Projecting Future Heat-Related Mortality under Climate Change Scenarios: A Systematic Review

1Exposure to extreme heat has been associated with both increased mortality and morbidity. A number of epidemiological studies have examined high temperatures in relation to total nonaccidental deaths, to cause-specific mortality, and to other health outcomes such as emergency department visits and hospitalizations. In fact, heat waves are the biggest cause of weather-related fatalities in many cities, responsible for more deaths annually than any other form of extreme weather [; World Health Organization (WHO) 2009a].

2Recently, heat-related mortality has become a matter of growing public health concern, especially because of climate change. The Intergovernmental Panel on Climate Change (IPCC) indicates that hot weather is likely to increase future heat-related mortality. Urban areas, home to more than half of the world’s population, can be particularly vulnerable to heat because of high concentrations of susceptible people, the urban heat island effect, poor urban design and planning, and the interaction between air pollution and heat.

3People with cardiovascular or respiratory disease, diabetes, chronic mental disorders, or other preexisting medical conditions are at greater risk from heat exposure. The effects of heat are particularly strong in the elderly. Other factors that influence the risks of heat-related mortality include social isolation, low income, low education, poor housing, lower access to air conditioning, and less availability of health care services.

4Many study designs have been used to examine the effects of temperature on mortality, including descriptive, case–control, case-only, case-crossover, time-series, spatial, and synoptic analyses. Generally, time-series and case-crossover are considered more efficient designs for examining the temperature–mortality relationships in single or multiple locations over time. These designs aim to investigate the health effects of temperature after controlling for potential confounders such as trends and seasonal cycles in mortality and, in some cases, humidity and air pollution.

5The health effects of heat can be estimated using the heat threshold and the heat slope. The temperature–mortality relationship is usually a nonlinear U-, V-, or J-shape. Many studies have quantified cold and heat effects separately, assuming a linear response below and above a threshold temperature. The heat threshold is the temperature at which the harmful effect of heat begins to occur, and the heat slope measures the size of this effect. A significant geographic variability has been observed in both heat thresholds and slopes. Heat thresholds tend to be higher in warmer locations, suggesting acclimatization.

6Many studies have found associations between high temperatures and mortality, but more research is needed on the impacts of climate change on future heat-related mortality. Scenario-based projections have been used as a key approach for policy making and planning in the context of uncertain future conditions. The IPCC has developed a set of scenarios in its Special Report on Emissions Scenarios (SRES; ). These scenarios are not assigned probabilities but, rather, can be deemed as possible futures, which depend on demographic, technological, political, social, and economic developments. Scenarios are used not to better predict the future but to better understand uncertainties in order to reach decisions that are robust under a range of possible futures. Informed by quantitative or qualitative evidence, projections provide decision makers with information on a variety of future trends, contexts, risks, and opportunities. Projecting heat-related mortality under climate change scenarios will therefore help decision makers in planning adaptation strategies and communicating the future health risks of climate change to the public and politicians.

7Understanding and managing uncertainty and complexity are the greatest challenges for projecting future heat-related mortality. The major difficulties are the long timescale over which climate change will occur; the diversity of potential impacts on health; and the complex interactions among demographic changes, socioeconomic development, technological innovation, and other environmental drivers. Currently, there are no guidelines for scenario-based projection research on heat-related mortality, and only a few studies have examined this issue. This article aims to fill this knowledge gap using a systematic review and to make recommendations for future research.

8A literature search was conducted in August 2010 using the electronic databases PubMed (National Library of Medicine 2010), Scopus and ScienceDirect,, and Web of Science. The search was limited to journal articles published in English from January 1980 through July 2010. The key words used were heat, temperature, mortality, death, climate change, projection, and scenario. References and citations of the articles identified were inspected to ensure that all relevant articles were included.

9Three inclusion criteria were used to select articles. First, articles had to include at least one projection of future heat-related mortality; studies of the climate change impact solely on future infectious diseases or air pollution were excluded. Second, in order to obtain authoritative information, this review included only peer-reviewed journal articles; books, reports, and conference abstracts were excluded. Third, we included only quantitative, empirical studies; reviews and qualitative studies were excluded.

10We found 14 studies, including three that considered both the future impacts of heat and air pollution on mortality. These studies are summarized in Table 1, with the most recent studies listed first.

11Main findings. In a study of the possible health impacts of climate change in 44 U.S. cities, estimated that increases in heat-related mortality would range from 70% to > 100% in 2050, relative to the baseline 1964–1991 summer mortality. Winter mortality would drop slightly, but this would not offset the increases in summer mortality to any significant degree. Large increases in heat-related mortality have also been projected for other cities in the United States. Heat-related mortality in Los Angeles by the 2090s was projected to increase by two to three times under the lower B1 emissions scenario and by five to seven times under the higher A1FI emissions scenario relative to 1961–1990. In the New York City metropolitan region, projected increases in heat-related mortality ranged from 47% to 95% by the 2050s, with a mean 70% increase compared with the 1990s. Annual average mortality rates in Chicago by the end of this century were projected to be equal to those of 1995-like heat waves under the B1 emissions scenario and to reach twice the 1995 levels under the A1FI emissions scenario. In Washington State, the largest projected excess deaths for the Seattle region occurred in those ≥ 65 years of age. Holding the population projection constant at the 2025 level, and using a moderate warming scenario, this age group was projected to have 96, 148, and 266 excess deaths due to heat events in 2025, 2045, and 2085, respectively.

12In three cities in Canada, projected a significant increase in temperature-related mortality in summer that was not offset by a significant but smaller estimated decrease in fall and winter mortality. projected that heat-related mortality in four Canadian cities would more than double by the 2050s and triple by the 2080s, and that cold-related mortality could decrease by 45–60% by the 2050s and by 60–70% by the 2080s.

13In Australia, applied the climate–mortality relationship estimated for 1979–1990 to scenarios for climate and demographic changes to predict potential impacts on mortality in five major cities in 2030. They estimated that changes in mortality due to direct climatic effects would be small after considering reduced winter mortality. However, we feel that this is probably because of the relative insensitivity of warming to emissions scenarios, over the short period to 2030.

14projected that climate change in Europe would have the greatest impact on three Mediterranean cities (Barcelona, Rome, and Valencia) and two continental cities (Paris and Budapest). The largest estimated impact according to age was on persons > 75 years of age, but in some cities relatively large proportions of heat-related deaths were also projected for younger adults. projected that annual heat-related deaths in Lisbon would increase from 5.4–6.0 per 100,000 in 1980–1998 to 5.8–15.1 in the 2020s, and to 7.3–35.6 in the 2050s. projected global rates of excess deaths due to heat would increase by 100% to 1,000% in the 2090s relative to the 1990s. The authors projected that the burden of future heat-related mortality would vary by country, with greater effects in China, India, and Europe because of the high population density of these regions. Their estimates were based on World Bank population estimates for the year 2000 and assumed no change in future population density or age composition.

15It has been estimated that population acclimatization could reduce future heat-related mortality by 20–25% in Los Angeles, 25% in New York City, and 40% in south-central Canada. In Lisbon, acclimatization could reduce heat-related deaths on average by 15% in the 2020s and 40% in the 2050s, relative to projections assuming no acclimatization. estimated that acclimatization to an extra 2°C in maximum temperature would reduce future heat-related mortality by 50%. However, complete acclimatization to high temperatures is unlikely because it would require extensive improvements to existing buildings (particularly in poor areas), and because the capacity for acclimatization will be reduced in vulnerable populations such as patients with advanced heart disease.

16In summary, most projections showed that climate change would result in a substantial increase in heat-related mortality. Also, most studies did not consider demographic changes that are expected to result in an aging population (who are more susceptible to heat), which could lead to an underestimation of future heat-related mortality. Acclimatization could reduce future heat-related mortality, but it would not entirely eliminate the impacts of climate change on mortality. The studies identified by our review used many different climate models, emissions scenarios, time periods, temperature exposures, and assumptions. It is therefore difficult to compare the different studies using standardized results that are stratified by periods and locations. It is also not possible to conduct a meta-analysis to create a combined estimate of the impacts of climate change on heat-related mortality.

17Methodological issues. Projecting heat-related mortality under climate change models and scenarios requires understanding of the historical temperature–mortality relationships and consideration of the future changes in climate, population, and acclimatization. Estimation of the possible health consequences of climate change is inherently difficult and involves numerous uncertainties, as outlined below and summarized in Table 2.

18Baseline temperature–mortality relationships. Mortality projections are based on historical exposure–response functions of temperature and mortality that are applied to climate change models and emissions scenarios to estimate future heat-related mortality. Therefore, it is important to consider which measure of temperature is the best predictor of mortality.

19Maximum temperature and mean temperature are commonly used measures of heat exposure. For example, modeled the relationship between maximum temperature and excess deaths in Lisbon during the summer months in 1980–1998 and then applied different climate and population change scenarios to the model to assess potential impacts on mortality in the 2020s and 2050s. projected the impacts of climate change on summer mortality using modeled daily mean temperatures for New York City. reviewed the literature on the relationship between mean temperature and mortality and derived a pooled estimate of the effect of temperature on mortality using meta-analysis.

20Others have used composite indices, which examine the combined effects of ambient temperature, humidity, and other meteorological variables. For example, the synoptic approach is an air-mass–based method that quantifies the effect of air and dew point temperatures, wind speed, cloud cover, barometric pressure, and others. The apparent temperature and Humidex combine the effects of temperature and humidity. used the Spatial Synoptic Classification to evaluate climate–mortality relationships in the U.S. cities. described the climate–mortality relationships by temporal synoptic indices in Australia. identified maximum apparent temperature thresholds that were associated with rising heat-related mortality in California. examined the historical relationship between age- and cause-specific mortality rates and heat events at the 99th Humidex percentile in Washington State.

21examined which measure of temperature is the best predictor of mortality. The authors compared seven temperature measures: maximum, mean, and minimum temperature; maximum, mean, and minimum apparent temperature; and the Humidex, based on daily data for 107 U.S. cities during 1987–2000. No one temperature measure was superior to the others. The strong correlation between different temperature measures suggests that they have a similar predictive ability. For the projection research of heat-related morality, we therefore propose that the temperature measure can be chosen based on practical concerns, such as using mean temperature, which may be commonly available from the climate models.

22The choice of model is probably a more important consideration than the choice of temperature measure. compared the predictive capacity of four approaches for identifying dangerous hot days: physiological classification, synoptic approach, temperature–humidity index, and temperature–mortality relationship. There was little agreement across the different approaches, but in general, modeling the temperature–mortality relationship most accurately identified days of highest excess mortality. For projection research, we recommend examining the heat threshold and slope using a temperature–mortality relationship based on a continuous scale of temperature values.

23Choosing a baseline time period for the temperature–mortality relationship is also important. Temperature–mortality relationships in the same city can be very different between the 1960s and the 2000s. Differences could be due to socioeconomic development, demographic change, and population acclimatization. Differences in the time periods used to estimate the historical temperature–mortality relationships also make it difficult to compare projections across studies. Figure 1 highlights the variability of different time periods used in each study. Because daily mortality data often are not available before 1990 in many cities, the time period 1996–2005, which centered on 2000, is recommended as the baseline.

24Climate change projections. Another fundamental issue for projecting heat-related mortality is the modeling of future climate. The IPCC has defined a set of 40 SRES scenarios that covered a wide range of the main driving forces of future greenhouse gas emissions. These scenarios are structured in four major families labeled A1, A2, B1, and B2. A1 represents rapid economic growth, global population peaking in midcentury, and rapid introduction of new and efficient technologies. A1 has three subgroups: A1FI (fossil intensive), A1T (nonfossil), and A1B (balanced). A2 represents high population growth, slow economic development, and slow technological change. B1 represents the same population growth as A1 but rapid changes in economic structures toward a service and information economy. B2 represents intermediate population and economic growth with local solutions to economic, social, and environmental sustainability. The emissions scenarios can be used to project future climates based on various general circulation models (GCMs).

25Selecting climate models is also not a trivial task, given the strengths and weaknesses of various GCMs. There are different types of GCMs, depending on whether they incorporate dynamics from the atmosphere, the ocean, or both. However, although all GCMs attempt to accurately represent climate processes, this gives rise to different GCMs adopting different representations and hence generates different climate change projections, even when assuming the same pathway of future emissions. For more detailed information on emissions scenarios and GCMs, see.

26In the existing studies, considered two of the emissions scenarios, A2 and B2, which assume relatively high and low future emissions, respectively. They used both scenarios to model daily mean temperatures in the 1990s and 2050s. applied the same scenarios to model daily maximum temperatures during 1961–1990 and 2070–2099. projected future climates based on the higher A1FI and lower B1 emissions scenarios. selected three climate change scenarios for high (A1B), low (B1), and moderate (A1B and B1 combined) summer warming.

27Adding to the uncertainties of emissions scenarios, different GCMs were often used to simulate the future or current climates. used three GCMs in their study: the GFDL model, the UKMO model, and the Max Planck model. used the CSIRO-Mk2 to project regional monthly mean changes in temperature, rainfall, and other climates. Also, because the spatial resolution of GCM results is too coarse to be used directly in the impact assessment at a local scale, downscaling methods were introduced. used a statistical downscaling approach to downscale daily outputs from five GCMs for selected cities. used the dynamic downscaling approach, in which the GCM outputs were used as initial and boundary conditions for finer-scale simulations by the climate model. Daily climate projections from regional climate models (RCMs) were also applied to climate–mortality studies. RCMs are higher-resolution climate models that can be nested within GCMs to provide more detailed simulations for a particular area. For example, used results from two RCMs (PROMES and HadRM2) that yield greater spatial details about climate.

28Climate change scenarios will determine the size of the predicted future heat-related mortality. Therefore, it is important to consider different emissions scenarios in the impact assessment, offering a range of possible future climates and health impacts. The uncertainty associated with future emissions has been recognized in the U.K. climate projections (UKCP09) by giving probabilistic projections that correspond to each of the three different emissions scenarios: high, medium, and low. These scenarios correspond to three of the commonly used emissions scenarios in SRES: A1FI, A1B, and B1, respectively.

29Because of the varying sets of strengths and weaknesses of different GCMs, the IPCC suggested that no single GCM can be considered the best and that multiple GCMs should be used to account for modeling uncertainties. Also, climate projection data at a higher spatial resolution will be more valuable, especially for information on the urban heat island effect. Better information on the probability of heat waves occurrences will also increase the accuracy of projections concerning the health impacts of climate change.

30Demographic changes. Challenges also arise from the uncertainties of future demographic changes that will modify the future sensitivity of populations to heat stress. Growing numbers of older adults will increase the proportion of the population at risk. In addition to having a diminished physiological ability to cope with heat, the elderly are more likely to live alone, have reduced social contacts, and experience poor health. Also, the effects of heat on mortality appear sometimes to be greater in women, especially elderly women.

31To project the effects of climate change independent of effects of population trends, one approach is to assume that the population size and age structure will remain constant. For example, assumed that population totals for each of the 31 counties in New York City, based on data obtained from the U.S. Census 2000 survey, were held constant throughout the modeling period. Baseline mortality rates for all age groups were also held constant. Similarly,,,, and did not account for population changes.

32If susceptible populations are considered, then future demographic trends should be addressed. used data on the projected population for 2030 that accounted for an aging population. obtained the county population estimates by age group for the years 2005–2030. The population was held constant for the 2025 projection, allowing differences in excess deaths between years to be interpreted as the component due to climate change. estimated population scenarios for Lisbon in line with the SRES. The population growth rates from each SRES storyline were applied to the 1990 Lisbon population to produce future population figures until 2100, and the median population from these calculations was used. National population projections were not used because they did not go far enough into the future.

33Population acclimatization. How populations may acclimatize to elevated temperatures over time is another issue affecting mortality projections. Acclimatization can be a physiological process of humans adjusting to changes in their environment. People may also adapt to extreme heat through increased use of air conditioning, modified behavior patterns, and improved building designs and urban planning.

34One approach is to assume that no acclimatization takes place in the future. For example, argued that epidemiological evidence of the extent to which short- or long-term acclimatization alters mortality risk is limited and sometimes discordant. For their projections the authors therefore assumed that no acclimatization occurred, and hence there would be no future change in the temperature–mortality relationship.,,,,, and also assumed no future acclimatization.

35To incorporate acclimatization, one approach is to use the exposure–response curves from analogue cities. These analogues represent cities whose present climate best approximates the estimated future climate of a target city. For example, modeled acclimatization in New York City using a temperature–mortality response function derived for Washington, DC, and Atlanta, Georgia (USA), which had mean summer temperatures for 1973–1994 that were within approximately 1°F of projected temperatures for the New York City region in the 2050s. However, estimates based on this approach may be biased if social, economic, and demographic characteristics related to mortality differ greatly between the target and analogue regions.

36Another approach involves the use of analogue summers from the same city to model population acclimatization. used analogue summers whereby future acclimatization was based on the temperature–mortality relationship only in the hottest summers on record. identified the five hottest and five coolest summers during 1953–2000 and attributed the differences in daily mean deaths between the hottest and coolest summers to acclimatization.

37Others have accounted for acclimatization by shifting current temperature–mortality relationships to the future. Using this method, the heat threshold increases with time but the slope of the temperature–mortality relationship remains unchanged. assumed that complete acclimatization to an extra 1°C warming in maximum temperature is reached every three decades. considered three possibilities of future acclimatization: no acclimatization, acclimatization to an increase of 2°C, and acclimatization to an increase of 4°C. How acclimatization might reduce the impacts of climate change is not well understood, and there is no consensus on how to estimate its effect. We recommend conducting sensitivity analyses using different approaches to model population acclimatization when projecting future heat-related mortality.

38The potential impacts of climate change on heat-related mortality are the subject of increasing public health concern. A variety of methods have been used to project future heat-related mortality. Although each of the methods has limitations, collectively they provide insight concerning projections of heat-related mortality under climate change scenarios. Projecting heat-related mortality under a changing climate requires analysis of historical exposure–response functions of temperature and mortality and consideration of the future changes in climate, population, and acclimatization.

39Reliable climate projections are now increasingly available for many regions of the world because of advances in climate modeling. The GCMs provide credible estimates of future climate change. Their credibility comes from the well-established physical basis of climate models, from the ability to simulate important aspects of the current climate, and from the ability to reproduce features of past climate changes. Using the multimodel ensembles, along with statistical and dynamical techniques for regionalizing GCM outputs, climate researchers have moved toward representing changes in future climate with probabilities. It is possible that regional climate projections for a given emissions scenario could soon become routine. There is an urgent need for the environmental health community to conduct evidence-based assessments of the health impacts of climate change by closer collaboration with climate researchers.

40Uncertainties in climate change projections should also be considered. Uncertainties may arise from model parameters, or from structural uncertainties as some processes in the climate system are not fully understood or are impossible to resolve because of computational constraints. Thus, the estimates of future heat-related mortality contain uncertainties that need to be carefully interpreted for policy implications.

41Another important consideration is the stability of temperature–mortality relationships over time. There are limitations to using present-day exposure–response functions to project future heat-related mortality levels, because the relationship between heat and mortality may change over time. Changes in mortality risk may occur because of an aging population or because of acclimatization, socioeconomic development, and adaptation strategies. If we do not consider adaptations in the modeling, it would likely lead to an overestimate of the future effects of heat. For instance, adaptation may occur through improved building design and better city planning and land-use patterns, such as green roofs, reflective surfaces on roads and buildings, tree planting, and preservation of regional green space. Adaptation may include changes in exposure patterns as people may spend less time outside, thereby altering the impact of heat. The prevalence of air conditioning has increased, and this trend is expected to continue. The implementation of heat health warning systems is becoming more widespread, and these systems may reduce the health risks from heat waves.

42There is conflicting evidence of air pollution being a confounder and an effect modifier of the temperature–mortality relationship. Some studies reported confounding effects of air pollutants on the association between temperature and mortality, whereas others found no evidence of confounding or effect modification. Recent evidence has indicated that the confounding effect of air pollution is relatively small, and there are independent effects of air pollution and temperature on mortality. However, because extreme heat events and increased levels of air pollution (e.g., ozone) often coincide, it is necessary to understand not only the independent effects of heat on mortality but also any combined effects of heat and air pollution. Air pollution is expected to increase in urban areas due to climate change, so the joint exposure of urban populations to high temperatures and air pollutants will increase in the future. Future heat-related mortality may also be due to indirect causes, such as deaths due to increased ozone caused by increased temperature, or synergistic effects of heat and air pollution.

43Short-term mortality displacement or “harvesting” is another important issue. Mortality displacement suggests that some heat-related deaths in already frail populations are only hastened by heat exposure. If most heat-related deaths were in the very elderly who had only a life expectancy in single years, the public health significance of heat-related deaths would be reduced. Few studies have investigated the degree of mortality displacement for heat-related deaths. The accurate estimation of years of life lost because of high temperatures remains unknown and would likely vary as a function of the severity and duration of heat events.

44As global average temperatures increase, heat-related mortality will increase, but overall effects on mortality could be offset somewhat by reductions in cold-related mortality. For example, estimated that a uniform 1°C warming results in a net mortality decline of 2.65 deaths per standard million per metropolitan areas, with 3.61 additional deaths in summer and 8.92 fewer deaths in winter in U.S. cities. Nevertheless, many scientists believe that the future increase in heat-related mortality is unlikely to be offset by the reduction in cold-related mortality, especially in the medium to long term. More studies are needed to understand how the balance of heat-related and cold-related mortality could change under different climate change and socioeconomic scenarios. However, because public health adaptation strategies could be different for heat waves and cold spells, it is better to separate the projections of heat- and cold-related mortality rather than presenting only a net mortality change.

45Although the magnitude of future climate change remains uncertain, climate modeling exercises indicate that future heat waves will be more frequent, more intense, and longer lasting. Efforts to better understand how climate change will affect population health, especially among the most vulnerable groups, are necessary. Given uncertainties in our understanding of the future population vulnerability to heat, it is important to use various methods to capture a plausible range of the health impacts of climate change. Further research is needed to provide a stronger theoretical framework for projecting heat-related mortality under climate change scenarios, including better understanding of socioeconomic development, adaptation strategies, land-use patterns, air pollution, and mortality displacement.

46Climate change is likely to cause increased heat-related mortality. A few studies have projected heat-related mortality under different climate change scenarios. Significant differences in projected mortality can be found in different emissions scenarios, suggesting that greenhouse gas mitigation policies are important for protecting human health. Although the methods used for projections are still in their early stages and have limitations, the need for evidence-based assessments of future health impacts of climate change is urgent. Such research will significantly contribute to assessing and managing the potential impacts of climate change on heat-related mortality.