Environmental Research 132 (2014) 290–296
Contents lists available at ScienceDirect
Environmental Research journal homepage: www.elsevier.com/locate/envres
A time series study on the effects of cold temperature on road traffic injuries in Seoul, Korea Won-Kyung Lee a,b, Hye-Ah Lee b, Seung-sik Hwang a, Ho Kim c, Youn-Hee Lim d, Yun-Chul Hong e, Eun-Hee Ha b, Hyesook Park b,n a
Department of Social and Preventive Medicine, Inha University School of Medicine, Republic of Korea Department of Preventive Medicine, School of Medicine, Ewha Womans University, 071, Anyangcheon-ro, Yangcheon-ku, Seoul 158-710, Republic of Korea c Department of Epidemiology and Biostatistics, Seoul National University, Seoul, Republic of Korea d Institute of Health and Environment, Seoul National University, Seoul, Republic of Korea e Department of Preventive Medicine, School of Medicine, Seoul National University, Seoul, Republic of Korea b
art ic l e i nf o
a b s t r a c t
Article history: Received 9 February 2014 Received in revised form 14 April 2014 Accepted 17 April 2014
Objective: Although traffic accidents are associated with weather, the influence of temperature on injuries from traffic accidents has not been evaluated sufficiently. The objective of this study was to evaluate the effect of temperature, especially cold temperatures, on injuries from traffic accidents in Seoul, Korea. We also explored the relationship of temperature with different types of traffic accident. Methods: The daily frequencies of injuries from traffic accidents in Seoul were summarized from the integrated database established by the Korea Road Traffic Authority. Weather data included temperature, barometric pressure, rainfall, snow, and fog from May 2007 to December 2011. The qualitative relationship between daily mean temperature and injuries from traffic accidents was evaluated using a generalized additive model with Poisson distribution. Further analysis was performed using piecewise linear regression if graph the showed non-linearity with threshold. Results: The incidence of injuries was 216 per 100,000 person–months in Seoul. The effect of temperature on injuries from traffic accidents was minimal during spring and summer. However, injuries showed a more striking relationship with temperature in winter than in other seasons. In winter, the number of injuries increased as the temperature decreased to o0 1C. The injuries increased by 2.1% per 1 1C decrease under the threshold of the daily average temperature 5.7 1C, which is 10-fold greater than the effect of temperature above the threshold. Some groups were more susceptible to injuries, such as young and male drivers, according to the types of traffic accident when the temperature decreased to below the freezing temperature. Conclusions: The incidence of injuries increased sharply when the temperature decreased below freezing temperature in winter. Temperature can be effectively used to inform high risk of road traffic injuries, thus helping to prevent road traffic injuries. & 2014 Elsevier Inc. All rights reserved.
Keywords: Injuries Temperature Traffic accidents Weather
1. Introduction In early 2012, there was a dramatic cold wave and heavy snow accumulation in Eastern and Northern Europe, leading to more than 800 deaths (Early 2013 European Cold Wave, 2014). Recordbreaking cold was also reported on other continents (Africa and Eastern Asia, including Korea). Twenty-six days had a minimum temperature of o10 1C in Seoul, Republic of Korea, during winter 2012, which was substantially longer than the previous record of 8.7 days between 2001 and 2010 (Korea Meteorological Administration, 2012). Meteorologists predict that the intensity and
n
Corresponding author. Fax: þ82 2 2652 8325. E-mail address: [email protected] (H. Park).
http://dx.doi.org/10.1016/j.envres.2014.04.019 0013-9351/& 2014 Elsevier Inc. All rights reserved.
duration of cold extremes will be severe in the future (U.S. Climate Change Science Program, 2008). Adverse weather conditions are associated with increased numbers of injuries from road traffic accidents (Andreescu and Frost, 1998). Several studies have reported that climate factors such as snow, rain, and temperature increase the risk of road accidents (Eisenberg and Warner, 2005; Asano and Hirasawa, 2003). To date, however, the effect of temperature on road traffic injuries has not been investigated extensively (Gill and Goldacre, 2009; Kim et al., 2012). In one study conducted in Montreal, Canada, temperature had a negative relationship with the number of accidents per day in winter, while it had a positive relationship in summer (Andreescu and Frost, 1998). However, potential confounders, such as climate factors, were not controlled. In addition, most previous studies on the effects of temperature have focused
W.-K. Lee et al. / Environmental Research 132 (2014) 290–296
291
a traffic accident is the person having caused the most culpable failure or the least injured among the persons concerned when their culpable failure is at the same level. The characteristics of the first party were classified by sex (male, female), age group (o 65 or Z 65 years old), motorcyclist, and drunken driving. Severe and mild injuries were defined as those that required medical treatment for 43 and r 3 weeks, respectively. Moreover, the types of traffic accident were classified as car-tocar and car-to-pedestrian accidents. Weather data were obtained from the Korea Meteorological Administration in 2007–2011 (downloaded from http://web.kma.go.kr/eng/index.jsp, November 2012). The meteorological observatory in Seoul is located in Jongro-gu, in the center of Seoul. Temperature (1C), humidity (%), barometric pressure (hPa), precipitation (mm), and fog are measured every hour. The depth of snow (cm) is collected every 3 h. In this study, temperature, barometric pressure, and humidity were summarized as daily mean values. The daily amount of precipitation and depth of snow were categorized as zero, less than median, median, and more than median. The median depth of precipitation was 3.5 mm, and that of snow was 1.78 cm. A foggy day was conditionally defined as a day on which fog was observed.
on health outcomes such as cardiovascular disease, pulmonary disease, and infectious disease (Lim et al., 2012; Grjibovski et al., 2013; Pica and Bouvier, 2013). Injury is an important public health problem because 5.8 million deaths occur each year, accounting for 10% of all-cause deaths worldwide (Injuries and violence: the facts. Geneva, World Health Organization, 2010). Road traffic injuries represent a major problem, as they constitute nearly one-quarter of all injuries (World report on road traffic injury prevention. Geneva, World Health Organization, 2004). Moreover, road traffic injuries have been increasing with respect to the associated mortality and disease burden. Such injuries caused 1.3 million deaths in 2010, which was 46.3% higher than the number of such deaths in 1990 (Murray et al., 2012). In terms of the global burden of disease, the total injuries in 2004 caused 279 million disability-adjusted life years (DALYs), accounting for 11.2% of all causes (Lozano et al., 2012). Road traffic injuries represented the majority at 27% of total injuries. In 1990, 57 million DALYs were caused by road traffic injuries, but by 2010, this number had increased by 33.2% to 75 million DALYs. Therefore, the mean rank of road traffic injuries increased from 12th to 10th during 1990–2010. It is important to explore environmental, human, and vehicle risk factors to prevent injuries (Haddon, 1999). Therefore, the purpose of this study was to evaluate the relationship between injuries from traffic accidents and temperature, especially in winter. We also evaluated whether this relationship changed according to the type of road traffic injury.
2.2. Analysis Korea has a temperate climate comprising four distinct seasons. In this study, the seasons were defined as spring (Mar–May), summer (Jun–Aug), fall (Sep–Nov), and winter (Dec–Feb) according to the meteorological definition in the northern hemisphere. The injury incidence was calculated as the number of injuries from traffic accidents divided by the resident registration population in Seoul during 2007–2011 to comprehend the magnitude of injuries. The analysis was conducted in the following order. First, a generalized additive model (GAM) with Poisson distribution was used to explore the relationship between daily temperature and injuries from traffic accidents by season. Next, the threshold value for temperature in winter was estimated, with a focus on the effect of cold temperature. Finally, the effects of the difference in cold temperature on road traffic injuries according to the characteristics of the first party, the severity of injuries, and the type of traffic accident were investigated. GAM is a flexible regression model that can be applied to evaluate the nonlinear relationship between independent variables and dependent variables by a smooth function (Hastie and Tibshirani, 1986; Wood, 2011). To control for confounders, the final model included barometric pressure, precipitation, snow, fog day, and calendar variables, such as days of the week and holidays.
2. Methods 2.1. Data We used the data from the integrated traffic accident database (DB) established by the Korea Road Traffic Authority (KoRoad). The KoRoad integrates traffic accident data that have been gathered and maintained by police, insurance companies, and mutual aid associations (buses, taxis, trucks, etc.). The integrated DB was established in 2007, and we used the number of injuries from traffic accidents per day in Seoul, Republic of Korea, from May 2007 to December 2011 (accessed and transmitted from http://taas.koroad.or.kr/index.jsp, November 2012). Seoul is the capital and the largest metropolis, with a population of more than 10 million, of the Republic of Korea. The definition and classification of traffic accidents were in accordance with those of the integrated DB. Traffic accidents were defined as accidents accompanied by injuries of persons occurring on account of road traffic. Traffic accidents were categorized by first party, severity of injury, and type of accident. The first party of
Log ðEðYÞÞ ¼ β0 þ S1 ðtemperatureÞþ S2 ðpressureÞ þ S3 ðdateÞþ factorðprecipitationÞ þ factorðsnowÞþ factorðfogÞ þ factorðday of the weekÞþ factorðholidayÞ Daily counts of road traffic injuries had the expected value of Y from the covariates. S1 , S2 , and S3 are the smoothing functions of temperature, barometric pressure, and date, respectively. Degrees of freedom were 5 for daily mean temperature and 15 for date of study period in winter. The relationship between temperature and injuries from traffic accidents were visualized in the graph by season. To estimate the effect of cold temperature on road traffic injuries, the following analysis was restricted to the winter season. It was performed using piecewise
Table 1 Meteorological characteristics in Seoul, 2007–2011. Spring (n¼399)
Summer (n¼ 460)
Fall (n¼ 455)
Mean
Range
Mean
Range
Mean
Range
Mean
Range
Temperature (1C) Mean Min Max Daily range
12.2 8.1 16.6 8.6
2.1, 24.4 5.3, 19.1 0.5, 31.1 0.0, 16.2
24.4 21.3 28.0 6.7
14.9, 30.0 12.3, 26.4 16.4, 35.1 0.0, 13.9
14.9 11.2 19.0 7.9
4.2, 27.0 7.2, 23.3 1.1, 32.2 1.6, 14.9
0.9 4.3 2.7 7.0
14.6, 12.2 17.7, 9.8 11.2, 17.2 1.5, 14.3
Humidity Relative (%)
56.1
19.4, 94.5
71.2
24.0, 93.8
61.7
25.7, 93.0
55.1
27.0, 96.3
Precipitation Mean depth (mm)
2.5
0.0, 69.0
11.3
0.0, 302.0
3.2
0.0, 260.0
0.7
0.0, 35.0
Snow Mean depth (cm)
0.04
0.0, 8.65
–
–
0.004
0,.0 0.61
1.18
0.0, 23.2
Pressure Barometric (hPa) Foggy day (%)
1005 34.0
984, 1020 (8.5)
998 55.0
987, 1009 (12.0)
1008 21.0
994, 1021 (4.6)
1013 17.0
988, 1027 (4.3)
n¼ Number of days. Spring (Mar–May), Summer (Jun–Aug), Fall (Sep–Nov), Winter (Dec–Feb).
Winter (n¼392)
292
W.-K. Lee et al. / Environmental Research 132 (2014) 290–296
linear regression. The model could be specified as follows: Log ðEðYÞÞ ¼ β0 þ β1 ðtemperature ξÞ þ þ β2 temperature þ S2 ðpressureÞ þ S3 ðdateÞ þ factorðprecipitationÞþ factorðsnowÞ þ factorðfogÞ þ factorðday of the weekÞ þ factorðholidayÞ We chose the threshold value with the lowest Akaike information criterion within 7 3 1C around the transition point suggested in the GAM. This study included neither lag nor moving average temperature because a better-fitted value was not provided. The relative risks of injuries for a 1 1C decrease below the threshold were expressed as the % change. Statistical analysis was performed using SAS version 9.3 (SAS Inc., Cary, NC, USA) and R 2.15.0 (R Foundation for Statistical Computing, Vienna, Austria). ‘mgcv’ packages of R software was used to perform GAM analysis. Statistical significance was defined as a P-value of o0.05.
3. Results In total, 1706 days were observed from 1 May 2007 to 31 December 2011. Table 1 shows the daily mean temperature, daily temperature range, humidity, barometric pressure, depth of precipitation, and snow during all four seasons in Seoul. The mean temperature ranged from 14.6 to 30 1C, and the relative humidity ranged from 19.4 to 96.3. The mean temperature in winter was 0.9 1C; it did not overlap with that of summer. However, spring and fall showed similar mean temperatures at 12.2 and 14.9 1C, respectively. It rained for 567 days, which comprised 33.2% of the study period. The average depth of precipitation was the highest in summer (11.3 mm) and lowest in winter (0.7 mm). In contrast, it snowed in winter on 132 days during the study period. The total number of injuries was 1237,694 from 870,330 accidents during the study period (Table 2). There were, on average, 696 daily injuries in spring, 729 in summer, 766 in fall, and 705 in winter. Considering the resident registration population, the incidence of injuries from traffic accidents was 216 per 100,000 person–months. Injuries from severe accidents comprised 10.3% of all injuries from traffic accidents. Moreover, injuries from
Table 2 Daily frequencies of cases of and injuries associated with traffic accidents in Seoul, 2007–2011. N
Spring (n¼ 399)
Summer (n¼ 460)
Fall (n ¼455)
Winter (n ¼392)
Mean SD
Mean SD
Mean SD
Mean SD
All traffic accidents Cases 870,330 495 Injuries 1237,694 696
85 512 108 729
87 537 109 766
103 492 136 705
116 159
Characteristics of the first party Sex Male 1015,589 568 Female 222,105 128
95 599 24 130
92 626 26 140
116 584 29 121
133 32
Age o65 years old 1197,449 673 Z65 years old 40,245 23 Motorcycle 36,810 22 Drunken driving 32,216 19
106 706 7 23 7 24 9 18
108 739 8 26 8 24 9 19
133 683 9 22 7 16 9 19
156 8 7 9
Type of accident Car-to-car 1039,556 578 Car-to-pedestrian 144,215 89
95 613 19 83
98 643 19 88
123 599 22 78
145 22
Severity of injury Mild 1110,273 620 Severe 127,421 75
100 654 13 75
101 687 14 79
125 636 16 69
148 16
n¼ Number of days. N ¼Total number of injuries in traffic accidents. Severe injuries were defined as those that required medical treatment for 43 weeks.
car-to-car and car-to-person accidents comprised 84.0% and 11.7%, respectively, of all injuries. Fig. 1 shows the non-linear relationship between temperature without a lag effect and the number of injuries from traffic accidents. Visual inspection shows that the number of injuries from traffic accidents decreased as the temperature decreased (but remained 40 1C). However, injuries increased as the temperature decreased to o0 1C. With respect to season, a clear J-shaped relationship was observed in fall and winter. In winter, the slope became steeper if the temperature decreased below approximately 5 1C. Fig. 2 shows the subtype (characteristics of the first party, severity of injury, and type of accident) curves of the relationship between temperature and daily injuries for winter. Male drivers were more frequently associated with accident-induced injuries than were female drivers in winter. The relationship between male sex and injuries in winter seemed to be steeper than that between female sex and injuries in winter. Among individuals of o65 years, road traffic injuries began to increase at o 0 1C, while an age of Z65 years showed no significant relationship. With respect to the severity of injury, a steep left-hand curve was observed in mild injuries compared with severe injuries. Moreover, injuries from car-to-car accidents showed a distinct relationship with temperature, while car-to-pedestrian accidents, drunken driving, and motorcycle driving showed no significant relationship. Total road traffic injuries showed the threshold of the daily mean temperature 5.7 1C in winter (Table 3). Below 5.7 1C, road traffic injuries increased by 2.07% per 1 1C decrease. Considering the subtypes, the temperature value below which injuries began to increase appeared to be lower in female drivers. Male drivers showed a 2.14% increase per 1 1C decrease under the threshold, while female drivers showed a 1.72% increase. Among the age groups, the effect reached statistical significant only in persons below the age of 65 years, in which it induced 2.10% more injuries. Mild injuries increased by 2.19% per 1 1C decrease, whereas severe injuries did not have a threshold. Low temperatures of below the threshold had the biggest effect, showing a 2.32% increase per 1 1C decrease in injuries from car-to-car accidents.
Table 3 Percent change in road traffic injuries for each 1 1C decrease below the temperature threshold by subtype during the winter season. Group
Threshold % 95% CI change
All
o 5.7 Z 5.7
Characteristics of the first party Sex Male o 5.7 Z 5.7 Female o 5.6 Z 5.6 Age o 65 years old o 5.7 Z 5.7 Z 65 years old Motorcycle Drunken driving Type of Car-to-car o 5.7 accident Z 5.7 Car-topedestrian Severity of Mild o 5.7 injury 4 5.7 Severe
P-value
2.07 0.20
2.43 1.71 o 0.001 0.34 0.06 0.006
2.14 0.18 1.72 0.26 2.10 0.21 0.05 0.11 0.54 2.32
2.53 0.33 2.59 0.60 2.47 0.35 0.72 0.67 1.26 2.70
0.30 0.33
0.45 0.15 o 0.001 0.02 0.69 0.067
2.19
2.57 1.82 o 0.001
0.26 0.11
0.40 0.11 o 0.001 0.26 0.49 0.552
1.74 0.03 0.85 0.07 1.74 0.07 0.63 0.89 0.17 1.93
o 0.001 o 0.001 o 0.001 0.126 o 0.001 o 0.001 0.894 0.777 0.135 o 0.001
W.-K. Lee et al. / Environmental Research 132 (2014) 290–296
All season unadjusting for seasonality 0.3
0.2
0.2 log RR
log RR
All season adjusting for seasonality 0.3
0.1 0.0
0.1 0.0
−0.1
−0.1 −10
0
10
20
−10
30
Spring
10
20
30
Summer
0.3
0.3
0.2
0.2 log RR
log RR
0
Daily mean temperature
Daily mean temperature
0.1
0.1 0.0
0.0
−0.1
−0.1 0
5 10 15 20 Daily mean temperature
15
25
20 25 Daily mean temperature
30
Winter
Fall 0.3
0.3
0.2
0.2 log RR
log RR
293
0.1
0.1 0.0
0.0
−0.1
−0.1 −5
0
5
10
15
20
25
Daily mean temperature
−15
−10
−5 0 5 Daily mean temperature
10
Fig. 1. Relationship between the frequency of all injuries from traffic accidents and temperature in Seoul, 2007–2011, (a) All seasons adjusting for seasonality, (b) All seasons unadjusting for seasonality, (c) Spring, (d) Summer, (e) Fall, (f) Winter. Generalized additive model of daily mean temperature and frequency of road traffic injuries adjusted for atmospheric pressure, precipitation, snow, fog, day of the week, holiday and date. The dotted lines represent 95% confidence bands. (a) The model plot adjusted for seasonality by smoothing spline of date, (b) the model adjusted for not seasonality but the number of population each year. It was presented for the comparison of the models with and without adjusting for seasonality.
4. Discussion The present study found direct evidence of the effect of cold temperature on road traffic injuries with considerations for climate conditions. In winter, the relationship between temperature and road traffic injuries demonstrated a clear J-shaped curve, and road traffic injuries increased by 2.07% per 1 1C decrease below the freezing temperature. The estimated effects of cold temperature could vary with the first party, severity of injury, and type of accident. Traffic accidents in winter could be influenced by road conditions such as snowy, icy, or wet roads. Near-freezing temperatures could also lead to dangerous road conditions and increase the risk of traffic accidents. Peak numbers of traffic accidents have been documented in early winter in Hokkaido, Japan (Asano and Hirasawa, 2003) and in the winter months in Canada (Andreescu and Frost, 1998). However, studies on traffic-induced injuries in
relation to cold temperature have received relatively little attention. Two previous studies reported the effect of temperature during the winter season on traffic accidents. These studies found a negative correlation in Southern Finland, Finland (Juga, 2012), and in Montreal, Canada (Andreescu and Frost, 1998). The number of injured persons delivered to hospitals reportedly had a positive correlation with temperatures above 0 1C, while a negative relationship between ambulance use due to traumatic injuries and outdoor temperatures was observed at temperatures below 0 1C in Korea (Kim et al., 2012). These findings indirectly suggest that low temperatures are associated with more injuries from traffic accidents because almost half (47%) of ambulance use due to traumatic injuries occurred because of traffic accidents. On the contrary, one study reported that higher temperatures were significantly associated with the occurrence of motor vehicle collisions, with a beta coefficient of 0.025 in an autoregressive integrated moving average model (Abe et al., 2008). With respect to injuries, several
294
W.-K. Lee et al. / Environmental Research 132 (2014) 290–296
Female
Male 0.3 log RR
log R
0.3 0.1 −0.2
0.1 −0.2
−15
−10
−5
0
5
−15
10
−10
Contents lists available at ScienceDirect
Environmental Research journal homepage: www.elsevier.com/locate/envres
A time series study on the effects of cold temperature on road traffic injuries in Seoul, Korea Won-Kyung Lee a,b, Hye-Ah Lee b, Seung-sik Hwang a, Ho Kim c, Youn-Hee Lim d, Yun-Chul Hong e, Eun-Hee Ha b, Hyesook Park b,n a
Department of Social and Preventive Medicine, Inha University School of Medicine, Republic of Korea Department of Preventive Medicine, School of Medicine, Ewha Womans University, 071, Anyangcheon-ro, Yangcheon-ku, Seoul 158-710, Republic of Korea c Department of Epidemiology and Biostatistics, Seoul National University, Seoul, Republic of Korea d Institute of Health and Environment, Seoul National University, Seoul, Republic of Korea e Department of Preventive Medicine, School of Medicine, Seoul National University, Seoul, Republic of Korea b
art ic l e i nf o
a b s t r a c t
Article history: Received 9 February 2014 Received in revised form 14 April 2014 Accepted 17 April 2014
Objective: Although traffic accidents are associated with weather, the influence of temperature on injuries from traffic accidents has not been evaluated sufficiently. The objective of this study was to evaluate the effect of temperature, especially cold temperatures, on injuries from traffic accidents in Seoul, Korea. We also explored the relationship of temperature with different types of traffic accident. Methods: The daily frequencies of injuries from traffic accidents in Seoul were summarized from the integrated database established by the Korea Road Traffic Authority. Weather data included temperature, barometric pressure, rainfall, snow, and fog from May 2007 to December 2011. The qualitative relationship between daily mean temperature and injuries from traffic accidents was evaluated using a generalized additive model with Poisson distribution. Further analysis was performed using piecewise linear regression if graph the showed non-linearity with threshold. Results: The incidence of injuries was 216 per 100,000 person–months in Seoul. The effect of temperature on injuries from traffic accidents was minimal during spring and summer. However, injuries showed a more striking relationship with temperature in winter than in other seasons. In winter, the number of injuries increased as the temperature decreased to o0 1C. The injuries increased by 2.1% per 1 1C decrease under the threshold of the daily average temperature 5.7 1C, which is 10-fold greater than the effect of temperature above the threshold. Some groups were more susceptible to injuries, such as young and male drivers, according to the types of traffic accident when the temperature decreased to below the freezing temperature. Conclusions: The incidence of injuries increased sharply when the temperature decreased below freezing temperature in winter. Temperature can be effectively used to inform high risk of road traffic injuries, thus helping to prevent road traffic injuries. & 2014 Elsevier Inc. All rights reserved.
Keywords: Injuries Temperature Traffic accidents Weather
1. Introduction In early 2012, there was a dramatic cold wave and heavy snow accumulation in Eastern and Northern Europe, leading to more than 800 deaths (Early 2013 European Cold Wave, 2014). Recordbreaking cold was also reported on other continents (Africa and Eastern Asia, including Korea). Twenty-six days had a minimum temperature of o10 1C in Seoul, Republic of Korea, during winter 2012, which was substantially longer than the previous record of 8.7 days between 2001 and 2010 (Korea Meteorological Administration, 2012). Meteorologists predict that the intensity and
n
Corresponding author. Fax: þ82 2 2652 8325. E-mail address: [email protected] (H. Park).
http://dx.doi.org/10.1016/j.envres.2014.04.019 0013-9351/& 2014 Elsevier Inc. All rights reserved.
duration of cold extremes will be severe in the future (U.S. Climate Change Science Program, 2008). Adverse weather conditions are associated with increased numbers of injuries from road traffic accidents (Andreescu and Frost, 1998). Several studies have reported that climate factors such as snow, rain, and temperature increase the risk of road accidents (Eisenberg and Warner, 2005; Asano and Hirasawa, 2003). To date, however, the effect of temperature on road traffic injuries has not been investigated extensively (Gill and Goldacre, 2009; Kim et al., 2012). In one study conducted in Montreal, Canada, temperature had a negative relationship with the number of accidents per day in winter, while it had a positive relationship in summer (Andreescu and Frost, 1998). However, potential confounders, such as climate factors, were not controlled. In addition, most previous studies on the effects of temperature have focused
W.-K. Lee et al. / Environmental Research 132 (2014) 290–296
291
a traffic accident is the person having caused the most culpable failure or the least injured among the persons concerned when their culpable failure is at the same level. The characteristics of the first party were classified by sex (male, female), age group (o 65 or Z 65 years old), motorcyclist, and drunken driving. Severe and mild injuries were defined as those that required medical treatment for 43 and r 3 weeks, respectively. Moreover, the types of traffic accident were classified as car-tocar and car-to-pedestrian accidents. Weather data were obtained from the Korea Meteorological Administration in 2007–2011 (downloaded from http://web.kma.go.kr/eng/index.jsp, November 2012). The meteorological observatory in Seoul is located in Jongro-gu, in the center of Seoul. Temperature (1C), humidity (%), barometric pressure (hPa), precipitation (mm), and fog are measured every hour. The depth of snow (cm) is collected every 3 h. In this study, temperature, barometric pressure, and humidity were summarized as daily mean values. The daily amount of precipitation and depth of snow were categorized as zero, less than median, median, and more than median. The median depth of precipitation was 3.5 mm, and that of snow was 1.78 cm. A foggy day was conditionally defined as a day on which fog was observed.
on health outcomes such as cardiovascular disease, pulmonary disease, and infectious disease (Lim et al., 2012; Grjibovski et al., 2013; Pica and Bouvier, 2013). Injury is an important public health problem because 5.8 million deaths occur each year, accounting for 10% of all-cause deaths worldwide (Injuries and violence: the facts. Geneva, World Health Organization, 2010). Road traffic injuries represent a major problem, as they constitute nearly one-quarter of all injuries (World report on road traffic injury prevention. Geneva, World Health Organization, 2004). Moreover, road traffic injuries have been increasing with respect to the associated mortality and disease burden. Such injuries caused 1.3 million deaths in 2010, which was 46.3% higher than the number of such deaths in 1990 (Murray et al., 2012). In terms of the global burden of disease, the total injuries in 2004 caused 279 million disability-adjusted life years (DALYs), accounting for 11.2% of all causes (Lozano et al., 2012). Road traffic injuries represented the majority at 27% of total injuries. In 1990, 57 million DALYs were caused by road traffic injuries, but by 2010, this number had increased by 33.2% to 75 million DALYs. Therefore, the mean rank of road traffic injuries increased from 12th to 10th during 1990–2010. It is important to explore environmental, human, and vehicle risk factors to prevent injuries (Haddon, 1999). Therefore, the purpose of this study was to evaluate the relationship between injuries from traffic accidents and temperature, especially in winter. We also evaluated whether this relationship changed according to the type of road traffic injury.
2.2. Analysis Korea has a temperate climate comprising four distinct seasons. In this study, the seasons were defined as spring (Mar–May), summer (Jun–Aug), fall (Sep–Nov), and winter (Dec–Feb) according to the meteorological definition in the northern hemisphere. The injury incidence was calculated as the number of injuries from traffic accidents divided by the resident registration population in Seoul during 2007–2011 to comprehend the magnitude of injuries. The analysis was conducted in the following order. First, a generalized additive model (GAM) with Poisson distribution was used to explore the relationship between daily temperature and injuries from traffic accidents by season. Next, the threshold value for temperature in winter was estimated, with a focus on the effect of cold temperature. Finally, the effects of the difference in cold temperature on road traffic injuries according to the characteristics of the first party, the severity of injuries, and the type of traffic accident were investigated. GAM is a flexible regression model that can be applied to evaluate the nonlinear relationship between independent variables and dependent variables by a smooth function (Hastie and Tibshirani, 1986; Wood, 2011). To control for confounders, the final model included barometric pressure, precipitation, snow, fog day, and calendar variables, such as days of the week and holidays.
2. Methods 2.1. Data We used the data from the integrated traffic accident database (DB) established by the Korea Road Traffic Authority (KoRoad). The KoRoad integrates traffic accident data that have been gathered and maintained by police, insurance companies, and mutual aid associations (buses, taxis, trucks, etc.). The integrated DB was established in 2007, and we used the number of injuries from traffic accidents per day in Seoul, Republic of Korea, from May 2007 to December 2011 (accessed and transmitted from http://taas.koroad.or.kr/index.jsp, November 2012). Seoul is the capital and the largest metropolis, with a population of more than 10 million, of the Republic of Korea. The definition and classification of traffic accidents were in accordance with those of the integrated DB. Traffic accidents were defined as accidents accompanied by injuries of persons occurring on account of road traffic. Traffic accidents were categorized by first party, severity of injury, and type of accident. The first party of
Log ðEðYÞÞ ¼ β0 þ S1 ðtemperatureÞþ S2 ðpressureÞ þ S3 ðdateÞþ factorðprecipitationÞ þ factorðsnowÞþ factorðfogÞ þ factorðday of the weekÞþ factorðholidayÞ Daily counts of road traffic injuries had the expected value of Y from the covariates. S1 , S2 , and S3 are the smoothing functions of temperature, barometric pressure, and date, respectively. Degrees of freedom were 5 for daily mean temperature and 15 for date of study period in winter. The relationship between temperature and injuries from traffic accidents were visualized in the graph by season. To estimate the effect of cold temperature on road traffic injuries, the following analysis was restricted to the winter season. It was performed using piecewise
Table 1 Meteorological characteristics in Seoul, 2007–2011. Spring (n¼399)
Summer (n¼ 460)
Fall (n¼ 455)
Mean
Range
Mean
Range
Mean
Range
Mean
Range
Temperature (1C) Mean Min Max Daily range
12.2 8.1 16.6 8.6
2.1, 24.4 5.3, 19.1 0.5, 31.1 0.0, 16.2
24.4 21.3 28.0 6.7
14.9, 30.0 12.3, 26.4 16.4, 35.1 0.0, 13.9
14.9 11.2 19.0 7.9
4.2, 27.0 7.2, 23.3 1.1, 32.2 1.6, 14.9
0.9 4.3 2.7 7.0
14.6, 12.2 17.7, 9.8 11.2, 17.2 1.5, 14.3
Humidity Relative (%)
56.1
19.4, 94.5
71.2
24.0, 93.8
61.7
25.7, 93.0
55.1
27.0, 96.3
Precipitation Mean depth (mm)
2.5
0.0, 69.0
11.3
0.0, 302.0
3.2
0.0, 260.0
0.7
0.0, 35.0
Snow Mean depth (cm)
0.04
0.0, 8.65
–
–
0.004
0,.0 0.61
1.18
0.0, 23.2
Pressure Barometric (hPa) Foggy day (%)
1005 34.0
984, 1020 (8.5)
998 55.0
987, 1009 (12.0)
1008 21.0
994, 1021 (4.6)
1013 17.0
988, 1027 (4.3)
n¼ Number of days. Spring (Mar–May), Summer (Jun–Aug), Fall (Sep–Nov), Winter (Dec–Feb).
Winter (n¼392)
292
W.-K. Lee et al. / Environmental Research 132 (2014) 290–296
linear regression. The model could be specified as follows: Log ðEðYÞÞ ¼ β0 þ β1 ðtemperature ξÞ þ þ β2 temperature þ S2 ðpressureÞ þ S3 ðdateÞ þ factorðprecipitationÞþ factorðsnowÞ þ factorðfogÞ þ factorðday of the weekÞ þ factorðholidayÞ We chose the threshold value with the lowest Akaike information criterion within 7 3 1C around the transition point suggested in the GAM. This study included neither lag nor moving average temperature because a better-fitted value was not provided. The relative risks of injuries for a 1 1C decrease below the threshold were expressed as the % change. Statistical analysis was performed using SAS version 9.3 (SAS Inc., Cary, NC, USA) and R 2.15.0 (R Foundation for Statistical Computing, Vienna, Austria). ‘mgcv’ packages of R software was used to perform GAM analysis. Statistical significance was defined as a P-value of o0.05.
3. Results In total, 1706 days were observed from 1 May 2007 to 31 December 2011. Table 1 shows the daily mean temperature, daily temperature range, humidity, barometric pressure, depth of precipitation, and snow during all four seasons in Seoul. The mean temperature ranged from 14.6 to 30 1C, and the relative humidity ranged from 19.4 to 96.3. The mean temperature in winter was 0.9 1C; it did not overlap with that of summer. However, spring and fall showed similar mean temperatures at 12.2 and 14.9 1C, respectively. It rained for 567 days, which comprised 33.2% of the study period. The average depth of precipitation was the highest in summer (11.3 mm) and lowest in winter (0.7 mm). In contrast, it snowed in winter on 132 days during the study period. The total number of injuries was 1237,694 from 870,330 accidents during the study period (Table 2). There were, on average, 696 daily injuries in spring, 729 in summer, 766 in fall, and 705 in winter. Considering the resident registration population, the incidence of injuries from traffic accidents was 216 per 100,000 person–months. Injuries from severe accidents comprised 10.3% of all injuries from traffic accidents. Moreover, injuries from
Table 2 Daily frequencies of cases of and injuries associated with traffic accidents in Seoul, 2007–2011. N
Spring (n¼ 399)
Summer (n¼ 460)
Fall (n ¼455)
Winter (n ¼392)
Mean SD
Mean SD
Mean SD
Mean SD
All traffic accidents Cases 870,330 495 Injuries 1237,694 696
85 512 108 729
87 537 109 766
103 492 136 705
116 159
Characteristics of the first party Sex Male 1015,589 568 Female 222,105 128
95 599 24 130
92 626 26 140
116 584 29 121
133 32
Age o65 years old 1197,449 673 Z65 years old 40,245 23 Motorcycle 36,810 22 Drunken driving 32,216 19
106 706 7 23 7 24 9 18
108 739 8 26 8 24 9 19
133 683 9 22 7 16 9 19
156 8 7 9
Type of accident Car-to-car 1039,556 578 Car-to-pedestrian 144,215 89
95 613 19 83
98 643 19 88
123 599 22 78
145 22
Severity of injury Mild 1110,273 620 Severe 127,421 75
100 654 13 75
101 687 14 79
125 636 16 69
148 16
n¼ Number of days. N ¼Total number of injuries in traffic accidents. Severe injuries were defined as those that required medical treatment for 43 weeks.
car-to-car and car-to-person accidents comprised 84.0% and 11.7%, respectively, of all injuries. Fig. 1 shows the non-linear relationship between temperature without a lag effect and the number of injuries from traffic accidents. Visual inspection shows that the number of injuries from traffic accidents decreased as the temperature decreased (but remained 40 1C). However, injuries increased as the temperature decreased to o0 1C. With respect to season, a clear J-shaped relationship was observed in fall and winter. In winter, the slope became steeper if the temperature decreased below approximately 5 1C. Fig. 2 shows the subtype (characteristics of the first party, severity of injury, and type of accident) curves of the relationship between temperature and daily injuries for winter. Male drivers were more frequently associated with accident-induced injuries than were female drivers in winter. The relationship between male sex and injuries in winter seemed to be steeper than that between female sex and injuries in winter. Among individuals of o65 years, road traffic injuries began to increase at o 0 1C, while an age of Z65 years showed no significant relationship. With respect to the severity of injury, a steep left-hand curve was observed in mild injuries compared with severe injuries. Moreover, injuries from car-to-car accidents showed a distinct relationship with temperature, while car-to-pedestrian accidents, drunken driving, and motorcycle driving showed no significant relationship. Total road traffic injuries showed the threshold of the daily mean temperature 5.7 1C in winter (Table 3). Below 5.7 1C, road traffic injuries increased by 2.07% per 1 1C decrease. Considering the subtypes, the temperature value below which injuries began to increase appeared to be lower in female drivers. Male drivers showed a 2.14% increase per 1 1C decrease under the threshold, while female drivers showed a 1.72% increase. Among the age groups, the effect reached statistical significant only in persons below the age of 65 years, in which it induced 2.10% more injuries. Mild injuries increased by 2.19% per 1 1C decrease, whereas severe injuries did not have a threshold. Low temperatures of below the threshold had the biggest effect, showing a 2.32% increase per 1 1C decrease in injuries from car-to-car accidents.
Table 3 Percent change in road traffic injuries for each 1 1C decrease below the temperature threshold by subtype during the winter season. Group
Threshold % 95% CI change
All
o 5.7 Z 5.7
Characteristics of the first party Sex Male o 5.7 Z 5.7 Female o 5.6 Z 5.6 Age o 65 years old o 5.7 Z 5.7 Z 65 years old Motorcycle Drunken driving Type of Car-to-car o 5.7 accident Z 5.7 Car-topedestrian Severity of Mild o 5.7 injury 4 5.7 Severe
P-value
2.07 0.20
2.43 1.71 o 0.001 0.34 0.06 0.006
2.14 0.18 1.72 0.26 2.10 0.21 0.05 0.11 0.54 2.32
2.53 0.33 2.59 0.60 2.47 0.35 0.72 0.67 1.26 2.70
0.30 0.33
0.45 0.15 o 0.001 0.02 0.69 0.067
2.19
2.57 1.82 o 0.001
0.26 0.11
0.40 0.11 o 0.001 0.26 0.49 0.552
1.74 0.03 0.85 0.07 1.74 0.07 0.63 0.89 0.17 1.93
o 0.001 o 0.001 o 0.001 0.126 o 0.001 o 0.001 0.894 0.777 0.135 o 0.001
W.-K. Lee et al. / Environmental Research 132 (2014) 290–296
All season unadjusting for seasonality 0.3
0.2
0.2 log RR
log RR
All season adjusting for seasonality 0.3
0.1 0.0
0.1 0.0
−0.1
−0.1 −10
0
10
20
−10
30
Spring
10
20
30
Summer
0.3
0.3
0.2
0.2 log RR
log RR
0
Daily mean temperature
Daily mean temperature
0.1
0.1 0.0
0.0
−0.1
−0.1 0
5 10 15 20 Daily mean temperature
15
25
20 25 Daily mean temperature
30
Winter
Fall 0.3
0.3
0.2
0.2 log RR
log RR
293
0.1
0.1 0.0
0.0
−0.1
−0.1 −5
0
5
10
15
20
25
Daily mean temperature
−15
−10
−5 0 5 Daily mean temperature
10
Fig. 1. Relationship between the frequency of all injuries from traffic accidents and temperature in Seoul, 2007–2011, (a) All seasons adjusting for seasonality, (b) All seasons unadjusting for seasonality, (c) Spring, (d) Summer, (e) Fall, (f) Winter. Generalized additive model of daily mean temperature and frequency of road traffic injuries adjusted for atmospheric pressure, precipitation, snow, fog, day of the week, holiday and date. The dotted lines represent 95% confidence bands. (a) The model plot adjusted for seasonality by smoothing spline of date, (b) the model adjusted for not seasonality but the number of population each year. It was presented for the comparison of the models with and without adjusting for seasonality.
4. Discussion The present study found direct evidence of the effect of cold temperature on road traffic injuries with considerations for climate conditions. In winter, the relationship between temperature and road traffic injuries demonstrated a clear J-shaped curve, and road traffic injuries increased by 2.07% per 1 1C decrease below the freezing temperature. The estimated effects of cold temperature could vary with the first party, severity of injury, and type of accident. Traffic accidents in winter could be influenced by road conditions such as snowy, icy, or wet roads. Near-freezing temperatures could also lead to dangerous road conditions and increase the risk of traffic accidents. Peak numbers of traffic accidents have been documented in early winter in Hokkaido, Japan (Asano and Hirasawa, 2003) and in the winter months in Canada (Andreescu and Frost, 1998). However, studies on traffic-induced injuries in
relation to cold temperature have received relatively little attention. Two previous studies reported the effect of temperature during the winter season on traffic accidents. These studies found a negative correlation in Southern Finland, Finland (Juga, 2012), and in Montreal, Canada (Andreescu and Frost, 1998). The number of injured persons delivered to hospitals reportedly had a positive correlation with temperatures above 0 1C, while a negative relationship between ambulance use due to traumatic injuries and outdoor temperatures was observed at temperatures below 0 1C in Korea (Kim et al., 2012). These findings indirectly suggest that low temperatures are associated with more injuries from traffic accidents because almost half (47%) of ambulance use due to traumatic injuries occurred because of traffic accidents. On the contrary, one study reported that higher temperatures were significantly associated with the occurrence of motor vehicle collisions, with a beta coefficient of 0.025 in an autoregressive integrated moving average model (Abe et al., 2008). With respect to injuries, several
294
W.-K. Lee et al. / Environmental Research 132 (2014) 290–296
Female
Male 0.3 log RR
log R
0.3 0.1 −0.2
0.1 −0.2
−15
−10
−5
0
5
−15
10
−10
