Background The seasonal trend of out-of-hospital coronary death (OHCD) and sudden cardiac death has been observed, but whether extreme temperature serves as a risk factor is rarely investigated. We therefore aimed to evaluate the impact of extreme temperatures on OHCDs in China. We obtained death records of 126 925 OHCDs from six large Chinese cities (Harbin, Beijing, Tianjin, Nanjing, Shanghai and Guangzhou) during the period 2009–2011.
Methods The short-term associations between extreme temperature and OHCDs were analysed with time-series methods in each city, using generalised additive Poisson regression models. We specified distributed lag non-linear models in studying the delayed effects of extreme temperature. We then applied Bayesian hierarchical models to combine the city-specific effect estimates.
Results The associations between extreme temperature and OHCDs were almost U-shaped or J-shaped. The pooled relative risks (RRs) of extreme cold temperatures over the lags 0–14 days comparing the 1st and 25th centile temperatures were 1.49 (95% posterior interval (PI) 1.26–1.76); the pooled RRs of extreme hot temperatures comparing the 99th and 75th centile temperatures were 1.53 (95% PI 1.27–1.84) for OHCDs. The RRs of extreme temperature on OHCD were higher if the patients with coronary heart disease were old, male and less educated.
Conclusions This multicity epidemiological study suggested that both extreme cold and hot temperatures posed significant risks on OHCDs, and might have important public health implications for the prevention of OHCD or sudden cardiac death.
- Clinical epidemiology
- CLIMATE CHANGE
- Environmental epidemiology
Statistics from Altmetric.com
Out-of-hospital coronary death (OHCD) is often used as a surrogate for sudden cardiac death (SCD).1 ,2 SCD accounts for almost one-half of all cardiac deaths,3 although its incidence rate and prevalence vary considerably throughout the world. SCD is frightening because it often occurs with little warning and leads to death within a few hours of symptom onset; the victims may not survive to be discharged from the hospital.4 Furthermore, some of the fatalities can appear in previously stable patients with coronary heart disease (CHD), or even among persons with no signs of preceding heart disorders.5 SCD always occurs outside the hospital environment, and its main underlying cause is CHD; therefore, OHCD represents the major manifestation of SCD.
The primary prevention of SCD remains a major public health challenge. Previous studies have indicated that psychological stress (eg, loss of, or separation from, a relative or a friend), physical activity, smoking, alcohol intake, obesity, diabetes mellitus and outdoor air pollution can increase the risk of SCD or other acute cardiac events.2 ,4 ,6–8 The seasonality of SCD or OHCD, characterised by a winter peak in these events has been reported in Japan,9 Berlin,10 Hungary11 and Israel.12 However, studies on the potential role of extreme ambient temperature as a risk factor are rare. Two prior studies with Poisson regression analyses in Washington13 and Minnesota14 revealed that the winter peak in out-of-hospital cardiac or coronary deaths could be accounted for by the cold temperature in winter, but the two studies were limited by a small number of daily deaths (7.7 and 0.2 on average).
Preliminary data suggested that there were more than 544 000 cases of SCD per year in China, and this estimate was expected to increase in recent years due to rapid socioeconomic progress and lifestyle changes.15 Given China’s large population, SCD constitutes a heavy burden of disease, especially in developed areas where the cardiovascular morbidity is usually high. Previously, a multicity study in China investigated the effects of extreme temperature on ischaemic heart disease mortality,16 but no knowledge was available about the effects on OHCDs.
Therefore, the objective of this epidemiological study was to examine the association between extreme temperature and OHCDs in six large Chinese cities.
This multicity study included participants from Harbin, Beijing, Tianjin, Nanjing, Shanghai and Guangzhou. These cities were chosen according to the different climatic zones of China, regional representativeness and data availability. As shown in figure 1, these cities vary according to geographical and climatic characteristics. For example, Harbin is located in cool temperate zone; Beijing and Tianjin are in warm temperate zone; Nanjing, Shanghai and Guangzhou are in subtropical zone. This analysis was limited to the urban areas because the death registry in rural areas of China may not be as reliable as in urban areas.
We obtained the daily coronary deaths for the period from 1 January 2009 to 31 December 2011 from the Death Register System from Chinese Centre for Disease Control and Prevention. Chinese government has mandated detailed quality assurance and quality control programmes at those institutions providing death data. According to the 10th revision of the International Classification of Diseases, coronary deaths were coded I20–I25. The coding was determined by physicians according to patients’ symptoms, inquiries, complaints and results of medical inspection or the description from the decedents’ relatives. The ascertainment of OHCD relied on standardised criteria from death certificates containing information on the location of death. OHCDs were defined as coronary deaths occurring outside of a medical institution (including hospitals and nursing homes).14 The medical conditions and the characteristics of patients in nursing homes were similar to those in hospitals; therefore, deaths from nursing homes were excluded. Specifically, OHCD refers to all coronary deaths that happen in a private home, public place, ambulance car, emergency room and others declared dead on arrival at a hospital. Those died in emergency room almost had a heart attack outside the hospital, thus were included in this analysis. In each city, death certificates were completed at the time of death either by community doctors for deaths at home or by hospital doctors for deaths at hospitals or in ambulance cars.
To allow for effect modification analyses, we also obtained personal characteristics according to the death certificates, including age, sex, education and marital status. The information on pre-existing cardiac or other relevant illnesses was unavailable because the data were not collected in China's routine death register system.
We collected data on daily mean temperatures and daily mean humidity (dew point temperature) during our study period from the Municipal Meteorological Bureau in each city. Weather data were from one meteorological station in each city, given the broad homogeneity of weather conditions in the same city. We also collected air pollution data to adjust for their potential confounding effects on cardiovascular health end points. Air pollution data were collected from the National Air Pollution Monitoring System and certified by the Ministry of Environmental Protection of China. The measurements from various air quality monitors (9–12 monitors per city) were averaged to represent the general exposure level of all decedents in each city. We incorporated three air pollutants in this analysis: particulate matter less than 10 μ in aerodynamic diameter (PM10), sulfur dioxide (SO2) and nitrogen dioxide (NO2). The Chinese government has mandated detailed quality assurance and quality control programmes at each monitoring station.
In environmental epidemiology, the time-series approach is widely used to investigate the association between acute exposure to environmental risk factors and daily aggregated health outcomes.17 ,18 This approach has the advantage of automatically controlling time-invariant confounders (such as age, sex, smoking, drinking and sociodemographic characteristics) by examining the same population repeatedly over time.19 We analysed the short-term associations between temperature and outcomes in each city with time-series methods. Because daily counts of deaths approximately follow a Poisson distribution and because the relationship between mortality and explanatory variables is generally non-linear, we applied generalised additive Poisson regression models allowing for overdispersion in the data. Previous studies have shown that the short-term association between ambient temperature and cardiovascular outcomes are non-linear and can last for several days20; therefore, we specified distributed lag non-linear models (DLNM) in studying the delayed effects of extreme temperature. The DLNM has the advantage of estimating cumulative effects of temperature on multiple lag days after adjusting for the collinearity of temperature on neighbouring days.21
Specifically, we used a natural cubic spline with five degrees of freedom (df) to account for the non-linear effect of temperature in the ‘cross-basis’ matrix of DLNM.16 ,22 Because previous studies showed that the mortality effects of low temperature could last for weeks, while the effects from hot temperatures tended to be shorter, we selected a maximum lag of 14 days as priori in the DLNM. Several covariates were incorporated: (1) a natural cubic smooth function of calendar time with 7 df/year to exclude any unmeasured long-term and seasonal trends (such as dietary structure and physical activities that may vary by season) in the time-series data set,19 (2) ‘cross-basis’ functions of air pollutants and humidity in the DLNM with a natural cubic spline of 3 df and a maximum lag of 3 days to control the possible non-linear and delayed confounding effects of air pollution and humidity and (3) an indicator variable for the day of the week to account for the in-week variation of deaths. For all of the ‘cross-basis’ matrixes, we used the natural cubic spline with 4 df for the lagged effects (lag space) of temperature, air pollutants and humidity. We did not incorporate barometric pressure and wind conditions, because they were rarely concerned in health studies.
We flexibly plotted the relative risks (RR) of the temperature-mortality association in each city with the RR defining as the risk at each temperature comparing with that at the ‘minimum-mortality temperature’. To quantitatively estimate the effects of extreme temperature, we calculated the RRs at extreme cold temperatures comparing the 1st centiles of city-specific temperature distribution with the 25th centiles, and the RRs at extreme hot temperatures comparing the 99th centiles of city-specific temperature distribution with the 75th centiles in each city.16 ,20
We then applied Bayesian hierarchical models to combine the city-specific RR estimates.8 ,18 ,23 This approach provides a flexible tool to pool effect estimates while accounting for within-city statistical error and between-city variability (heterogeneity) of the ‘true’ effects. The model generated a posterior probability distribution of the pooled mean estimates, from which we reported the combined RRs as the posterior mean and 95% posterior interval (PI), similar to the mean and 95% CI in classic meta-analyses. We implemented this model by using two-level normal independent sampling estimation, which assumed uniform priors on the overall mean estimates and the covariance describing the between-city variations for the cities. We calculated the I2 statistic to evaluate the heterogeneity of city-specific estimates.
To examine the potential modifying roles of individual characteristics on the effects of extreme temperatures on OHCDs, we performed subgroup analyses stratified by a characteristic, such as age, sex, education and marital status.
In addition to the above main analyses, we further performed five sensitivity analyses. First, because it is hard to determine the true lag time of temperature's effects, we estimated the effects of extreme temperature over the lag periods of 0–3, 0–7, 0–21 and 0–28 days. Second, we changed the df per year within the range of 4–8 in the smoothness of time to check the stability of our main findings. Third, we re-examined the associations between extreme temperature and OHCDs by excluding air pollutants from the main models. Fourth, in order to examine the sensitivity of our results to influenza, we included a dummy variable of influenza epidemics by taking the value of one when the 7-day moving average of the respiratory mortality was greater than the 90th centile of its city-specific distribution.24 Fifth, we further examined the additional effects of heat waves and cold spells by including a dummy variable in the regression models. The heat wave was defined as a period of three or more consecutive days with daily mean temperature above the 97.5th of the temperature distribution during the study period. The cold spell was defined as a period of three or more consecutive days with daily mean temperature below the 2.5th of the temperature distribution during the study period.
The statistical tests were two-sided, and values of p<0.05 were considered statistically significant. All models were fitted in the R software (V.2.15.1, R Foundation for Statistical Computing, http://cran.r-project.org/) with the generalised additive Poisson regression models using the ‘mgcv’ package, the DLNM using the ‘dlnm’ package,22 and the Bayesian hierarchical models using the ‘tlnise’ package.
Table 1 summarises the descriptive statistics for daily mean OHCDs, temperatures and air pollution in each city from 2009 to 2011. We identified a total of 126 925 OHCDs, with the daily OHCDs varying from 6 in Guangzhou and Shanghai to 39 in Tianjin. There were no missing data in our data set and no days with zero deaths. The city-specific population size and the total number of OHCDs were provided in online supplementary table S1. We also summarised the distribution of demographic variables (age, gender, education and marital status) in each city in online supplementary table S1. The annual mean temperature ranged from 4.9°C in Harbin to 23°C in Guangzhou, which fully captured the temperature variations in China. According to the time trend plots of daily coronary deaths in each city (curves not shown), there appeared to be a clear seasonal trend for daily OHCD mortality with a peak in winter and a trough in summer.
Figure 2 provided the exposure–response relationship curves for temperature and daily OHCD mortality over lags 0–14 days in each city. Curves of Harbin, Nanjing, Shanghai and Guangzhou were almost U-shaped; curves of Beijing and Tianjin were almost inverse J-shaped. The RRs of all the curves were statistically significant at temperature extremes, that is, the lower 95% confidence limits of RRs above 1.
According to the city-specific temperature cut-offs, we calculated the cumulative RRs of extreme cold and hot temperature on OHCD mortality during the previous 2 weeks. At the averaged levels, the pooled RRs of extreme cold and hot temperatures were 1.49 (95% PI 1.26–1.76) and 1.53 (95% PI 1.27–1.84), respectively. The city-specific effect estimates (see online supplementary table S2) were heterogeneous with the I2>50% for both extreme cold and hot temperature. The corresponding attributable fractions of extreme cold and hot temperatures for OHCDs were 33% (95% PI 21%, 43%) and 35% (95% PI 21%, 46%).
Table 2 summarises the pooled risk estimates associated with extreme temperatures in stratification analyses classified by age groups, gender, educational attainment and marital status. The RR estimates were significant only for elders over 65 years of age. Extreme temperatures were significantly associated with OHCDs among both males and females, but the associations were stronger for males. The significant associations were restricted within those with low education status. We did not detect appreciably different effects of extreme temperatures for those who were singles and married.
We performed five sensitivity analyses to check our main findings. First, we changed the maximum lag of temperature in the DLNMs from 3 to 28 days, which still supported a significant association of extreme temperatures with OHCDs (see table 3). The 14 days in main analysis and 7–28 days in sensitivity analyses produced similar results, demonstrating that there were no apparent ‘harvesting effects’. Second, using alternative df in smoothness of time would not change our main results substantially. Third, the main estimates for the associations between extreme temperature and daily OHCDs were almost unchanged after excluding air pollutants from the models. Fourth, our results were not apparently changed when the influenza indicators were controlled in the models (data not shown). Fifth, the pooled additional RRs of OHCD morality associated with heat wave and cold spell were 1.11 (95% CI 0.97 to 1.27) and 1.08 (95% CI 0.94 to 1.24). The lack of statistical significance suggested that the effect estimates of extreme temperature in main models were robust against additional effects of heat waves and cold spells.
Based on the large sample size, this multicity time-series study suggested that the temperature-OHCD association curves were U-shaped or J-shaped, and that extreme temperatures could significantly increase the risk of OHCDs.
A winter peak in cardiac events has been widely noted in the literature.14 Statistics concerning SCD or arrest, and out-of-hospital cardiac arrest in Japan,9 Berlin,10 Hungary,11 Israel12 and the USA13 ,14 have indicated a remarkable seasonal variation, with an apparent increase in winter; however, these observations were limited to seasonal or month-to-month comparisons. These prior findings have led to the speculation that ambient temperatures may be an important risk factor that could potentially contribute to the onset of OHCD.
The significant and non-linear (U-shaped, V-shaped or J-shaped) associations between temperature and cardiovascular outcomes have been well documented.20 The current analysis found that both low and high outdoor temperature had significant impacts on the daily occurrences of OHCDs. This was, in part, consistent with two time-series analyses performed in Washington13 and Minnesota14 that demonstrated a significant negative association between daily average temperature and out-of-hospital cardiac mortality. Cagle and Hubbard found that a 5°C increase in temperature resulted in a decrease in out-of-hospital cardiac mortality by a factor of 0.97 (95% CI 0.96 to 0.98) in King County, Washington between 1980 and 2001.13 Temperatures below 0°C were strongly associated with unexpected SCD (RR=1.38, 95% CI 1.10 to 1.73), whereas high temperatures had no association with SCD. After excluding seasonal trends of mortality using smoothing functions, we identified a significant association between extreme hot temperatures and the risk of OHCD; this is the first report of its kind in the literature.
We found significant heterogeneity of city-specific effect estimates. Explaining these differences was difficult, and various factors may contribute to the heterogeneity, such as latitudes, the proportion of the elderly, number of air conditioning per household and income per capita. However, limited by the number of cities, we could not fully explore the reasons for the observable differences in each city.
Several mechanisms have been postulated to be responsible for the acute effects of cold and heat on OHCD. Experimental and epidemiological studies have associated cold exposure with increased arterial pressure and blood viscosity and with oxygen deficiency in the cardiac muscle, which can induce myocardial ischaemia or arrhythmias in coronary patients.25 ,26 Furthermore, blood cell counts, plasma cholesterol, C reactive protein and fibrinogen concentrations, all of which may be thrombogenic, appear to be raised when the body is exposed to cold temperatures.27 Low outdoor temperatures have been found to be strongly associated with high blood pressure in Chinese adults across a range of climatic conditions, thus contributing to an increased cardiac risk, especially among coronary patients. However, an accumulation of behavioural risk factors could also trigger an acute coronary event in the winter, such as higher fat intake and lower levels of physical activity.27 Low winter temperatures may also be associated with the flu season, and an increase in upper respiratory tract infections could result in additional stress on the heart. In parallel, when ambient temperatures rise, the body diffuses heat by thermoregulatory means, such as the elevation of heart rates, vasodilation and sweating. These processes can reduce blood supply to vital internal organs, including the heart.28 Heat-induced dehydration can also increase blood viscosity and cholesterol levels and further increase the likelihood of microvascular thrombosis and subsequent heart attacks.29 ,30 In summary, it is plausible biologically that both cold and hot weather may have the potential to increase the risk of OHCD.
We found that the RRs of OHCD associated with extreme temperatures were higher in the elderly, males and persons with low educational attainment. The vulnerability for elders may be related to their fragility to external stresses such as extreme temperatures. SCDs or out-of-hospital arrests were reported to be more common in males than in females.15 ,16 Likewise, we found that OHCD occurred more frequently among males than females, but the differences were not significant; however, the reasons for this sex difference remain unclear. Our results show that a low educational level in coronary patients was associated with a higher probability of dying outside of the hospital, a finding consistent with the results of a national longitudinal study.31 Education level is a main indicator of socioeconomic status, which has been associated with a range of physical and mental health problems.
Our findings may have important public health significance. Knowledge of the incidence and risk factors of SCD in China is scarce. Data on SCD in general are rare in China. Preliminary data suggest that the annual incidence rate of SCD in China is 41.8/100 000 population (lower than that reported in the USA) and ranges from 53 to 117 out of 100 000/year.3 This article contributed knowledge of the risk factors of SCD to the scientific literature. Subsequently, public health efforts can be designed for the primary prevention of SCD by minimising exposure to cold or hot weather conditions. In the presence of extreme temperatures, patients of CHD, especially if they are old, male, less educated, can be taught to stay in an environment with a comfortable temperature (eg, increased use of air conditioning) and wear more or fewer articles of clothing as relevant; the health sectors should also be well prepared to provide prompt and good care for those experiencing a cardiac arrest when cold or hot temperatures are forecasted. If adequate preventive measures were taken, a large public health benefit could be expected.
The limitations of our study should also be noted. First, given that people spend most of their time indoors, the real association between extreme temperature and OHCD should be investigated further. Second, the study period (3 years for each city) was relatively short. Third, the death certificates were completed by different doctors, and thus diagnosis error may occur in such large multicity studies. Fourth, we failed to analyse the modifying effects of pre-existing cardiac diseases because of the data unavailability. Sixth, residual confounding may occur because ozone was not controlled in the models, especially when the effects of high temperature were investigated.
This multicity epidemiological study suggested that both extreme cold and extreme hot temperatures posed significant risks on OHCDs, especially for those patients with CHD who were old, male, less educated. Our findings might have important public health implications for the prevention of OHCD or SCD.
What is already known on this subject
Out-of-hospital coronary death is often used as a surrogate for sudden cardiac death, and its primary prevention remains a major public health challenge.
Few studies have examined outdoor temperature as a potential risk factor.
What this study adds
Both extreme cold and extreme hot temperatures posed significant risks on out-of-hospital coronary deaths (OHCDs) in six large Chinese cities.
The coronary patients who are old, male, less educated might be susceptible.
These results might have important public health implications for the prevention of OHCDs or sudden cardiac death.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online supplement
RC, TL and JC contributed equally.
Contributors RC, TL and JC performed the statistical analysis and drafted the paper. MY collected and organised the data. ZZ and HK revised the manuscript. HK designed and supervised the study.
Funding The study was supported by the National Basic Research Program (973 program) of China (2011CB503802), China Medical Board Collaborating Program (13–152), the Gong-Yi Program of the China Ministry of Environmental Protection (201209008) and National Natural Science Foundation of China (81222036).
Competing interests None.
Ethics approval Institutional Review Board of School of Public Health, Fudan University.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.