Our results confirm that extreme heat days have a clear impact on total mortality in Spain. This has also been found in other multi-city European Studies [12, 15, 16], even in temperate climates . To our knowledge, this is the first study using official thresholds defined by a heat health warning system.
These have been mainly based on the 95th percentile (though with modifications in some scenarios, as previously described) of the 20-years historic series of minimum and maximum daily summer temperatures in each capital city. However, this criterion was only previously associated with health effects in the largest cities of Madrid , Barcelona  and Seville , whereas in others was not validated. Although the number of excesses was relatively homogeneous between the provincial capital cities in each Autonomous Region, there were substantial differences between regions. There were also a few with a surprisingly lower or higher than expected in comparison to other cities of the same Region.
There are several reasons that cities might show atypical relative risks during the supra-threshold days. Firstly, the official thresholds may not be a good indicator, for that city, of days at high risk. Specifically, the 95th percentile may not be a good indicator of risk or the exceptions made to the 95th percentile in the official thresholds criteria were inadvisable. Some of the cities with unusually high risks do appear to have been such exceptions. For example, Oviedo (RR = 2.08, 95%CI = [1.30-3.31]) had both the minimum and maximum threshold revised upwards to their 'floor' levels of 20 C and 33 C, and Zamora (RR = 2.70, 95%CI = [1.50-4.86]) had a minimum temperature threshold (22 C) appreciably higher that the 95th percentile (18.6 C), indicating that the special rule for continental climates had operated. More generally, the meta-regression analysis showed that low mean summer temperature, which was where the exception rules applied, was a strong predictor for high relative risks. Together, these suggest that the upwards revision of thresholds in cooler cities lower than would be appropriate if the intention was to identify days which the same excess risk as other cities on supra-threshold days.
To shed further light on performance of the official thresholds in identifying days of greatest heat-related events, we compared the percentage increase in risk of death on extreme heat days in 2003, calculated in our study (results not shown) with a previous study, by Martinez-Navarro et al. , where observed mortality for the summer of 2003 (from 1st June to 31st August) was compared with the mortality that would have been expected on the basis of historic series since 1990. We found a very weak correlation (r = 0.09) between estimates of both studies, perchance due to imprecision in the two estimates. However, if we only consider the largest cites (over 400,000 inhabitants) this correlation becomes considerably larger (r = 0.59). Indicating that in all these the original thresholds given by 95th percentiles were not revised upwards.
Identifying heat-health thresholds is a practical decision that should respond to credibility, precision and cost criteria . In this study we have not sought to explore all aspects of this nor review all the ways this can be done, so that we retain focus on the simple question of whether the official thresholds have been successful in identifying, in each provincial capital city, days with elevated risks of mortality and whether extent of elevation was about the same level across cities. For the same reason we did not explore factors that might modify the extreme heat impact, such period of time in the summer and duration of the extreme heat events, nor did we compare extremely hot to all other days, as done in some studies. Therefore our estimates are not entirely comparable with other published studies that quantified health effects of extreme hot temperatures.
The geographical variability in the distribution of the extreme heat-mortality effects roughly identifies the climatic zones in the Iberian Peninsula. This could partially be explained by the fact that city-specific risk estimates are calculated using official thresholds, which are at different percentiles of the temperature distribution, which may have added 'artificial' heterogeneity between climatic zones. However, when we used a fixed 95% percentile thresholds heterogeneity in relative risks did not differ substantially (I2 = 51.4% vs. 54.9%). Thus there may still be a case for different thresholds for provincial capital cities based on their climatic region. Effects of high daily temperatures on mortality in English regions have been predicted from the region's climate at the 93 rd percentile of summer mean temperature . Similarly, regional thresholds have been defined for heat-health warning systems in France  and Australia . Age, urban environment and low socio-economic level have been identified as risk factors associated with heat-related mortality [22, 23].
A major strength of the current study was the availability of long time series data sets from all provincial capital cities of Spain, providing enough power to identify and estimate extreme heat effects in all climatic regions of Spain. This study also includes the 2003 summer, which was an unusually hot summer in Europe. For this reason, we tested the sensitivity of our results by removing 2003 from the analysis. However, the overall risk estimate did not differ substantially (RR = 1.21; 95%CI = [1.16-1.27]). This may be explained by the fact that 1996 summer also recorded similar high temperatures in Spain, especially in Central Regions. Overall risk estimate for 1996 (RR = 1.29; 95%CI = [1.08-1.53]) was close to 2003 (RR = 1.31; 95%CI = [1.22-1.40]). Therefore, in a 10 years period we found another extremely hot summer, besides 2003, with a high impact on mortality. This agrees with the latest report issued by the Intergovernmental Panel for Climate Change indicates that climate change will lead to an increase in the frequency and intensity of heat waves . Predictions for the Iberian Peninsula, using general circulation models, indicate a uniform increase in temperature over the course of the 21st century, with an average upward trend every 10 years between 0.6 C to 0.7 C in summer, a greater number and higher frequency of days with extreme heat temperatures in summer .
However, ecological time-series studies can be affected by various biases, in particular potential confounding due temporal patterns. In this study well-established methods were used to control for trend and seasonality. Additionally, some air pollutant levels, which are higher in summer in many European areas , are associated with an increase in mortality and may be associated with hot temperatures as well. In our case, it is difficult to ascertain whether the triggering factor for death in extreme heat days is exclusively due to an increase in temperature or to an increase of air pollution levels that tend to accompany such high temperatures. Unfortunately, due to lack of data our study was unable to assess the influence of air pollutants and other weather variables, such as humidity (although there is little evidence that it is associated with mortality) .
Although official thresholds gave consistent relative risk in the large capital cities, mainly because from previous studies [4–7] the 95th percentile seems to be a 'natural' threshold of triggering mortality for heat days, in some other cities heat thresholds should be updated to better represent impact on health. In this sense, a very recent study conducted in Castilla La Mancha region showed how these 'natural' heat thresholds for maximum temperature range from the 92nd percentile in Cuenca up to 97th percentile in Albacete and Toledo . Therefore, criterion for the choice of thresholds must be based on epidemiological instead only on meteorological basis. In this sense, further research needs to be done to describe and understand how is the relationship of heat and mortality in these cities.