341–6. Kim SY, Lee JT, Hong YC, Ahn KJ, Kim H. Determining the threshold effect of ozone on daily mortality: an analysis of ozone and mortality in Seoul, Korea, 1995–1999. Environ Res 2004;94:113–9. Kim H, Ha J, Park J. High temperature, heat index, and mortality in 6 major cities in South Korea. Arch Environ Occup Health 2006;61(6):265–70. KMA. Regulations of forecast services. Seoul: Korea Meteorological Administration, 2008; 24/03/2008. Downloaded in November 2008, website: http://web.kma.go.kr/open/pds/legal/legal02/ legal02-06.html. Kosatsky T. The 2003 European heat waves. Euro Surveill 2005;10:148–9. Lee JT, Schwartz J. Reanalysis of the effects of air pollution on daily mortality in Seoul, Korea: a case-crossover design. Environ Health Perspect 1999;107(8):633–6.
2164
S CI EN CE OF T H E T OTAL EN V I RO N M EN T 4 0 7 ( 2 0 09 ) 21 5 8–2 16 4
Lee JT, Dockery DW, Kim CB, Jee SH, Chung Y. A meta-analysis of ambient air pollution in relation to daily mortality in Seoul, 1991–1995. J Prev Med Pub Health 1999a;32(2):177–82. Lee JT, Shin DC, Chung Y. Air pollution and daily morality in Seoul and Ulsan, Korea. Environ Health Perspect 1999b;107(2):149–54. Loudon RG. Weather and cough. Am Rev Respir Dis 1964;89:352. McKee CM. Deaths in winter: can Britain learn from Europe? Eur J Epidemiol 1989;5:178–82.
Menne B, Ebi KL, editors. Climate change and adaptation strategies for human health. Heidelburg: Steinkopff Verlag, Darmstadt; 2005. Pattenden S, Nikiforov B, Armstrong BG. Mortality and temperature in Sofia and London. J Epidemiol Community Health 2003;57:628–33. Russell WT. The relative influence of fog and low temperature. Lancet 1928;2:1128–30.