Study area
The study area is Tianjin, the third largest municipality in China. It has a population of approximately 12.9 million, with the elderly (age ≥ 65 years) accounting for approximately 8.52% at baseline (Tianjin Statistical Information Site, 2006–2011, http://stats.tj.gov.cn/Category_29/Index.aspx). Tianjin has a typical temperate climate characterized by hot, rainy summers and cold, dry winters.
Data collection
Daily mortality data on IHD in the elderly (age ≥ 65 years) from Jan 1st 2006 to Dec 31st 2011 in Tianjin were collected from Death Registration and Reporting System of the Chinese Centre for Disease Control and Prevention. The data were include all the residents of Tianjin, and the mortality data were representative of the study area. The causes of IHD were classified and coded according to the International Classification of Disease, 10th version (IHD: I20-I25). Data permission was obtained and the study was approved by the Ethics Committee/Institutional Review Board of Peking University Health Science Center.
Indicator of YLL was used in this study. The method to calculate YLL was used in previous studies as the follow equation [19, 20].
$$ \mathrm{YLL}=\sum Yi\times Li $$
where Yi is the death number for a specific age group i, and Li is the remaining life expectancy for a specific age group i.
First, we matched each person’s age at death to the World Health Organization (WHO) standard life table (Additional file 1: Table S1). Then, the daily YLL of the elderly from IHD in our study area was estimated by summing the YLL of all individuals who aged ≥65 years and died from IHD on the same day.
Daily meteorological data, including the daily relative humidity and maximum temperature, were obtained from the Tianjin Meteorological Bureau, which were match with the daily mortality data of IHD during the study period from 2006 to 2011. The average daily concentrations of particulate matter with aerodynamic diameter ≤ 10 μm (PM10) were also collected from the Tianjin Environmental Monitoring Centre to allow for the evaluation of the confounding effects of air pollutants.
Statistical analyses
First, the baseline temperature-YLL relationships from 2006 to 2011 were established, then future assessments were made in combination with future temperature projections.
Distributed lag non-linear models (DLNMs) were used to measure the non-linear and delayed effects of temperature on YLL from IHD during the baseline period [21]. Seasonal and long-term trend were adjusted using a natural cubic spline function with 7 degrees per year, and day of the week as a categorical variable. The daily PM10 and relative humidity were adjusted using a natural cubic spline with 3 degrees of freedom. A natural cubic B-spline basis with 5 degrees of freedom for temperature and a maximum lag of 15 days between temperature and YLL with 5 degrees of freedom were chosen in the baseline analysis. Optimal temperature (OT) was used as a reference to calculate the temperature-related YLL related to high or low temperatures. The OT was determined according to the exposure-response curve for temperature in YLL for IHD during the baseline period.
Temperature projections were made according to Representative Concentration Pathways (RCPs) reported in the IPCC AR5. Three RCPs scenarios were selected in our analysis, including RCP2.6, RCP4.5 and RCP8.5, which represent the mild, the medium and the high emission scenarios. The future daily temperatures for periods of 2046 to 2065 centred on 2050s, and 2061 to 2080 centred on 2070s, were developed from 19 global-scale climate models (GCMs) from the World Climate Research Programme (WCRP) Coupled Model Intercomparison Project Phase 5 (CMIP 5) multi-model dataset (Additional file 1: Table S2).
As for the temperature calibration, we used method as follows: it started with the projected change in a weather variable (i.e. maximum temperature in June). This is computed as the (absolute or relative) difference between the output of the GCM run for the baseline years and for the target years (e.g. 2050s–2070s). These changes are then added to the observed baseline to create the projected temperature in a given year for a specific emission scenario (in this case the WorldClim Database, http://www.worldclim.org/downscaling). Through this method, we could capture the trend of temperature change in future projection. And this approach was also used in previous study [22].
The projection of temperature-related YLL from IHD in the elderly was calculated by integrating the temperature projections under different RCPs with baseline exposure-response relationships. The projections and changes of annual heat-related, cold-related and total temperature-related YLL were calculated.
Furthermore, adaptation of the population was taken into consideration because people could adapt to warmer climatic conditions through a number of measures [23]. A 25% acclimatization factor was assumed according to the previous study of a U.S. city, which reported that excess mortality related with heat reduced by approximately 25%, indicating population adaptation to heat in recent decades through increased using of air conditioning, greater awareness of the risks by high temperature, and introduction of heat-warning systems, etc. [24].
Considering the temperature-YLL relationship may not remain stable over time due to population adaptation, we modelled the adaptation by shifting the OT and shape of temperature-YLL curves [25]. The model combining the absolute threshold shift of OT with the reduction in the slope of the heat exposure-response function was used in this study because it may balance the uncertainty among adaptation models, climate models and emissions [26]. Previous studies have suggested that OT could continue to rise with increasing temperature due to adaptation [26, 27]. Absolute threshold shift of OT is a popular method for modeling adaptation [26], and applying a shift in absolute threshold of OT between 1.0 °C and 4.0 °C in future is recommended because this is broadly within the range of shifts in threshold temperature observed in previous epidemiological studies [27,28,29]. Thus, we conservatively estimated that the OT will increase by 1.0 °C in 2050s, and by 1.2 °C in 2070s relative to the baseline [30].
Demographic change was also taken into account. Based on the low, medium and high variant scenarios of population growth among the population aged 65 years and above in China employed by the UN, this population size will be 3.1 and 3.2 times greater than the baseline population in 2010 by the 2050s and 2070s, respectively [12].
Sensitivity analyses were performed to test whether the results were robust to changes in the parameters in the model, including using 4 degrees of freedom of relative humidity, and a natural cubic B-spline with 5 degrees of freedom for temperature and a maximum lag of 15 days between temperature and YLL with 6 degrees of freedom.
R software was used to perform all analyses (version 3.3.3, http://www.R-project.org/).