Much of the excess mortality from heat waves is related to cardiovascular and other chronic diseases [6,7] and is concentrated in the elderly [1,2,8]. The impact of heat waves on mortality seems also to be higher in urban areas, due to the ‘urban heat island effect’ [9,10,11]. Some studies suggest that this effect could be due to interaction between temperatures and air pollution [12].

In 1991 and 1995 Spain experienced two heat waves, both associated with excess mortality [13-17]. However, heat-related mortality was not considered to be a public health priority in Spain, and specific warning and/or surveillance systems were not implemented [18].

During the summer of 2003 Europe experienced a heat wave that was remarkable both in the magnitude and the duration of the high temperatures recorded. Thousands of deaths were associated with this meteorological phenomenon, highlighting the current inabiligy to deal with this kind of health threat.

This paper presents a summary of the results of the study to describe mortality and detect any excess mortality experienced in summer 2003 in Spain that was carried out by the Instituto de Salud Carlos III.

The heat wave of 2003 in Spain

Summer temperatures in Spain are usually high. The mean daily temperature during the period 1971-2000 for June, July and August for 48 out of the 50 provincial capitals was 21.8ºC. Mean maximum and minimum temperatures for the same period were 28ºC and 15.7ºC respectively (Instituto Nacional de Meteorología).

However, Spain experienced an increase in temperatures during the summer 2003. Mean daily temperatures for the period June-August, in the same group of cities, were 12.9% (2.7ºC) higher than the observed mean of the period 1971-2000. Mean maximum and minimum temperatures for the same period were 11.2% (3ºC) and 16.2% (2.3ºC) higher respectively compared with the series 1971-2000.

Increased mean and mean minimum temperatures during this period were registered in all 48 provincial capitals (range: 3.7% to 33.1% for mean temperatures and range: 2.7% to 24.8% for minimum temperatures) and all but one registered increased mean maximum temperatures (range: -0.6% to 23.7%) during the period June-August 2003.

Mean daily temperatures of 33ºC and over were recorded for at least half of the days (46/92 days) of the period in 15 out of 48 cities. In 8 of these 15 cities temperatures over 33ºC were registered for more than 60 of the 92 days in the period.

Methods
In order to estimate any possible excess in mortality in Spain in summer 2003, we compared observed mortality during the period July-August 2003 with expected mortality in the provincial capitals of the 50 provinces of Spain.

Through a query made to the database at the Ministerio de Justicia (Ministry of Justice), observed mortality was obtained for 27 computerised death registers . The remaining 23 death registers were not computerised, and teams of two people travelled to the provincial offices to obtain the desired data, which was then inserted into the study database. We collected information on date of birth and death, place of death, and place of residence at death for every death certificate entered in the death register between 1 June and 20 August 2003.

To estimate the expected mortality, the Instituto Nacional de Estadística (National Statistics Institute, INE) provided time series of deaths from 1980 to 2002 (2002 data were provisional) for the 50 cities included in the study. For the prediction of deaths in 2003 we have fitted Poisson regression models for different time periods including the year of death as a continuous variable. Age in five year groups (with the oldest group being ‘85 years and over’) and month of death were also included in the model to adjust predictions for variations in age structure of the population and seasonality. Models better fitting observed mortality in the first 5 months of 2003 were based on the 1996-2002 and 1990-2002 time periods. We used models using both time series for each one of the 50 cities included in the study and for each one of the following age groups: 64 years and under, 65 to 74 years, 75 to 84 years, 85 years and over.

Observed number of deaths were compared with expected and the percent variation was calculated for every city ([O-E/E]*100). An overall weighted mean percent variation, using expected number of deaths for each city as weighting variable, was also calculated.

Results
The 50 cities included in the study represent 35% of the total population and all climate spectra in Spain. Median population per city is 152 690 inhabitants with a range between 31 506 and 3 016 788 inhabitants. Of all deaths registered in Spain in the period 1980-2002, 48.7% were registered in this group of cities. In the year 2002 this percentage was 48%.

There is great variability in the daily number of deaths registered in Spain. However, time series show a marked seasonality with peaks in winter months. Smaller peaks are observed in summer months. The increasing trend observed since 1980, probably due to the aging of the population, and to a lesser degree to population growth, seems to have stabilised in recent years [FIGURE 1]. The year of mortality trend stabilisation varies from one city to another.

During June-August 2003 a total of 43 212 deaths were registered in the 50 provincial capitals under study. Of these, 14 236 (33%) occurred in June, 13 895 (32%) in July and 15 081 (35%) in August. For 141 deaths, the date of birth was not available. For the 43 071 deaths for which age could be calculated, 13 039 (30%) were of people aged 85 years or older, 13 831 (32%) were of people aged 75 to 84 years, 7888 (18%) were of people aged 65 to 74 years, and 8312 (19%) were of people aged 64 years and under.

Figure 2 shows daily deaths, a 5 day moving average of daily deaths, and mean daily temperatures in the 50 cities under study during summer 2003. In this figure we observe a peak in daily deaths during the second week of August. Two smaller peaks, in mid-June and during the second week of July are also observed. The three peaks observed in daily deaths coincide with the three waves of high temperatures suffered along the summer. Although there is an important variability in daily deaths and temperatures registered among the provincial capitals included in the study, the pattern observed in figure 2 is applicable to most of them.

Tables 1 and 2 show the total expected and observed deaths and estimated excess mortality percent for the 50 provincial capitals and for the two time periods better predicting expected deaths (1996-2002 and 1990-2002). Excess deaths in summer 2003 compared with expected is 10.6% (4151 deaths) higher than expected using as predictor the model based on the1996-2002 time series and 7.9% (3166 deaths) using as a predictor the model based on the 1990-2002 time series.

Assuming that 48% of deaths were registered in the provincial capitals (data for 2002) we can estimate between 6595 and 8648 excess deaths in Spain from 1 June to 20 August 2003, using the 1990-2002 or 1996-2002 time series respectively for our model.

Although results using both models, based on the 1990-2002 and 1996-2002 time series, do not differ very much, we consider that the later time series (1996-2002) better estimates expected deaths because of the stabilisation of the mortality trend in the last few years. The following results refer only to the comparison with the model based on the 996-2002 time series.

The excess deaths were higher in August (17% more than expected), but observed deaths were more than expected for the three months under study (9% in June and 5% in July) [TABLE 3]. Only the elderly were affected and important decreases in mortality were registered in people under 65 years throughout the entire period, while we observed 29% excess deaths among people 85 years and over [TABLE 3].

Table 4 shows expected and observed deaths and estimated mortality difference percent for the period January- August 2003. Deaths registered in January and February, and to a lesser extent in March, were fewer than expected. However, the excess detected in the summer period overcompensated for this difference and for the first 8 months of 2003 we detected 1964 (1.7%) more deaths than expected.

Discussion
Observed deaths in June, July and August 2003 in Spain were between 7.9% and 10.6% higher than expected. Although excess mortality was more important in August, excess deaths were observed from June. Significant excess mortality was observed only in the elderly (75 years and older), while among those 64 years and younger, mortality decreased during this period. Access to air conditioning at work and use of swimming pools and other practices that lower body temperature among younger people could account for part of this reduction. Further studies of mortality in people aged 74 years and under are needed to explain these findings.

The objective of this study was to estimate any excess mortality experienced in Spain during summer 2003. An association between mortality and temperatures or other variables such as ozone and other pollutants has not been tested. However, the known association between high temperatures and mortality, the fact that the three heat waves experienced in Spain in the summer of 2003 occurred near in time to the three periods of high mortality registered and the distribution of this high mortality throughout the country reinforce the hypothesis of this association.

Based on these results, estimated excess deaths in Spain could be between 6595 and 8648. However, this study included data from provincial capitals and therefore, although they cover a wide range of city sizes, the urban heat island effect [9,10,11] described in the literature could account for part of this excess mortality, and results would not be representative of mortality in Spain, because it may differ in rural areas.

A second study of mortality carried out by the Instituto de Salud Carlos III in a random sample of 107 out of 7458 rural villages with fewer than 10 000 inhabitants, representing a total of 140 807 people, estimated a 40% increased mortality compared with the mean mortality observed in the three previous years (2000-2002) [18]. This study strengthens the magnitude of the results presented in this article.

Mortality experienced in Spain during the winter of 2002-2003 was marginally less than mortality observed during the previous winter. It could be argued that people who unexpectedly did not die during the preceding winter were the people who died during the summer. Data from a newly established mortality surveillance system (not presented in this article) show an excess mortality in the months following the summer, probably associated to an early appearance of the 2003-2004 influenza season. These results would not support the ‘harvesting’ theory.

Population groups at high risk, such as people 85 years and older, as identified in this study and possibly because of the mechanisms explained by Kenney et al [19] and Foster et al [20], can be identified and specific preventive measures aiming exposure reduction can be implemented if good coordination between surveillance and alert systems and social and health services is achieved.

Results of several studies showing the association between high temperatures and mortality during heat waves in several cities in Spain were published during the 1990s. However, the magnitude and media coverage, and therefore the social impact of the heat wave experience in summer 2003 transferred the debate from academia to the community and political arena.

This debate has altered decision makers’ perception of heat-related health problems and the control of heat-related mortality has become a priority. For summer 2004 the Spanish government launched the Plan de acciones preventivas contra los efectos del exceso de temperaturas sobre la salud (Prevention plan against adverse effects on health of high temperatures) [21], improved for the summer 2005 and created an interministerial steering commission for this initiative. This plan included:

- A temperature-based alert system using as alert threshold the 95th centile of observed daily maximum temperatures during the last 25 years.
- Awareness campaigns addressed to high risk groups, the general population and healthcare and social services professionals.
- A voluntary register of people at high risk who could benefit from support activities delivered by the Red Cross and other social organisations.
- Development of conduct protocols during heat waves for healthcare and social services professionals.
- A daily mortality surveillance system.