Study population and setting
The EuroHEAT project involves nine European cities (Athens, Barcelona, Budapest, London, Milan, Munich, Paris, Rome, Valencia), with a total population of around 25 million citizens, which represent a variety of climatic, socio-economic, and air pollution characteristics. Available data for daily mortality, meteorological and air pollution were provided for each city between 1990 and 2004. Only summer months (June-August) were included in the study. Summer 2003 was analyzed separately to assess the impact of this exceptional heat wave episode that affected most of the European cities, and results were compared with heat waves from other years included in the study period.
Data base
Mortality data were daily counts of primary deaths for all natural causes (ICD-9: 1-799; ICD-10: group A-R), cardiovascular (ICD-9: 390-459; ICD-10: group I), cerebrovascular (ICD-9: 430-438; ICD-10: group I 600-699), and respiratory causes (ICD-9: 460-519; ICD-10: group J), by gender and age groups (65-74, 75-84, 85 +), except from Paris where only total mortality was available in 2003.
Meteorological data including air temperature, dew point temperature, sea level pressure, total cloud cover, wind speed and wind direction were collected for the entire study period at the city airports, every 3 hours.
To investigate potential confounding by air pollution, data were also collected for the following variables: SO2 (24-hour mean), TSP or Black Smoke (24-hour mean), PM10 (24-hour mean), PM2.5 if available (24-hour mean), NO2 (maximum 1 hour, 24-hour mean), O3 (maximum 1 hour, maximum 8-hour moving average), and CO (maximum 8-hour moving average).
Exposure definition
Exposure to heat waves considered both extreme day time values in terms of maximum apparent temperature (Tappmax) and high night time temperatures through minimum temperature (Tmin).
Tappmax is a discomfort index based on air and dew point temperature, calculated using the following formula [15, 16]:
Heat waves were thus defined as
-
1)
periods of at least two days with Tappmax exceeding the 90th percentile of the monthly distribution
or
-
2)
periods of at least two days in which Tmin exceeds the 90th percentile and Tappmax exceeds the median monthly value.
Daily counts of total and cause specific mortality (cardiovascular, cerebrovascular and respiratory) in the heat wave periods were considered as outcome variable, stratifying by gender and age groups (65-74 yrs, 75-84 yrs, 85 + yrs).
City specific analysis
As the first stage, a city specific analysis was conducted using Generalized Estimating Equations (GEE models)[17] to analyze longitudinal data. A Poisson distribution was assumed for the outcome variable (mortality) and days were characterized as "heat wave" or "non heat wave" days as the exposure variable to estimate the effect on mortality.
Observations from different years were assumed to be independent, while observations within the same summer were correlated. A similar approach has already been suggested in other studies [18–21]. Since the number of clusters (summers) was small, and equal to the number of years in the study period, we used the model-based estimator for the coefficients' standard errors, as recommended in the presence of few large clusters [22]. A first order autoregressive structure within each year was chosen, based on an exploratory analysis similar to the one described by Chiogna & Gaetan [23, 24].
A common model was specified for each city taking into account the following as potential confounders: holidays, day of the week and calendar month, linear terms for barometric pressure (lag 0-3) and wind speed, linear and quadratic terms for time trend and the maximum 1-hour daily value of NO2 (μg/m3) at (lag 0-1). NO2 was chosen to adjust for air pollution as it has shown to be a good indicator of traffic in a large European collaborative project [25] which assessed the impact of air pollution on mortality, using meteorological variables as potential effect modifiers.
The effect was estimated as the percent increase in daily mortality during heat wave compared to non heat wave days. The effect of each heat episode was also investigated with respect to specific heat wave characteristics such as duration, intensity and timing within the season.
Duration was categorized using the city specific median values of heat wave duration (number of consecutive heat wave days) as the cut-point:
-
short heat wave if duration was shorter than the median,
-
long heat wave if duration was equal to or longer than the median.
Intensity was also categorized into two levels according to the extreme values of Tappmax reached during the heat wave:
-
low intensity heat wave if Tappmax was below the monthly 95th percentile,
-
high intensity heat wave if Tappmax was equal to or above the monthly 95th percentile.
Timing within the season was defined according to the time interval between different heat waves:
-
the first heat wave of each summer was considered separately,
-
heat waves that occurred between 1 and 3 days after the previous one,
-
heat waves that occurred three or more days after the previous one.
Pooled analysis
To summarize city-specific results, cities were grouped into two regions, according to geographical and climatological criteria in order to control for heterogeneity, as defined in the PHEWE study [18]: "Mediterranean" (Athens, Barcelona, Milan, Rome, and Valencia) and "North-Continental" (Budapest, London, Munich and Paris) City-specific estimates were combined through a random effect meta-analysis using the method described by DerSimonian and Laird [26]. To estimate the impact in each region, a GEE model was used, similar to the city-specific model, but adding a city indicator variable and interaction terms of the exposure variable with the confounders.