ORIGINAL PAPER Climate change and heat-related mortality in six cities Part 1: model construction and validation Simon N. Gosling & Glenn R. McGregor & Anna P��ldy Received: 5 October 2006 /Revised: 6 February 2007 /Accepted: 12 February 2007 / Published online: 9 March 2007 # ISB 2007 Abstract Heat waves are expected to increase in frequency and magnitude with climate change. The first part of a study to produce projections of the effect of future climate change on heat-related mortality is presented. Separate city- specific empirical statistical models that quantify significant relationships between summer daily maximum temperature (Tmax) and daily heat-related deaths are constructed from historical data for six cities: Boston, Budapest, Dallas, Lisbon, London, and Sydney. ���Threshold temperatures��� above which heat-related deaths begin to occur are identified. The results demonstrate significantly lower thresholds in ���cooler��� cities exhibiting lower mean summer temperatures than in ���warmer��� cities exhibiting higher mean summer temperatures. Analysis of individual ���heat waves��� illustrates that a greater proportion of mortality is due to mortality displacement in cities with less sensitive temperature��� mortality relationships than in those with more sensitive relationships, and that mortality displacement is no longer a feature more than 12 days after the end of the heat wave. Validation techniques through residual and correlation anal- yses of modelled and observed values and comparisons with other studies indicate that the observed temperature���mortality relationships are represented well by each of the models. The models can therefore be used with confidence to examine future heat-related deaths under various climate change scenarios for the respective cities (presented in Part 2). Keywords Mortality. Climate change . Temperature . Mortality displacement . Heat waves Introduction Both warm and cold extremes of temperature have adverse effects on health. A non-monotonic ���V-shaped��� relationship is often observed between temperature and mortality��� annually (Huynen et al. 2001) and for the separate warm and cold seasons (Ballester et al. 1997). Hajat et al. (2006) have shown that, although linear relationships exist be- tween temperature and mortality, during extreme heat events mortality exceeds that expected from a linear association and is better represented non-linearly. Kalkstein and Davis (1989) describe a ���threshold temperature��� beyond which mortality increases above the baseline level. Differ- ent thresholds have been identified for a variety of causes of death (P��ldy et al. 2005 Huynen et al. 2001). Thresholds may be confounded by other meteorological variables, e.g. Saez et al. (2000) illustrated a 2��C higher threshold (23��C) on very humid days when the relative humidity was above 85% in Barcelona, Spain, but there is also evidence that humidity may have insignificant effects on mortality (Dessai 2002, 2003 Ballester et al. 1997 Braga et al. 2001). Thresholds have also been found to vary temporally for a single location (Davis et al. 2003 Ballester et al. 1997), and according to age, with elderly populations being most susceptible to changes in temperature (Conti et al. 2005 Donaldson et al. 2003 Huynen et al. 2001). Interestingly, there is evidence that cardiovascular fitness may be more important than age in determining individual vulnerability to heat (Havenith 1997). Havenith et al. (1995) examined the response to heat stress across a heterogeneous sample of 56 individuals aged 20���73 years Int J Biometeorol (2007) 51:525���540 DOI 10.1007/s00484-007-0092-9 S. N. Gosling (*) : G. R. McGregor Department of Geography, King���s College, London WC2R 2LS, UK e-mail: simon.gosling@kcl.ac.uk A. P��ldy Jozsef Fodor National Institute of Environmental Health, Budapest, Hungary

in a warm humid climate of 80% relative humidity and 35��C air temperature. The effect of age was negligible compared with effects related to fitness, which was measured by maximum oxygen uptake. Comparative studies have shown the occurrence of geographical varia- tion in thresholds. Heat-related/cold-related mortality thresholds occur at higher/lower temperatures in locations with a relatively warmer/colder climate, and the gradient (or steepness) of the temperature���mortality relationship for increasing/decreasing temperature is often found to be lower in warmer/colder locations than colder/warmer ones (Donaldson et al. 2003 Pattenden et al. 2003 Keatinge et al. 2000 Eurowinter 1997). For example, Curriero et al. (2002) illustrated across 11 US cities that threshold temperatures were higher in warmer southern cities, where the temperature���mortality association was less sensitive, than in cooler northern cities. The variation of thresholds and temperature���mortality gradients has led to inference on how populations may acclimatise to changing climatic conditions (Donaldson et al. 2003 Curriero et al. 2002 Braga et al. 2001 Saez et al. 2000). The direct effects of extreme temperature on health are not always immediate���a lag is often observed between the temperature event and resultant mortality whereby separate previous days��� temperatures or lagged moving averages are associated with the current day���s mortality. Lags of less than 3 days are most commonly associated with heat- related mortality (Hajat et al. 2002 Michelozzi et al. 2005 Conti et al. 2005) but different lags may be associated with disease-specific mortalities (Gemmell et al. 2000 McGregor 1999 P��ldy et al. 2005 Ballester et al. 1997). Some studies present a negative relationship between hot temperatures and mortality for lags above 3 days, which compensates some of the deaths caused by heat during the initial days of the heat event (Hajat et al. 2002, 2005 Pattenden et al. 2003 Braga et al. 2001). This is known as ���mortality dis- placement���, whereby the heat principally affects individuals whose health is already compromised and who would have died shortly anyway, regardless of the weather. Estimates of mortality displacement vary considerably���Sartor et al. (1995) estimated that 15% of total deaths during the Belgium 1994 heat waves were due to displacement. Gouveia and Fletcher (2000) estimate mortality displace- ment as about 50% during the 1994 heat waves in the Czech Republic. Estimates varied between 1% and 30% in France during the summer heat wave of 2003 (Le Tertre et al. 2006) and evidence from the United States estimates the value as between 25% and 50% (Kalkstein 1993). A number of studies point to increases in heat-related mortality under climate change scenarios. Donaldson et al. (2001) estimate a 253% increase in annual heat- related mortality by the 2050s for the United Kingdom, and Dessai (2003) estimated the heat-related mortality rate to increase from between 5.4 and 6.0 (per 100,000) for 1980���1998 to between 5.8 and 15.1 for the 2020s, and 7.3 to 35.6 for the 2050s for Lisbon. The range in values was due to the combined uncertainties inherent in climate change projections, potential acclimatisation, and methodologies. Assessments of the impacts of climate change on heat- related mortality need to be location specific because it has been shown that the relationship is not evenly distributed in space (Davis et al. 2004 Kalkstein and Davis 1989). Attention also needs to be paid to the inherent uncertainties in impact assessments, especially those arising from climate projections, so that a range of possible impacts are illustrated. This paper summarises the first part of a study aimed at producing projections of the effect of future climate change on heat-related mortality. The research is published in two parts (Fig. 1). In this paper (Part 1) separate empirical���statistical non-linear regression models based on the aggregate dose-response relationship between daily maximum temperature (Tmax) and heat-related deaths (the difference between observed and expected deaths) are Fig. 1 The adopted methodology for this research (adapted from Dessai 2002) 526 Int J Biometeorol (2007) 51:525���540

developed for six cities in order to model the current relationship between weather and heat-related mortality. In Part 2, climate change and population change scenarios are applied to the models developed here to estimate the heat- related mortality burden attributable to climate change for each city. This includes an exploratory uncertainty analysis to examine the uncertainties in the projections due to climate modelling, which is considered as a major source of uncertainty in climate-health modelling (Dessai 2003). Uncertainties concerning acclimatisation and those inherent in the temperature���mortality models are also included. Additional uncertainties such as population ageing and use of air-conditioning/heating units exist, but will not be examined due to the added complexities in modelling them. Materials and methods Selection of cities The cities selected for this study were Boston, Budapest, Dallas, Lisbon, London and Sydney (Table 1). The aim was to include cities in different climates such as Continental Cool Summer, Temperate, Humid Subtropical and Medi- terranean (McKnight and Hess 2000) so that any regional differences in exposure���response could be examined. Another important consideration was that data for at least 10 years was available to provide a reliable representation of the cities��� climates, and that it was available at reasonable cost. Mortality data Daily total deaths from all causes were obtained for each city to include both heat stroke and any possible comorbid factors (Davis et al. 2003 Kilbourne 1997 Kunst et al. 1993). The maximum available data record for each city was examined because this gives a more reliable represen- tation of climate and any associations with mortality, and gives more precise regression coefficients than if shorter periods were used (Davis et al. 2003 Horst 1966). Therefore this study assumes that exposure���response relationships remain constant over time. Davis et al. (2004) state that this stationary nature is often assumed in assessments such as this, but it is noted that some evidence points to the possibility of non-stationarity in several US cities (Davis et al. 2003). Mortality data was missing for Dallas 1990, which was excluded from the analysis. Any other missing data was replaced by linear interpolation. An anomaly resulting from 137 extra deaths caused by an airliner accident at Dallas Fort- Worth International Airport on 2 August 1985 (Wikipedia 2006) was excluded from the analysis. As the focus of the research is heat-related mortality, only the summer months [June, July, August (December, January, February for Sydney) hereafter referred to as ���summer���] were used for analysis. For inter-city analysis and estimation of future mortality burdens under climate- and population-change scenarios, mid-year population estimates were obtained or calculated by linear interpolation between census years as denominators for the computation of crude mortality rates Table 1 Sources of data used for each city City K��ppen classification (McKnight and Hess 2000) Meteorological station used Meteorological data source Mortality and population data source Population data available Period of study Boston (US) Cooler Humid (Continental Cool Summer) Boston-Logan International Airport US National Climatic Data Centre US National Centre for Health Statistics Census Years 1970, 1980, 1990, 2000 1975���1998 Budapest (Hungary) Warmer Humid (Temperate) Kitaibel P��l Street Hungary National Meteorological Service Hungary National Statistical Office Annual 1970���2000 Dallas (US) Warmer Humid (Humid Subtropical) Dallas Fort-Worth International Airport US National Climatic Data Centre US National Centre for Health Statistics Census Years 1970, 1980, 1990, 2000 1975���1998 Lisbon (Portugal) Warmer Humid (Mediterranean) Lisboa Geof��sico Portuguese Meteorology Institute Portuguese National Institute of Statistics Census Years 1981, 1991 1980���1998 Greater London (UK) Warmer Humid (Temperate) Heathrow Airport British Atmospheric Data Centre Office of National Statistics Annual 1976���2003 Sydney (Australia) Warmer Humid (Humid Subtropical) Observatory Hill Australian Bureau of Meteorology Australian Bureau of Statistics Annual 1988���2003 Int J Biometeorol (2007) 51:525���540 527