Assessment of the total excess mortality in France

Methodology

In France, the physician fills in the death certificate and after folding the paper part containing the medical description of the cause of death to make it secret, sends it to the “mairie” (town council) . The town council sends the census information (first name, last name age address ) to the Institut National des Statistiques et Etudes Economiques (INSEE, national insititute of statistics and economics) and the health information to the physician at the Departmental (‘county’) Health Office (Direction Départementale des Affaires Sanitaires et Sociales). The departmental health office checks the cause of death and sends it to INSERM to record national statistics. This process takes several months. During the heat wave, the town councils and the departmental health offices were asked to send their respective death certificates to INSERM and INSEE daily, and to give a daily count to InVS for August and September 2003. Regular cross-checking of the three sources meant that accurate information could be obtained [1].
InVS compared the observed number of deaths from 1 to 15 August 2003 with the average rate for the years 2000, 2001 and 2002, modified by population projections for 2003 (the last census was in 1999) [2]. The INSERM study compared the number of deaths by sex and age from 1 to 20 August 2003 with the mean daily number of deaths observed during July, August and September 2000, 2001 and 2002 [1]. More recently, INSERM produced additional results based on validated mortality data between 1 August and 31 December [3].. The reference used to assess the number of deaths in France for this period was the mortality rate by sex and age observed during the period 2000-2002, modified by an estimation of the evolution of death rates and population size for the period 2000-2003.
Meteo-France (the national meteorological service) produced minimal, maximal and average 24 hour temperatures on the basis of a sample of 180 stations representative of French cities [1].

Results

The temperatures increased between 1 August (daily maximal temperature of 25°C) and 5 August (37°C) and maintained themselves at very high levels up to 13 August 2003. They fell abruptly to 28°C between 13 and 16 August [Figure]. Moreover, the high temperatures and the stagnant atmospheric conditions significantly increased ozone levels, with observed concentrations ranging between 130 and 200 µg/m3 in almost every town between 3 and 13 August [1,2].
The increase in the number of excess deaths followed the same pattern as the increase in temperatures. Nationwide, the impact hit on 4 August, when there were 300 excess deaths. The daily excess rose progressively, reaching 1800 deaths on 8 August and about 2200 deaths on 12 August. It regressed quickly on 13 of August to return to normal levels on 19 August (1) [FIGURE].

The analysis of death certificates given by the departmental health offices allowed InVS to produce a first estimate on 28 August of 11 435 excess deaths (excess of 55%) between 1 and 15 August 2003 [2]. On 25 September, INSERM estimated the cumulative excess deaths between 1 and 20 August at 14 800 (excess of 60%) [1].The impact was greater for women (70% increase in excess total mortality) than for men (40% increase in excess mortality)(1). This was the case even for same age groups. Excess mortality reached 20% in the 45-74 year age group, 70% in the 75-94 year age group and 20% in people aged 94 years and over [1].
INSERM also showed that during the last third of the month of August and the month of September the mortality had reached the usual level [3]. October and November 2003 showed the usual death rates in every region.

Assessment of the total mortality in excess between cities

Methodology

For 13 cities, InVS received [TABLE] the death certificates for all deaths of local residents, except for fetal deaths, from the town councils [4] and meteorological data from representative stations in towns. The towns were representative of the different regions. For the Eastern third of the country,from South to North, Nice, Grenoble, Lyon, Dijon, and Strasbourg were chosen. For the middle third and from South to North Marseille and Paris were chosen. For the western third and from South to North Toulouse, Bordeaux, Poitiers, Le Mans, Rennes and Lille were chosen. The excess mortality rate was calculated as the division of the number of deaths 2003 minus the mean of 1999-2002 deaths by the mean of 1999-2002 deaths, for the period between 1 and 19 August.

Results
Dijon, Paris, Poitiers, Le Mans and Lyon clearly showed the highest difference between the usual (1999-2002) and 2003 daily mean temperatures from 1 to 19 August 2003 (>6.7°C) [TABLE]. Toulouse and Bordeaux, presented similar meteorological characteristics but with a milder delta (respectively 6.6° and 6.2° C). Grenoble and Strasbourg, with deltas of 6.3°C and 5.9°C, did not experience very high minimal temperatures. Rennes was less affected by the heat wave, with lower numbers of days with high maxima or high minima and a mean delta of 5.6°C. Marseille and Nice had numerous days with very high minima, but their mean delta was relatively small (4.3°C). The number of excessively warm days in Lille was very low, as was the delta (4.0°C).

In the thirteen cities in the study, a lag of 1 to 3 days between the start of the heat wave and the increase in the number of deaths was observed [4]. An excess in mortality was found in every city, and the disparity of the impact of the heat wave depending on the city appeared clearly. In Dijon, Paris, Poitiers, Le Mans and Lyon, the excess in mortality was particularly marked (>78%) whereas in Lille, the excess of deaths was very low (+4%). Contrasts in the excesses of death could be noted between some towns with quite similar meteorological situations. The excess of deaths in Nice was of 53% whereas in Marseille it was ‘only’ 26%. Strasbourg suffered a 51% excess of deaths whereas Grenoble ‘only’ had an excess of 28%.

Assessment of the excess mortality by diseases

Methodology

The analysis of death causes was done by INSERM for the period between 1 and 20 August for the whole of France. This information was compared to the cause of deaths of 2000 and 2001 for the same period. Data for 2002 were not fully validated. The initial cause of death was taken into account with the classification in 65 categories and 17 CIM chapters from Eurostat. A chapter on death directly linked to heat was created by merging the categories of dehydration, hyperthermia and heatstroke [3].

Results
The analysis of deaths by causes of death between 1 and 20 of August from INSERM [3] did not show any significant impact of heat on deaths under ages of 5 years. For people aged under 45 years the observed excess were mild (19%), and only for deaths caused by undefined conditions, heatstroke, dehydration and hyperthermia, and only for men. Among the 2565 excess deaths observed in the 45-74 year age group, 439 were due to heat related illnesses, 418 to undefined causes, 365 to cardiovascular diseases, and 249 to cancer. The highest relative increase concerned heat related deaths (with 434 cases registered in 2003 compared to a mean of 9 cases for the same period in 2000-2001), mental illness (170%), undefined causes (110%), diseases of the nervous system (70%), genito-urinary diseases (70%), endocrine diseases (60%), infectious diseases (60%) and pulmonary diseases (50%). The relative increase was lower for cardiovascular diseases, cancers and accidental deaths (< 20%). Among the 11 891 excess deaths in people aged 75 years and over, 2852 were directly linked to heat, 2633 to cardiovascular diseases, 1265 to undefined causes, 1213 to respiratory diseases, and 781 to diseases of the nervous system. The relative progressions were most important for heat related diseases (1860%), infectious diseases (130%), accidental falls (130%) and undefined causes (110%). Relative excesses by causes of death were generally more pronounced in women than in men for the same age periods.

Discussion
France was very heavily affected by the 2003 heat wave. It suffered 14 800 excess deaths between 1 and 20 August (+60%). The calculation of this excess can be considered reliable. It is based on the cross-checking of different sources of information (InVS counts, INSEE data and INSERM data). This epidemic event was not immediately followed by any deficit in mortality. INSERM showed that the excess of death observed during the August heat wave had not yet been compensated for at the end of 2003.
No other country in Europe reached such a toll, but other countries differred from France in terms of geographic and temporal extent of the very intense heat wave. Throughout France, 2/3 of the meteorological stations recorded temperatures above 35°C [2]. It is also worth emphasising that elevated odds ratios have been observed in towns where the climatic phenomenon was similar in length and intensity. For example the relative excesses of deaths in Paris, Barcelona and Torino were important [2]. Results from a number of different studies all favour an important role for this exceptional heat stress in the toll registered. The INSERM study showed a strong correlation between the number (n=0-1; n=2-5; n=5-10) of consecutive very hot days (Tmax > 35°C and Tmin > 20°C) and the relative risk of excess deaths among administrative “departments” (RR=1.3 ; RR=1.5 ; RR=1.8) [1]. There is also a trend between the relative excess mortality among the thirteen cities and the delta between the usual temperatures and the observed ones for August 2003 as shown in the table.

Other factors can explain the heterogeneity between towns and regions regarding the impact of extreme temperatures. A chronological study of deaths, temperature, and ozone in 9 cities showed that the proportion of observed deaths explained by these last two variables was very low for Lille, Strasbourg, Marseille, Toulouse, moderate for Bordeaux, Rouen and very important for Paris and Lyon [5]. This result is in favour of a geographical heterogeneity of vulnerability to heat wave. Sociodemographic factors can partly explain this difference. For example, the percentage of ages over 74 years is more elevated in Nice (12.7% in 1999) than in Marseille (9.2% in 1999). Other factors certainly intervene, such as the size of the cities, the urban heat island, cultural habits, or adaptation to very hot temperatures. For example, Marseille suffered a heat wave in 1983, and in 2003, an ‘emergency plan’ to help the public and the hospitals prevent extreme heat effects already existed. This meant that the population of Marseilles was more likely to cope better with a heat wave. The influences of those factors have been analysed by the InVS in specific studies focusing on pollution [5], and heat related risk factors [2].

Based on those results, the French government decided to develop a National Heat Health Watch Warning System (Système d’Alerte Canicule Santé (SACS)) adapted for each 'département' (county). The objectives are to anticipate the health effects of heat waves and to alert the authorities in time to allow the setting up of preventive actions [6]. It has been developed on the basis of a retrospective analysis of mortality and minimal and maximal temperatures data in fourteen pilot cities. The cut-offs have been set in order to anticipate large scale events three days in advance, resulting in an excess mortality above 50% in Paris, Lyon, Marseille and Lille and above 100% in the smallest cities. The system was extended 'département'-wide using the 98th centiles of minimal and maximal temperatures. The national action plan that integrates this watch warning system has four levels. They correspond to various degrees of activations of actors concerning public health surveillance, social supports, and medical preventive actions. It runs from 1 June to 31 September and results in a close cooperation between the meteorological services and the public health agencies. During 2004 no heat wave was observed but the climatologic predictions estimate that summers as hot as 2003 could be more frequent in the future [7].
The efficiency of the heat health watch warning systems has never been put to the test completely. However, some published results support the hypothesis of their effectiveness in the short term, as well as the possibility of adaptation of the population to hot temperatures in the long term [8-10].