Epidemiological studies have demonstrated a significant association between ambient temperature and mortality.1 2 This relationship was generally found to be V-shaped, U-shaped or J-shaped, with increased mortality at high and low temperatures and minimum mortality occurring at various temperature points.3–8 Exposure to high temperature may cause dehydration and increase blood viscosity, which can result in some adverse health outcomes, for some vulnerable subpopulations, it is likely to lead to excess cardiovascular and respiratory deaths;9 while during the cold period, the ability to regulate body temperatures is reduced, the cardiorespiratory system cannot adjust well to the outside temperature change, especially for those persons with pre-existing cardiovascular and respiratory diseases.10

The health impacts of temperature variation are likely to be heterogeneous across regions with different geographic conditions, climate and population characteristics.11 12 Identification of factors affecting the health susceptibility to temperature variation has been recognised as an important step to protect the vulnerable population.13 Some studies have also reported that some individual-level factors were associated with the health effects of temperature variation, such as age, sex and socioeconomic status.3 13–19

Being the largest developing country, China has experienced noticeable climate change over the past decades.20 The annual average temperature has risen by 0.5–0.8°C during the past century.21 Meanwhile, some extreme weather events, such as heat waves and cold spells, have also been witnessed in China, resulting in substantially excessive mortalities.10 22 23 However, few studies have examined how temperature–mortality relationships differ among different subpopulations and regions in China.24

One of our recent analyses examined the health effects of heat waves and associated effect modifiers based on a national database comprising data from 66 Chinese communities;25 the present study aimed to examine the effect modifiers of temperature–mortality relationship in China. We evaluated whether individual characteristics (age, sex, place of death, education level and cause of death) and community-level factors (average temperature, population density, sex ratio, percentage of older population, health access, average annual income and latitude) could modify the temperature–mortality relationship. This is one of the largest studies of temperature–mortality relationship to date in China.

Online supplementary material table S1 summarised the daily mortality and weather statistics by community. During the study period, a total of 1 260 913 deaths were reported in the 66 communities, including 552 866 and 198 777 deaths from cardiovascular and respiratory systems. The communities differed in population size and daily mortality count. The community-specific mean number of daily deaths ranged from 2 to 23. Most of the deceased were above 65 years of age (71.5%), and male (58.2%). A substantial fraction of deaths (72.8%) occurred out of a hospital.

Online supplementary material table S1 and figure S1 illustrated the distribution of daily temperature in the 66 communities. The 66 communities in this study had a wide range of climates, with annual mean temperatures ranging from 4.6°C in Yi'an County in Northeast China (community ID, 230223) to 22.9°C in Hepu County in the Southeast Region (community ID, 450521). Among the seven regions, the Southern region had the highest mean temperature, and the Northeastern region had the lowest temperature.

Online supplementary material figures S2 and S3 showed the diagnostic graphs of the model, including the plot of the residuals and the plot of PACF for a few communities. There were no discernible seasonal patterns and no autocorrelation in the residuals, indicating acceptable goodness of fit of the models.

Online supplementary material figure S4 showed the temperature-mortality association and the identified MMT in a few communities. Online supplementary material figure S5–S8 showed the association for males, females, cardiovascular and respiratory mortalities, respectively. The results showed that the curves varied across communities, but generally was U-shaped or V-shaped. We found significant mortality effects from low and high temperatures, with a most comfortable temperature value; in general, the MMT was found to be higher in the southern regions. The overall effect estimates for all the 66 communities and each of the seven regions were illustrated in table 1. The strongest cold effect was observed for the North and Central regions, and the Northeast Region had the lowest mortality effect from cold weather. For the mortality effect of high temperature, the highest effect was observed in South China, and the lowest effect in the Northwest Region.

In this large multicommunity study, we provided unique evidence on individual-specific and community-specific indicators that were related to the vulnerability to temperature-related mortality risk in China. Our analysis found that most of the effect modifiers were generally similar for individual-level and community-level factors, which enhanced the possibility that these factors were valid potential markers of vulnerability, which contributed to the advantage of this study. The underlying mechanisms for the observed excess mortality during days with high temperature may be related to the stress placed on the respiratory and circulatory systems to increase heat loss through skin surface blood circulation.36 This, coupled with an increase in blood viscosity and cholesterol levels, may lead to elevated mortality risk. The possible mechanisms for the cold-related mortality risk have also been suggested. One underlying reason might be the higher prevalence of respiratory infection during the winter season, particularly influenza epidemics and an increase in plasma cholesterol and plasma fibrinogen with low temperatures, coupled with a higher blood pressure in cold weather, could lead to thrombosis through haemoconcentration and trigger an acute mortality event.37 Moreover, the extremely cold weather might have also contributed to the excess mortality by reducing access to a hospital.

Socioeconomic inequalities at community and individual levels have been linked with a higher burden of environmental risks, including those from weather variation.3 13 38 39 For example, higher mortality effects of high temperature were observed for those at an older age, dying out of a hospital and with existing cardiopulmonary conditions.13 Regional differences in the association between ambient temperature and mortality may be partly due to the dissimilar population structure in terms of age, sex, education attainment and other socioeconomic status.3

Our study found that low socioeconomic conditions were generally associated with higher mortality risk from high and low temperatures, which was generally consistent with several previous studies. For example, a study about the heat-related mortality risk in Chicago in 1995 found that people with existing medical conditions and low socioeconomic status were the vulnerable population.40 In Texas, heat-related excess mortality was higher among the elderly and persons involved in heavy labour.41 An analysis examining the effect indicators of the temperature-mortality association in 107 US cities found that age, cause of death, place of death, median household income, unemployment rate, usage of air conditioning and education level were potential markers of vulnerability to temperature extremes.3 Similar findings have also been observed in other studies.14 17 42 43

Our study suggested that the temperature-mortality association increased with age, with those aged over 85 years being the most susceptible population, which was consistent with several previous studies.44 45 The older persons were generally at poor health status, had pre-existing medical conditions and diminished ability to detect changes in their body temperature, and thus had reduced thermoregulatory capacity, which might be the major reason for their increased susceptibility to temperature extremes.14 22 46

Education level was a good indicator of socioeconomic status that conveyed susceptibility.42 For cold and heat effects, our study found a generally higher mortality risk for areas with a lower education level, which was consistent with a study of 7 US cities14 47 and Curriero et al's45 study of 11 US cities. Persons with more educational attainment were more likely to have greater wealth and better health access, dietary intake, better domestic heating system in winter and air conditioning usage in summer.48 Higher mortality rates among the low-educated population have also been observed in several other studies, such as a cohort study in the USA.49

Location of death was found to be an important effect modifier of the mortality effect of cold and hot weathers in this study, which has also been reported in a few studies;13 14 however, these two studies13 14 found that the effect modification was mainly for heat effects, while our study found that it was more obvious for cold effects; the underlying reason for this discrepancy remained unknown. The finding of higher vulnerability for those dying out of hospitals was biologically plausible. It was possible that people dying out of a hospital had higher intensity of exposure to an ambient environment, and it was also partly due to the better healthcare received by the residents living in a hospital, as it was likely that those dying out of a hospital usually had lower socioeconomic status.14

The factors leading to the community vulnerability to mortality effects of temperature were found to be community-specific temperature, urbanisation, education level, healthcare facility, household income and latitude and the effect modification was particularly noteworthy for cold effects. Previous studies also found that group-level factors could explain some heterogeneity in community-specific effects.3 45 50 51 This phenomenon may be due to the general health and nutrition status, healthcare access, and overall knowledge, attitude and adaptation capacity to the health effects of temperature. A study examining the temperature-related mortality in Chicago also showed that areas with lower socioeconomic status had higher mortality risk.50 Anderson et al's3 study including 107 US communities also examined the community-specific effect modifiers, and found that communities with a higher unemployment rate, more population with lower education attainment, and more population living in an urban environment were more vulnerable to the temperature extremes.

This study found that southern communities with lower latitude and higher annual average temperature had temperature-related mortality risk, which has also been reported elsewhere.4 16 45 The higher cold effect in southern areas may suggest the evidence for acclimatisation. Though the winter temperature was higher in south China than in northern areas, the lack of a centralised heating system in south China made people in southern areas suffer more from the cold weather, causing excess mortality risk,22 while higher heat effect in south region observed in this study might be due to more exposure opportunities and higher intensity of heat exposure by the population in south China. However, some other studies have reported that low latitude areas had lower risk from heat effects, which was inconsistent with findings from our study.44 52

Though previous studies suggested that hot weather is an important health hazard in the context of global climate change,53 our study suggested that cold weather was also an important health hazard, which should not be neglected. This finding also indicated that the health impact of cold weather, particularly its long delayed effects, should be considered in the health effect analysis and projecting health impact of future climate change.

A few limitations should be noted when interpreting the findings from this study. One limitation was that we were not able to control for air pollution and influenza epidemics in the analysis due to data unavailability. Our previous work suggested that the health effects of temperature extremes remained similar after controlling for various air pollutants and influenza epidemics.22 However, O'Neill et al54 examined the mortality impact of temperature controlling for ozone and particulate matter and found that while the effect of high temperature on mortality reduced by about 30%, the associations persisted. Goldberg's study in Canada also suggested that air pollution could not distort the temperature–mortality relationship to a great extent.55 Another limitation was the use of ambient temperature as a surrogate for personal exposure. The misclassification derived from this assumption will be largely determined by the extent to which ambient and microenvironmental temperatures are correlated,16 which will most likely be affected by adaptive mechanisms such as the use of air conditioning. Furthermore, for both individual-level or community-level analyses of effect modifiers, we cannot control the potential confounding effect of one modifier by another. Owing to the high correlation between some variables, we cannot simply put them in the same model, making the observed effect modification difficult to attribute to one specific factor. It should also be noted that small number of daily mortality in some communities might lead to imprecise risk estimation, and for this reason we only examined the effects on mortality from respiratory and cardiovascular systems; further studies on more specific diseases are warranted in future studies. Another limitation was that this study used weather data from only one weather observation station for each community; however, a community in this study covered a relatively small geographic area, and the weather data from one station could generally represent the local weather pattern.

The findings from this study have some public health implications for policymakers and the scientific community by providing evidence of the temperature–mortality relationship in China and related effect modifiers at the individual and community levels. Climate adaptation planning should be taken as one local issue, which means adaptation strategy development should take into consideration local weather conditions, population characteristics and socioeconomic status. The identified susceptible subpopulations from this study provided evidence for the targeted temperature-mortality prevention and adaptation measures. The community-level effect modifiers indicate the necessity of a multicity study and indicate that community-specific prevention strategies are warranted.

In summary, this study identifies the susceptibility of mortality effects of temperatures based on both individual-level and community-level effect modifiers; more attention should be given to these vulnerable individuals and vulnerable communities to reduce adverse health effects of extreme temperatures.