Heat-related mortality risk model for climate change impact projection
© The Japanese Society for Hygiene 2013
Received: 31 January 2013
Accepted: 11 July 2013
Published: 9 August 2013
We previously developed a model for projection of heat-related mortality attributable to climate change. The objective of this paper is to improve the fit and precision of and examine the robustness of the model.
We obtained daily data for number of deaths and maximum temperature from respective governmental organizations of Japan, Korea, Taiwan, the USA, and European countries. For future projection, we used the Bergen climate model 2 (BCM2) general circulation model, the Special Report on Emissions Scenarios (SRES) A1B socioeconomic scenario, and the mortality projection for the 65+-year-old age group developed by the World Health Organization (WHO). The heat-related excess mortality was defined as follows: The temperature–mortality relation forms a V-shaped curve, and the temperature at which mortality becomes lowest is called the optimum temperature (OT). The difference in mortality between the OT and a temperature beyond the OT is the excess mortality. To develop the model for projection, we used Japanese 47-prefecture data from 1972 to 2008. Using a distributed lag nonlinear model (two-dimensional nonparametric regression of temperature and its lag effect), we included the lag effect of temperature up to 15 days, and created a risk function curve on which the projection is based. As an example, we perform a future projection using the above-mentioned risk function. In the projection, we used 1961–1990 temperature as the baseline, and temperatures in the 2030s and 2050s were projected using the BCM2 global circulation model, SRES A1B scenario, and WHO-provided annual mortality. Here, we used the “counterfactual method” to evaluate the climate change impact; For example, baseline temperature and 2030 mortality were used to determine the baseline excess, and compared with the 2030 excess, for which we used 2030 temperature and 2030 mortality. In terms of adaptation to warmer climate, we assumed 0 % adaptation when the OT as of the current climate is used and 100 % adaptation when the OT as of the future climate is used. The midpoint of the OTs of the two types of adaptation was set to be the OT for 50 % adaptation.
We calculated heat-related excess mortality for 2030 and 2050.
Our new model is considered to be better fit, and more precise and robust compared with the previous model.
In 2002, the World Health Organization reported for the first time the health impact of climate change . In the report, however, heat-related impact was not included in the final aggregate total number, because it was not easy to model the relationship between ambient temperature and mortality for projecting future impact of climate change.
Although the above projection was one of the first projections of climate change impact on heat-related mortality, there were some weaknesses to the method, including (1) the applicability of the OT estimation method to other regions of the world, and (2) that the risk function was categorical. In this paper, we solve problem (2) by using a nonparametric risk function. This should be a substantial improvement, because the average of the temperatures in the highest temperature category was actually very close to the lower boundary of the category and huge underestimation was expected for the risk of this category. Regarding problem (1), we examine in this paper the applicability using cities in Korea, Taiwan, Europe, and the USA as well as 47 Japanese prefectures.
Another concern is the temporal pattern of exposure–mortality effects. This can be regarded in two ways . One is the carry-over effect (lag effect), and the other is mortality displacement or harvesting. The former effect has been considered not to last long for heat-related mortality. In the latter case, heat just kills people who would die in a very short period of time anyway, and avoiding heat would extend the lives of people for only a short period of time. In the case of the 2003 heat wave in Europe, however, there was little evidence of mortality displacement . Here, we address this lag effect using a distributed lag nonlinear model developed by Armstrong .
Materials and methods
Description of mortality data by country
47 prefectures, 1973–2008 for Okinawa
A set of mortality projections for the 2030s and 2050s was developed for this project by the World Health Organization. The approach built upon previous methods developed by Mathers and Loncar . The method uses a series of regression equations that quantify the current and historical relationships between mortality and a set of independent variables. The major independent variables that were shown to be structurally related to mortality were (i) gross domestic product (GDP)/capita, (ii) years of education, and (iii) time, which is assumed to be a proxy for health benefits arising from technological developments. In addition, specific assumptions are made regarding future patterns of human immunodeficiency virus (HIV)/acquired immunodeficiency syndrome (AIDS), tuberculosis, malaria, smoking, and body mass index (BMI).
Development of the new risk function
We developed the risk function based on all 47 prefectures in Japan. Then, we examined the applicability of the assumptions in combining these 47 prefectures described below. For this purpose, we used the data for the selected cities in Korea, Taiwan, the USA, and Europe. We did not include data other than from Japan, because, although these cities somewhat represent large cities in each region, the data are not systematically collected and are not suitable for calculating average values.
Applicability of OT estimation method
OT, 84th percentile of T max, and ratio of RMOT over RMav
84th percentile of T max
To improve precision, use of long-term data and combining area-specific data are effective, provided that the chronological trend is well controlled and the heterogeneity among areas is reasonably small.
To control for chronological trend and population size, we introduced relative risk. The reference mortality for each year is the average number of deaths during the days with daily maximum temperature from the 75th to 85th percentile for each year. We chose this reference because this reference temperature range usually covers the OT, and because this reference is less affected by influenza epidemic than the annual average daily number of deaths or adding time trend in regression models as a nonparametric term.
Lag effect inclusion
As explained earlier, lag effect should be controlled for to obtain the overall heat-related effect. For this purpose, we used a distributed lag nonlinear model . In the actual calculation, we used the dlnm package developed by Gasparrini  in R 2.15.1 . In their paper, they describe dlnm as “a modelling framework that can simultaneously represent non-linear exposure–response dependencies and delayed effects. This methodology is based on the definition of a ‘cross-basis’, a bi-dimensional space of functions that describes simultaneously the shape of the relationship along both the space of the predictor and the lag dimension of its occurrence.” In our model, we used OT-offset daily maximum temperature (expressed as “T max − OT”) as the temperature index. Another choice would be the percentile value of daily maximum temperature, because most prefectures have OT close to the 84th percentile value. However, the range of daily maximum temperature is very wide in northern prefectures, whereas that in southern prefectures is narrow, so a one-unit increase of the percentile corresponds to a different increase in temperature.
From relative risk to excess number of deaths
To examine the robustness of the above index, we used selected data from cities in Korea, Taiwan, the USA, and Europe. Table 2 presents the values of this ratio in these countries selected for this manuscript. Although cities in Korea, Taiwan, and the USA showed a higher ratio than Japanese prefectures, two European countries showed a lower ratio. Considering that the European data are for all ages combined, the ratios for the 65+-year-old age group is expected to be even lower, because <65-year-old age groups are usually more resistant to heat; heat-related relative risk is lower for younger age groups. Based on this observation, we decided to use the Japanese ratio, 0.88, for global projection.
Projection of future impact of climate change on heat-related mortality
We need the daily maximum temperature distribution for risk estimation. So, we chose NCC (= NCEP Corrected by CRU, where NCEP stands for National Centers for Environmental Prediction, USA, and CRU stands for Climate Research Unit, University of East Anglia) daily data  as the baseline climate data. For the future climates, we used one of the general circulation models provided by the WHO Global Burden of Diseases project, i.e., BCM2. Here, we added the monthly average difference between the current climate and the future climate from models to each of the baseline daily maximum temperatures within the same month. The grid resolution is 1°. As the socioeconomic scenario, we used SRES A1B.
Adaptation is another concern. When the OT is estimated as the 83.6th percentile value of daily maximum temperature as of the current climate, we assume no adaptation (0 % adaptation). If the future climate is used for OT estimation, we assume 100 % adaptation. The midpoint of the 0 and 100 % adaptation OTs is assumed to be the OT for 50 % adaptation. V-shape flattening is another possible type of adaptation, but our preliminary analyses showed no evidence of flattening (this will be published in another paper), and we did not include this type of adaptation in this manuscript.
Asia Pacific, High Income
Latin America, Andean
Latin America, Central
Latin America, Southern
Latin America, Tropical
North America, High Income
North Africa/Middle East
Sub-Saharan Africa, Central
Sub-Saharan Africa, East
Sub-Saharan Africa, Southern
Sub-Saharan Africa, West
Projection of heat-related excess deaths
In some regions, the heat-related excess deaths constitute up to 0.6 % of deaths, and some high-income regions also suffer a 0.1–0.4 % excess.
Because this manuscript aims at developing a better projection model for heat-related excess deaths, the problems and their solutions were already addressed in “Materials and methods” section. Here, we evaluate how our goal was achieved, i.e., how much the fit, precision, and robustness of the proposed model have been improved from the previous projection . As shown in Figs. 2 and 5, it is obvious that the fit is much better for the present model. To show the improvement of precision, we used the present model and applied it to the observation period of the previous model, i.e., 1972–1995. As an example, when the temperature was 10 °C higher than OT, the risk estimate (95 % confidence interval) of this calculation was 1.241 (1.218–1.265), whereas the present calculation gave 1.173 (1.156–1.189); the range of the confidence interval was 30 % smaller. As for the robustness, Table 2 clearly shows the similarity of the OT and the ratio of RMOT over RMav across a wide range of areas, including Asia, the USA, and Europe. Hence, our model that sets OT to be around the 84th percentile value of daily maximum temperature can be generalizable to all over the world. One additional concern may be the applicability of our OT estimation to tropical areas, where the annual temperature difference is very small. Our analyses covered very cold areas to subtropical areas, but did not cover tropical areas or countries in the Southern Hemisphere. In this regard, McMichael and coworkers’ report  is worth mentioning, because they observed the heat-related effect for 12 cities including cities in tropical countries and those in the Southern Hemisphere. Their model used a different method from ours; For example, they used daily mean temperature instead of daily maximum temperature, and they averaged out lag effect for lag 0–1 day and for lag 0–13 days. Still, the relation of the 0–1 averaged lag model was V-shaped for the majority of the cities. Considering that the 84th percentile value is roughly equal to the summer mean temperature, we tried to identify summer average temperature and optimum temperature from the graphs they provided. Among the cities with a V-shaped relation, most of the cities, including a tropical city (Bangkok, Thailand), had optimum temperature similar to their respective summer average temperature. Although this procedure is not accurate, at least use of the 84th percentile value in predicting OT cannot be falsified even by observations of cities in tropical zones or in the Southern Hemisphere. Another supporting report is the 4th Assessment Report of the Intergovernmental Panel on Climate Change ; temperature rise in tropical areas is expected to be less prominent compared with temperate areas and high-latitude areas. Hence, our projection would not be wide of the mark.
Compared with our previous projection, this new projection has advantages in terms of fit, precision, and robustness. Also, we included lag effect, which addresses mortality displacement. This implies that, although there may be some mortality displacement effect, the overall risk exists and climate change would increase the burden due to heat-related mortality.
Because the purpose of this paper is to show how we have improved the model for the global projection, we showed the projection results using only one global circulation model. However, the model described in this paper was adopted in the new WHO Global Burden of Diseases attributable to climate change, which will appear somewhere soon, and several models will be used and compared.
Compared with the total number of deaths, the proportion of deaths due to heat-related mortality attributable to climate change appears small. However, considering that even huge cyclones do not kill many people nowadays in high-income countries, heat waves are a major environmental hazard. Of course, we need to compare the heat-related impact and cold-related impact to obtain the net impact of global warming. However, new papers have warned that we should not naively apply the V-shaped temperature–mortality relation to projections for future cold-related excess deaths [17, 18]. We will try to develop our model for cold-related excess mortality after taking into account the points these papers raised in the near future.
This study was supported by the Environment Research and Technology Development Fund (S-8 & S10) of the Ministry of the Environment, Japan, and the Global Research Laboratory (grant K21004000001-10A0500-00710) through the National Research Foundation of Korea (NRF), funded by the Ministry of Education, Science, and Technology, Korea.
Conflict of interest
All authors have no conflicts of interest.
- WHO. The World Health Report 2002-reducing risks, promoting healthy life. Geneva: WHO; 2002.Google Scholar
- Kunst AE, Looman CW, Mackenbach JP. Outdoor air temperature and mortality in the Netherlands: a time-series analysis. Am J Epidemiol. 1993;137:331–41.PubMedGoogle Scholar
- El-Zein A, Tewtel-Salem M, Nehme G. A time-series analysis of mortality and air temperature in Greater Beirut. Sci Total Environ. 2004;330:71–80.PubMedView ArticleGoogle Scholar
- Honda Y, Ono M, Sasaki A, Uchiyama I. Shift of the short-term temperature mortality relationship by a climate factor—some evidence necessary to take account of in estimating the health effect of global warming. J Risk Res. 1998;1:209–20.View ArticleGoogle Scholar
- Curriero FC, Heiner KS, Samet JM, Zeger SL, Strug L, Patz J a. Temperature and mortality in 11 cities of the eastern United States. Am J Epidemiol. 2002;155:80–7.PubMedView ArticleGoogle Scholar
- Honda Y, Kabuto M, Ono M, Uchiyama I. Determination of optimum daily maximum temperature using climate data. Environ Health Prev Med. 2007;12:209–16.PubMed CentralPubMedView ArticleGoogle Scholar
- Takahashi K, Honda Y, Emori S. Assessing mortality risk from heat stress due to global warming. J Risk Res. 2007;10:339–54.View ArticleGoogle Scholar
- Armstrong B. Models for the relationship between ambient temperature and daily mortality. Epidemiology. 2006;17:624–31.PubMedView ArticleGoogle Scholar
- Fouillet A, Rey G, Laurent F, Pavillon G, Bellec S, Guihenneuc-Jouyaux C, et al. Excess mortality related to the August 2003 heat wave in France. Int Arch Occup Environ Health. 2006;80:16–24.PubMed CentralPubMedView ArticleGoogle Scholar
- Bell ML, Dominici F, Samet JM. A meta-analysis of time-series studies of ozone and mortality with comparison to the national morbidity, mortality, and air pollution study. Epidemiology. 2005;16:436–45.PubMed CentralPubMedView ArticleGoogle Scholar
- Mathers CD, Loncar D. Projections of global mortality and burden of disease from 2002 to 2030. PLoS Med. 2006;3:e442.PubMed CentralPubMedView ArticleGoogle Scholar
- Gasparrini A, Armstrong B, Kenward MG. Distributed lag non-linear models. Stat Med. 2010;29:2224–34.PubMed CentralPubMedView ArticleGoogle Scholar
- Ihaka R, Gentleman RR. A language for data analysis and graphics. J Comput Graph Stat. 1996;5:299–314.Google Scholar
- Ngo-Duc T. A 53-year forcing data set for land surface models. J Geophys Res. 2005;110:D06116.Google Scholar
- McMichael AJ, Wilkinson P, Kovats RS, Pattenden S, Hajat S, Armstrong B, et al. International study of temperature, heat and urban mortality: the “ISOTHURM” project. Int J Epidemiol. 2008;37:1121–31.PubMedView ArticleGoogle Scholar
- Solomon S, Qin D, Manning M, Chen Z, Marquis M, Ave KB. IPCC. Summary for Policymakers. In: Climate Change 2007. The physical science basis contribution of working group I to the fourth assessment report of the intergovernmental panel on climate change. Cambridge, UK and New York, USA: Cambridge University Press; 2007. p 18.Google Scholar
- Kinney PL, Pascal M, Vautard R, Laaidi K. Winter mortality in a changing climate: will it go down? Bull Epidémiol Hebd. 2012;12–13:148–51.Google Scholar
- Ebi KL, Mills D. Winter mortality in a warming climate: a reassessment. WIRES Climate Change. 2013;4:203–12.View ArticleGoogle Scholar