You are here

Article Text

Article menu

Download PDFPDF

Research report
Impact of heat on mortality in 15 European cities: attributable deaths under different weather scenarios
  1. M Baccini1,2,
  2. T Kosatsky3,4,
  3. A Analitis5,
  4. H R Anderson6,
  5. M D'Ovidio7,
  6. B Menne4,
  7. P Michelozzi7,
  8. A Biggeri1,2,
  9. the PHEWE Collaborative Group
  1. 1Department of Statistics “G. Parenti”, University of Florence, Florence, Italy
  2. 2Biostatistics Unit, ISPO Cancer Prevention and Research Institute, Florence, Italy
  3. 3Environmental Health Division, BC Centre for Disease Control, Vancouver, Canada
  4. 4Global Change and Health, WHO Regional Office for Europe, Rome, Italy
  5. 5Department of Hygiene, Epidemiology and Medical Statistics, Medical School, University of Athens, Athens, Greece
  6. 6Division of Community Health Sciences and MRC-HPA Centre for Environment & Health, St. George's, University of London, London, UK
  7. 7Department of Epidemiology, Local Health Authority Rome E, Rome, Italy
  1. Correspondence to Michela Baccini, Department of Statistics “G. Parenti”, University of Florence, Viale Morgagni 59, 50134 Florence, Italy; baccini{at}ds.unifi.it

Abstract

Background High ambient summer temperatures have been shown to influence daily mortality in cities across Europe. Quantification of the population mortality burden attributable to heat is crucial to the development of adaptive approaches. The impact of summer heat on mortality for 15 European cities during the 1990s was evaluated, under hypothetical temperature scenarios warmer and cooler than the mean and under future scenarios derived from the Intergovernmental Panel on Climate Change Special Report on Emission Scenarios (SRES).

Methods A Monte Carlo approach was used to estimate the number of deaths attributable to heat for each city. These estimates rely on the results of a Bayesian random-effects meta-analysis that combines city-specific heat-mortality functions.

Results The number of heat-attributable deaths per summer ranged from 0 in Dublin to 423 in Paris. The mean attributable fraction of deaths was around 2%. The highest impact was in three Mediterranean cities (Barcelona, Rome and Valencia) and in two continental cities (Paris and Budapest). The largest impact was on persons over 75 years; however, in some cities, important proportions of heat-attributable deaths were also found for younger adults. Heat-attributable deaths markedly increased under warming scenarios. The impact under SRES scenarios was slightly lower or comparable to the impact during the observed hottest year.

Conclusions Current high summer ambient temperatures have an important impact on European population health. This impact is expected to increase in the future, according to the projected increase of mean ambient temperatures and frequency, intensity and duration of heat waves.

  • Bayesian
  • climate
  • health impact assessment
  • meta analysis me

https://doi.org/10.1136/jech.2008.085639

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    Introduction

    An important association between daily deaths and relatively higher ambient summer temperatures has been demonstrated, albeit with substantial evidence of among-city heterogeneity, in analyses performed as part of the Assessment and Prevention of Acute Health Effects and Weather Conditions in Europe (PHEWE) project.1 2 Based on counts of daily deaths and on weather and air quality data for 15 European cities over 11 years, the PHEWE summer analyses provide city-specific air-quality-adjusted estimates of mortality risk by maximum apparent temperature. For most cities, the warm season heat-mortality curve appears J-shaped, with Mediterranean cities showing a higher temperature threshold and a steeper slope above threshold than continental and northern European cities. These findings are consistent with studies comparing cities at different latitudes and climates.3–5

    The realisation that summer heat influences rates of daily death, even in northern Europe,6 coupled with the likelihood that a warmer world and an ageing population will increase vulnerability to summer weather, makes it crucial that we apply our understanding of the population heat response to planning for adaptation to heat.7–10

    Here, we assess the impact of summer heat on mortality in 15 European cities. The existence of between-year temperature variability, within and among cities, provides opportunity to assess how urban populations respond to summers, which overall are either warmer or cooler than the mean over the 1990s. We use this variability to contrast the daily mortality population response to heat during the warmest summers to that occurring during the coolest summers.

    Projections of future impact of heat on mortality are also derived on the basis of climate scenarios developed by the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report 2007.7 The IPCC provides climate scenarios that encompass different combinations of the main driving forces of future greenhouse gas trajectories, including demographic change, and social, economic and technological development. We used three of the proposed IPCC scenarios to project the impact of heat on mortality at 2030.

    Methods

    Our impact assessment is based on a daily temperature-mortality relationship presented previously.2 The result is an estimate of the number of heat-attributable deaths in each city over the 11-year study period, 1990–2001. Then, based on (1) the range of the observed data and (2) three IPCC scenarios, we developed estimates of heat exposure to assess the response of European urban populations to summers either cooler or warmer than the observed 11-year record overall.

    Mortality and exposure data

    We considered daily meteorological and mortality data from 15 cities enrolled in the PHEWE project: Athens, Barcelona, Budapest, Dublin, Helsinki, Ljubljana, London, Milan, Paris, Prague, Rome, Stockholm, Turin, Valencia and Zurich. The study period was different among cities, ranging from 5 years in Athens to 11 years in Dublin, Helsinki, Milan and Stockholm.1

    The impact evaluation was carried out for the warm season, defined as the period from 1 April to 30 September. We considered mortality for all natural causes (International Classification of Diseases-9 codes 1–799). Daily counts of deaths by broad age group (15–64, 65–74, 75+ years) were available for all 15 cities. These age groups have been used by the PHEWE investigators due to their consistency with the epidemiological time series literature.11

    Exposure was measured by daily maximum apparent temperature, calculated from 3-hourly air temperature and humidity data.12 13 For Barcelona, the daily average apparent temperature was used as 3-hourly data were not available.

    Assessment of the exposure–mortality relationship

    As with our earlier assessment of the effect of heat on mortality over 11 warm seasons overall, the current exposure measure incorporated temperatures cumulated up to lag 3 (lag 0–3).

    Given that our earlier results2 highlighted a typical “V” or “J” shape for the relationship between maximum apparent temperature at lag 0–3 and mortality during the warm season, the current assessment of heat-attributable deaths likewise summarises the heat effect by two parameters:

    1. a threshold, corresponding to the minimum of the exposure–response curve, which is interpretable as the exposure corresponding to the minimum risk of death;

    2. a slope above the threshold, which expresses the effect of exposures exceeding the threshold (on a logarithmic scale).

    The statistical analysis consisted of two steps. First, city-specific estimates of the parameters of interest were obtained specifying a GEE Poisson model with first-order autoregressive variance–covariance matrix for the daily count of deaths.14 Separate estimates of the slope above the threshold were obtained by age class (15–64, 65–74, 75+ years), while a common threshold was assumed for all age classes. To adjust for the confounding effect of air pollution, a linear term was introduced in the city-specific models. We used the maximum hourly concentration of nitrogen dioxide as a proxy for the overall daily air pollution level. The choice of NO2 as the air pollution indicator was based on comparability, availability and completeness of daily measurements among cities.1

    Bayesian random-effects meta-analysis models were used to combine the city-specific results.15 16 To reduce heterogeneity among cities, the city-specific thresholds and slopes were pooled into two groups defined a priori on the basis of meteorological and geographical criteria (Mediterranean cities: Athens, Barcelona, Ljubljana, Milan, Rome, Turin and Valencia; North-Continental cities: Budapest, Dublin, Helsinki, London, Prague, Paris, Stockholm and Zurich).1

    For assessment of attributable deaths, we used the posterior city-specific distributions or shrinkage distributions of threshold and slope above the threshold obtained from the Bayesian meta-analysis.17 Using posterior city-specific distributions of the effects estimates is recommended in the presence of heterogeneity.18–20 Posterior city-specific distributions updated the first stage estimates, taking into account information from all other cities. These represent a compromise between first-stage city-specific results, which can be affected by large sampling variability, and the overall meta-analytic estimate, which, although usually stable, is difficult to interpret in the presence of heterogeneity among cities.21 22

    Posterior city-specific distributions have been summarised by their means (posterior means) and appropriate credibility intervals. Lower and upper bounds of a (1−α)% credibility interval are defined as the (α/2)th and (1−α/2)th percentiles of the posterior distribution, respectively. For example, the 10th and the 90th percentiles define the 80% credibility interval for the parameter of interest.

    Health impact measure

    The impact of heat on mortality was quantified in terms of number and fraction of deaths attributable to all apparent temperatures exceeding the city-specific threshold.

    We used a Monte Carlo approach to estimate uncertainty around threshold and slope of city-specific effects of heat.23 24 For each city, we sampled 1000 values from the city-specific posterior distribution of the slope (bc, c=1,2…1000) and from the city-specific posterior distribution of the threshold (hc) obtained through the Bayesian meta-analysis. Then, for each sampled pair (bc, hc), we calculated a daily time series of attributable deaths according to the following formula:

    ADct=Yt−Yctbaseline;Yctbaseline=Yexp(bc(Tt−hc))ifTt>hcADct=0ifTt<hc

    where Tt and Yt are the observed daily maximum apparent temperature (lag 0–3) and the observed daily number of deaths at time t (t=1, 2….t*), respectively; Yctbaselineis the estimated baseline level of mortality at time t; and ADct is the number of attributable deaths at time t for sample c. Attributable deaths were calculated by age group (15–64, 65–74, 75+ years). For each city, the total number of attributable deaths was obtained by summing over the three age classes.

    In the Monte Carlo simulations, we assumed independence between threshold and slope. In a sensitivity analysis, different levels of correlation were assumed, with similar results obtained (results not shown).

    Definition of temperature–mortality scenarios by resetting data observed

    For each city, four alternative scenarios were defined by selecting specific daily series of baseline mortality and exposure.

    The extreme scenarios consisted of the 6-month daily time series of maximum apparent temperature with corresponding baseline mortality obtained by selecting over all the observed years, for each month and day:

    H1, the second to hottest day

    C1, the second to coldest day.

    For example, for the H1 scenario, we considered the hypothetical summer constituted by the second to hottest 1 April, the second to hottest 2 April….the second to hottest 30 September. We selected the second to coldest and the second to hottest days (instead of the hottest and the coldest days) to avoid scenarios that markedly over-represent extreme values.

    In addition, we defined two scenarios by selecting:

    H2, the summer characterised by the highest mean level of apparent temperature

    C2, the summer characterised by the lowest mean level of apparent temperature.

    These four scenarios were used for evaluating the impact of heat during summers hypothetically cooler and warmer than the overall (up to) 11-year mean.

    IPCC scenarios

    The impact of a projected future climate was evaluated using temperature projections derived from the IPCC Special Report on Emission Scenarios (SRES).7 In this report, six families of scenarios are defined based on different future levels of greenhouse gas emission to which different projected warming levels are associated.

    We selected three scenarios:

    • scenario B1 (low-emission scenario): best estimate of the temperature change equal to 1.8°C at 2090–2099 relative to 1980–1999.

    • scenario A1B (middle-emission scenario): best estimate of the temperature change equal to 2.8°C at 2090–2099 relative to 1980–1999.

    • scenario A2 (high-emission scenario): best estimate of the temperature change equal to 3.4°C at 2090–2099 relative to 1980–1999.

    To project heat-mortality impacts in 2030, we assumed a constant rate of temperature change over the period 1999–2099, and for each scenario we calculated the attributable number of deaths, by adding to each observed daily apparent temperature Tt the corresponding projected average temperature change ΔT.

    Results

    A large variability in mortality and exposure was observed among cities (table 1). The mean total daily deaths during the warm season ranged from 6.3 in Ljubljana to 149 in London (crude rate 2.39 and 2.19 per 100 000, respectively). The mean daily maximum apparent temperature over the warm season ranged from 14.3°C and 14.7°C in Helsinki and Dublin to 27.9°C and 29.4°C in Athens and Valencia, respectively.

    View this table:
    Table 1

    Population size, percentage of older people; study period; daily average deaths for natural causes during the warm season (April–September) over the whole study period and during the coolest and the warmest years (years with the lowest and the highest average daily level of maximum apparent temperature, respectively); descriptive statistics of daily maximum apparent temperature and mean of daily maximum NO2 concentration during the warm season (April–September)

    The city-specific posterior distributions of the thresholds and of the per cent change in mortality associated with a 1°C increase in exposure above the threshold, by age class, are summarised in table 2 (posterior means and 95% credibility intervals). The posterior mean of the above-threshold temperature mortality variation in the over 75s ranged from 1% in Valencia to 7.3% in Athens, for Mediterranean cities, and from 1.4% in Zurich to 3.2% in Paris, for North-Continental cities. A large heterogeneity in the threshold was observed among Mediterranean cities, while threshold estimates for North-Continental cities appeared more homogeneous. The effect of heat was largest in the 75+ years age group.2

    View this table:
    Table 2

    Means and 95% credibility intervals (see Methods for definition) of the city-specific posterior distributions of maximum apparent temperature threshold and per cent variation in mortality (lag days 0–3) for all natural causes associated with 1°C increase in maximum apparent temperature above the threshold, by age group (15–64, 65–74, 75+ years)

    In table 3, we report for each city the average maximum apparent temperature (lag 0–3) and the percentage of days with exposure exceeding the threshold, as observed over the 11-year study period and as predicted by each of the four hypothetical scenarios.

    View this table:
    Table 3

    Mean and SD (within brackets) of daily maximum apparent temperature (lag days 0–3) and percentage of days with lag 0–3 maximum apparent temperature exceeding the city-specific threshold observed during the study period and under the hypothetical low-exposure and high-exposure scenarios, by city

    There is a large inter-city variability in the observed percentages of days above the threshold: from 1.3% and 6.4% (Dublin, Helsinki) to 60.2% and 58.1% (Barcelona, Valencia).

    The mean exposure levels under the warmest and coolest year (H2, C2) scenarios were less extreme than under the H1 and C1 (second to hottest and second to coldest day) scenarios. Under the hypothetical low-exposure scenarios (C1, C2), the percentages of days with exposure levels exceeding the threshold are lower than the actual ones. When the second to coldest days are considered (C1), the percentages exceed 10% only for Barcelona, Valencia and Budapest. As expected, under the high-exposure scenarios (H1, H2), the percentages are larger than those observed for the (up to) 11 summers overall. Under the hottest year scenario (H2), in 11 cities more than 1 day in four contributes to attributable deaths. There are 14 (all except Dublin) such cities under the second to hottest day scenario (H1).

    The average number of attributable deaths per warm season (183 days) is reported in table 4. The average number of attributable deaths per warm season over the 11-year study period varies from 0 in Dublin to 423 (more than two extra deaths per day) in Paris. However, an appreciable level of imprecision can be seen by the breadth of the 80% credibility intervals.

    View this table:
    Table 4

    Mean number of attributable deaths per year and 80% credibility intervals (see Methods for definition) calculated under the observed series and the four hypothetical low-exposure and high-exposure scenarios, by city

    In some cases, the second to hottest day scenario (H1) and the hottest year scenarios (H2) produced highly divergent impact estimates (see Budapest, Milan, Paris, Prague, Rome and Stockholm). For London and Athens (and to a slight degree for Barcelona), there were more attributable deaths during the warmest year than under the second to hottest day scenario (H1). An explanation of this apparently paradoxical result is that the impact estimate depends on the exposure distribution and the baseline rate of death. Even if the H1 scenario is more extreme than the warmest year (H2) scenario in terms of exposure, it can happen that the baseline number of events (which depends on several factors apart from weather) is lower under the H1 than under the H2 scenario, resulting in a lower estimated impact.

    Under the second to coldest day scenarios (C1), the number of attributable deaths was much lower. Less extreme results were obtained considering the coolest year (C2). While for cities such as Athens, London, Milan, Prague and Turin, the mortality impact was strongly higher under the warmer than under the coolest scenarios, for others, in particular Valencia and Barcelona, appreciable attributable mortality was estimated under both warmer and cooler scenarios.

    We calculated for each city the average number of attributable deaths divided by the average total number of deaths per warm season over the 11-year study period (attributable fraction). The total and age-specific attributable fractions are reported in table 5 with the corresponding counts of attributable deaths. The highest impact fractions were found for four Mediterranean cities (Barcelona, Rome, Turin and Valencia) and for two North-Continental cities (Budapest and Paris).

    View this table:
    Table 5

    Total and age-specific percentages of attributable deaths over the total number of deaths during the study period, by city

    In most cities, the highest impact was observed for the 75+ years age group; attributable fractions larger than 2% were observed for the 15–64 years age group in Budapest (AD=71, AF=2.02), Rome (AD=40, AF=2.17) and Valencia (AD=13, AF=2.55).

    The results of the projections in 2030 under the IPCC SRES scenarios are reported in table 6. The number of attributable deaths under the SRES scenarios is generally lower or comparable to the number of attributable deaths observed during the warmest year of the up to 11-year series.

    View this table:
    Table 6

    Mean number of attributable deaths per year, by city, calculated for the study period and projected at 2030 under low (B1), middle (A1B) and high (A2) greenhouse gas emission scenarios developed by IPCC (2007)

    In comparing the impact under different scenarios, account should be taken of the presence of Monte Carlo error, which is due to the use of a sample of values from the posterior city-specific distributions of threshold and slope. The amount of this error is large if the spread of the posterior distributions is large. We found that this error is not substantial for most of the cities included in the study; however, it can explain, for example, the fact that the mean number of attributable deaths per year in 2030 for Valencia is lower than the mean number of attributable deaths calculated for that city during the study period.

    Discussion

    Warm ambient temperatures have an important impact on population mortality in the 15 European cities enrolled in the PHEWE study.

    We investigated the potential health impacts of changing summer weather, developing scenarios warmer and cooler than the mean, based on the observed (5–11 years) time series of exposure.

    The hottest year and the coldest year scenarios describe realistic and moderate warming and cooling situations. Cooler summer scenarios are not realistic under any climate change projection but are useful for comparison purposes. For most cities, an appreciable number of heat-attributable deaths was observed even during the coolest summer.

    The second to hottest and the second to coldest day scenarios are made up of days extracted in isolation and out of the time series in which they occurred. Consequently, the variability of the resulting maximum apparent temperature time series can be, for some cities, unrealistically low. The number of attributable deaths calculated under these two scenarios provides, in some sense, upper and lower bounds for the impact of heat in the cities studied, even if a certain underestimation of the impact is probable for the H1 scenario because cumulative exposure to heat could have stronger impacts on health than the effect of isolated hot days.25

    In general, the impact of heat strongly increases under the warming scenarios, except in cities that are characterised by low variability in apparent temperature between and within summers (eg, Valencia).

    For several cities, we observed large discrepancies between the impact under H1 and H2 scenarios. This happens when the simulated hottest time series is a collection of very extreme temperatures, while, during the actual warmest summer overall, high apparent temperature exposures are partly balanced by cooler days. In this sense, the simulated hottest time series represent future scenarios where hot days and heat waves are more frequent than at the present.

    On the contrary, for those cities where it is already possible to observe summers that are a collection of very hot days (eg, Barcelona, Valencia and Athens), the impact under H1 and H2 scenarios is similar, indicating that the second to hottest day scenario is not particularly extreme in those situations.

    Paris and Budapest are the North-Continental cities for which the calculated impact of a warmer summer is highest in absolute and relative terms. In general, we found a certain evidence that high NO2 concentrations are related to high relative impact (table 2), as expected in the presence of an interaction between heat and air pollutant concentration.26 However, differences in air pollutant concentration can only partly explain the higher impact observed for Budapest and Paris. Further studies should investigate this point.

    The major mortality impact was on older people, both in absolute and relative terms. However, in Budapest, Rome and Valencia, a large attributable fraction of deaths was also observed in the youngest age class. This is an important finding, as deaths in younger individuals are associated with greater loss of potential life years.27

    The projections for 2030 are based on specific greenhouse gas emission scenarios derived from assumptions about demographic, economic and technological growth.7 The impact under the SRES scenarios is slightly lower or comparable to the impact during the observed warmest year. This result is consistent with projections that in the future, very hot summers will be more usual.26

    Issues related to model assumptions

    According to the PHEWE study, we modelled the exposure–response relationship by a threshold-slope model.2 The gain in terms of fitting that we would have obtained by introducing two or more cut-points or a flexible function for describing the relationship would have been to the detriment of the stability and the general validity of the results.

    We assumed that the threshold does not vary by age. Empirical evidence from Madrid would support this assumption.28 Anyway, as the threshold for the 75+ years age class is likely lower than the common threshold (reflecting the higher susceptibility of older people) and as the total number of attributable deaths is strongly influenced by those in the 75+ years age class, our estimates of all-age heat-mortality impact are probably conservative.

    We could argue that the variation from 23°C to 32°C in city-specific thresholds reflects population acclimatization, physiologic and behavioural to the diverse climates across Europe.5 However, epidemiological evidence of the extent to which short-term or long-term acclimatization mechanisms alter population risk is limited and sometimes discordant.2 29–32 In our assessment, we therefore assumed that no adaptation occurs modifying the exposure–response curve under the hypothesised temperature scenarios. Probably, this assumption conveys a degree of overestimation of the impact of warmer summers in future.

    Evidence that the effect of heat during the first week after exposure is partially compensated by harvesting phenomena has been reported in the literature.33 34 In the PHEWE study, mortality displacement was found in the Mediterranean and north-continental regions, with a cumulative effect at lag 25, which was around 30% of the cumulative effect at lag 3.2 Despite this result, here we have given equal weight to all deaths attributable to average maximum apparent temperatures above the threshold recorded up to 3 days before, regardless of whether mortality displacement was a few days or several years. Future analyses should quantify the net impact of heat, taking into account harvesting mechanisms, and should quantify the impact in terms of years of life lost, which may be of greater relevance to public health than crude attributable mortality.

    What is already known on this subject

    • There is an important association between daily deaths and high summer ambient temperatures registered in European cities.

    • Future climate changes involve increase of mean ambient temperatures and frequency, intensity and duration of heat waves.

    • From a public health point of view, it is crucial to assess the health impact of summer heat under current and predicted future warming scenarios.

    What this study adds

    • Heat has a great impact on mortality in European populations, with a mean attributable fraction of deaths during summer of around 2%.

    • The largest impact was observed among older people; however, in some cities, an important impact was also found for younger adults.

    • Heat-attributable deaths markedly increase under the predicted future warming scenarios.

    Acknowledgments

    We thank Ben Armstrong for the useful suggestions.

    References

    1. ↵
      1. Michelozzi P,
      2. Kirchmayer U,
      3. Katsouyanni K,
      4. et al
      . Assessment and prevention of acute health effects of weather conditions in Europe, the PHEWE project: background, objectives, design. Environ Health 2007;24:6–12.
      OpenUrl
    2. ↵
      1. Baccini M,
      2. Biggeri A,
      3. Accetta G,
      4. et al
      . Heat effects on mortality in 15 European cities. Epidemiology 2008;19:711–19.
      OpenUrlCrossRefPubMedWeb of Science
    3. ↵
      1. Keatinge WR,
      2. Donaldson GC,
      3. Cordioli E,
      4. et al
      . Heat related mortality in warm and cold regions of Europe: observational study. BMJ 2000;321:670–3.
      OpenUrlAbstract/FREE Full Text
    4. ↵
      1. Curriero FC,
      2. Heiner KS,
      3. Samet JM,
      4. et al
      . Temperature and mortality in 11 cities of the eastern United States. Am J Epidemiol 2002;155:80–7.
      OpenUrlAbstract/FREE Full Text
    5. ↵
      1. McMichael AJ,
      2. Wilkinson P,
      3. Kovats RS,
      4. et al
      . International study of temperature, heat and urban mortality: the ‘ISOTHURM’ project. Int J Epidemiol 2008;37:1121–131.
      OpenUrlAbstract/FREE Full Text
    6. ↵
      1. Rocklöv J,
      2. Forsberg B
      . The effect of temperature on mortality in Stockholm 1998–2003: a study of lag structures and heatwave effects. Scand J Public Health 2008;36:516–23.
      OpenUrlAbstract/FREE Full Text
    7. ↵
      IPCC. 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: Cambridge University Press, 2007.
    8. ↵
      1. McMichael AJ,
      2. Woodruff RE,
      3. Hales S
      . Climate change and human health: present and future risks. Lancet 2006;367:859–69.
      OpenUrlCrossRefPubMedWeb of Science
    9. ↵
      1. Kovats RS,
      2. Campbell-Lendrum D
      . Climate change and human health: estimating avoidable deaths and disease. Risk Anal 2005;25:1409–18.
      OpenUrlCrossRefPubMed
    10. ↵
      1. Kovats RS,
      2. Hajat S
      . Heat stress and public health: a critical review. Annu Rev Public Health 2008;29:41–55.
      OpenUrlCrossRefPubMedWeb of Science
    11. ↵
      1. Kelsall JE,
      2. Samet JM,
      3. Zeger SL,
      4. et al
      . Air pollution and mortality in Philadelphia, 1974–1988. Am J Epidemiol 1997;146:750–62.
      OpenUrlAbstract/FREE Full Text
    12. ↵
      1. Steadman RG
      . The assessment of sultriness: part I: a temperature-humidity index based on human physiology and clothing science. J Appl Meteorol 1979;18:861–73.
      OpenUrlCrossRef
    13. ↵
      1. Kalkstein LS,
      2. Valimont KM
      . An evaluation of summer discomfort in the United States using a relative climatological index. Bull Am Meteorol Soc 1986;7:842–8.
      OpenUrl
    14. ↵
      1. Diggle P,
      2. Liang KY,
      3. Zeger SL
      . Analysis of longitudinal data. Oxford: Oxford University Press, 1994.
    15. ↵
      1. Sutton AJ,
      2. Abrams KR
      . Bayesian methods in meta-analysis and evidence synthesis. Stat Methods Med Res 2001;10:277–303.
      OpenUrlAbstract/FREE Full Text
    16. ↵
      1. Sutton AJ,
      2. Higgins JPT
      . Recent developments in meta-analysis. Statist Med 2008;27:625–50.
      OpenUrlCrossRef
    17. ↵
      1. Gelman A,
      2. Carlin JB,
      3. Stern HS,
      4. et al
      . Bayesian data analysis. London: Chapman and Hall, 2004.
    18. ↵
      1. Post E,
      2. Hoaglin D,
      3. Deck L,
      4. et al
      . An empirical Bayes approach to estimating the relation of mortality to exposure to particulate matter. Risk Anal 2001;21:837–42.
      OpenUrlCrossRefPubMedWeb of Science
    19. ↵
      1. Biggeri A,
      2. Bellini P,
      3. Terracini B
      . Meta-analysis of the Italian studies on short-term effects of air pollution—MISA 1996–2002. Epidemiol Prev 2004;28(4–5 Suppl):4–100.
      OpenUrlPubMed
    20. ↵
      1. Le Tertre A,
      2. Schwartz J,
      3. Touloumi G
      . Empirical bayes and adjusted estimates approach to estimating the relation of mortality to exposure of PM10. Risk Anal 2005;25:711–18.
      OpenUrlCrossRefPubMedWeb of Science
    21. ↵
      1. Ades AE,
      2. Lu G,
      3. Higgins JPT
      . The interpretation of random-effects meta-analysis in decision models. Med Decis Making 2005;25:646–54.
      OpenUrlAbstract/FREE Full Text
    22. ↵
      1. Mindell J,
      2. Boaz A,
      3. Joffe M,
      4. et al
      . Enhancing the evidence base for health impact assessment. J Epidemiol Community Health 2004;58:546–51.
      OpenUrlAbstract/FREE Full Text
    23. ↵
      1. Burmaster DE,
      2. Anderson PD
      . Principles of good practice for the use of Monte Carlo techniques in human health and ecological risk assessments. Risk Anal 1994;14:477–81.
      OpenUrlCrossRefPubMedWeb of Science
    24. ↵
      1. Grigg OA,
      2. Spiegelhalter DJ,
      3. Jones HE
      . Local and marginal control charts applied to methicillin resistant Staphylococcus aureus bacteraemia reports in UK acute National Health Service trusts. J R Statist Soc A 2009;172:49–66.
      OpenUrlCrossRef
    25. ↵
      1. Hajat S,
      2. Armstrong B,
      3. Baccini M,
      4. et al
      . Impact of high temperatures on mortality: is there an added heat wave effect? Epidemiology 2006;17:632–8.
      OpenUrlCrossRefPubMedWeb of Science
    26. ↵
      Menne and Matthies. Public health responses to extreme weather events: EuroHEAT technical summary. Kopenhagen, Denmark: WHO Regional Office for Europe, 2009.
    27. ↵
      1. Lopez AD,
      2. Mathers CD,
      3. Ezzati M,
      4. et al
      . Global burden of disease and risk factors. New York: Oxford University Press, 2006.
    28. ↵
      1. Linares C,
      2. Díaz J
      . Impact of heat waves on daily mortality in distinct age groups. Gac Sanit 2008;22:115–19.
      OpenUrlCrossRefPubMedWeb of Science
    29. ↵
      1. Donaldson GC,
      2. Keatinge WR,
      3. Nayha S
      . Changes in summer temperature and heat-related mortality since 1971 in North Carolina, South Finland, and Southeast England. Environ Res 2003;91:1–7.
      OpenUrlPubMed
    30. ↵
      1. Carson C,
      2. Hajat S,
      3. Armstrong B,
      4. et al
      . Declining vulnerability to temperature-related mortality in London over the twentieth century. Am J Epidemiol 2006;164:77–84.
      OpenUrlAbstract/FREE Full Text
    31. ↵
      1. Medina-Ramón M,
      2. Schwartz J
      . Temperature, temperature extremes, and mortality: a study of acclimatization and effect modification in 50 United States cities. Occup Environ Med 2007;64:827–33.
      OpenUrlAbstract/FREE Full Text
    32. ↵
      1. Barnett AG
      . Temperature and cardiovascular deaths in the US elderly: changes over time. Epidemiology 2007;18:1–4.
      OpenUrlCrossRefPubMedWeb of Science
    33. ↵
      1. Braga ALF,
      2. Zanobetti A,
      3. Schwartz J
      . The time course of weather-related deaths. Epidemiology 2001;12:662–7.
      OpenUrlCrossRefPubMedWeb of Science
    34. ↵
      1. Hajat S,
      2. Armstrong BG,
      3. Gouveia N,
      4. et al
      . Mortality displacement of heat-related deaths: a comparison of Delhi, Sao Paulo, and London. Epidemiology 2005;16:613–20.
      OpenUrlCrossRefPubMedWeb of Science

    Footnotes

    • This work was done on behalf of the PHEWE Collaborative Group: Department of Epidemiology, Local Health Authority RM/E, Rome, Italy: P Michelozzi, U Kirchmayer, F de'Donato, M D'Ovidio, D D'Ippoliti and C Marino; School of Geography, Geology and Environmental Science, University of Auckland, Auckland, New Zealand: G McGregor; Department of Statistics, University of Florence, Florence, Italy: A Biggeri and M Baccini; Biostatistics Unit, ISPO Cancer Prevention and Research Institute, Florence, Italy: G Accetta; WHO Regional Office for Europe, Rome, Italy: B Menne and T Kosatsky; Department of Hygiene, Epidemiology and Medical Statistics, Medical School, University of Athens, Athens, Greece: K Katsouyanni and A Analitis; Department of Astrogeophysics, University of Joannina, Joannina, Greece: P Kassomenos; Municipal Medical Research Institute, Barcelona, Spain: J Sunyer; Division of Community Health Sciences, St. George's, University of London, London, UK: H R Anderson and R Atkinson; National Institute for Public Health Surveillance, Saint Maurice, France: S Medina; National Centre of Public Health, Institute of Environmental Health, Budapest, Hungary: A Paldy; Department of Epidemiology Health Authority Milan, Milan, Italy: L Bisanti; Regional Environmental Protection Agency of Piedmont, Grugliasco, Italy: E Cadum; Department of Epidemiology, Charles University, Prague, Czech Republic: B Kriz; Department of Environmental Health, Institute of Public Health, Ljubljana, Slovenia: A Hojs; St. James's Hospital, Dublin, Ireland: L Clancy and P Goodman; Department of Environmental Health, Umea University, Umea, Sweden: B Forsberg; Unit of Environmental Epidemiology, National Public Health Institute, Kuopio, Finland: J Pekkanen; Department of Medical Statistics, National Institute of Hygiene, Warsaw, Poland: B Woityniak; Department of Mathematical Sciences, University of Aberdeen, Aberdeen, UK: I Jolliffe; German Meteorology Service, Freiburg, Germany: G Jendritzky; Department of Climatology, Institute of Geography and Spatial Organization, Warszawa, Poland: K Blazejczyk; Institute of Atmospheric Physics, Academy of Sciences, Prague, Czech Republic: R Huth; Climatological Department, Meteorological Office, Environmental Agency, Ljubljana, Slovenia: T Cegnar; Institute of Social and Preventive Medicine, University of Basel, Basel, Switzerland: C Schindler; Valencia School of Health Studies, Valencia, Spain: F Ballester; French Meteorology Service, Roissy Charles De Gaulle, France: G Monceau; Department of Geography and Regional Studies, University of Miami, Miami, Florida: L S Kalkstein.

    • Funding This study was funded by the European Commission, DG Research, FP5 (contract QLK4-CT-2001-00152); by the PH project 2004322, EuroHEAT; and supported by WHO Regional Office for Europe.

    • Competing interests None.

    • Patient consent Obtained.

    • Provenance and peer review Not commissioned; externally peer reviewed.