France’s heat health watch warning system

Abstract

In 2003, a Heat Health Watch Warning System was developed in France to anticipate heat waves that may result in a large excess of mortality. The system was developed on the basis of a retrospective analysis of mortality and meteorological data in fourteen pilot cities. Several meteorological indicators were tested in relation to levels of excess mortality. Computations of sensibility and specificity were used to choose the meteorological indicators and the cut-offs. An indicator that mixes minimum and maximum temperatures was chosen. The cut-offs were set in order to anticipate events resulting in an excess mortality above 100% in the smallest cities and above 50% in Paris, Lyon, Marseille and Lille. The system was extended nationwide using the 98th percentile of the distribution of minimum and maximum temperatures. A national action plan was set up, using this watch warning system. It was activated on 1st June 2004 on a national scale. The system implies a close cooperation between the French Weather Bureau (Météo France), the National Institute of Health Surveillance (InVS) and the Ministry of Health. The system is supported by a panel of preventive actions, to prevent the sanitary impact of heat waves.

Modifications of the system for summer 2005

The experience of the first year of running of the system has lead to several changes in its design. Among these adjustments, the definition of the warning thresholds for the biometeorological indicators was slightly modified. Other modifications were done regarding the practical organisation of the system and especially the management of situations at-risk.

In 2005, the thresholds have been re-defined for two reasons. First, the use of the 98th percentile of temperature was not consistent with the use of a temperature averaged over three days. The thresholds have to be based on the distribution of moving average of temperature over three days. Second, it was assumed that the method used to define the thresholds was not applicable unless specific conditions were met. These conditions are a sufficient number of heat waves (>5) and a baseline of mortality stable enough (i.e. main cities only). Pilot cities that did not meet these criteria were then excluded from the study: Bordeaux, Toulouse and Nice because they had too few heat waves; Grenoble, Lille, Limoges, Dijon because the mortality baseline was too noisy (here the choice was qualitative). Using the remaining cities, new percentiles, based on the distribution of the averaged temperatures were computed and tested. The 99.5th percentile was chosen for its optimal combining of sensitivity and specificity. Yet, it has to be underlined that these notions of sensitivity and specificity are not readily applicable to this topic, since heat waves are not reproducible events. They only give indications on the capability of the thresholds, but the final choice is mainly subjective.

Using the 99.5th percentile new thresholds were proposed, as shown in Table 1. In most cities, the difference with the previous thresholds is lower than 1°C, expect in Lille, Grenoble and Limoges where the minimal threshold was substantially raised, and in Toulouse where the maximal threshold was lowered.

The system is still imperfect, as the temperatures are not the only factors to consider regarding heat load. The humidity may also be interesting, but cannot be used in a predictive system due to the difficulties of forecasting. The characteristic of the summer before the heat wave may also be of importance. Indeed, the dramatic impact of the 2003 heat wave may partly be explained by a very warm summer, with elevated temperatures in June and July, which created a state of exhaustion in the most vulnerable groups of population. In addition, the forecasting of meteorological parameters is always associated with an uncertainty that has to be dealt with. For instance, it was observed in 2004 that the maximal temperatures were slightly underestimated, while the minimal temperatures were overestimated. These errors were small and infrequent (<1°C for 76% of the days) but still interfered with the specificity of the system. Indeed, situations when the indicators are close to the thresholds are usual and highly sensitive to even minor uncertainties in the forecasting. This led to the definition of additional qualitative criteria to consider: humidity, wind, air pollution, intensity and duration of the heat wave, sanitary signals. For borderline situations, the additional information is discussed with the health scientists and with the meteorologists. It is believed that this will help in improving the communicability and the efficiency of the system, which were the main concerns for the users of this system.

Cities	2004		2005
	Threshold minimal	Threshold maximal	Threshold minimal	Threshold maximal
Bordeaux	22	36	21	35
Dijon	19	34	19	34
Grenoble	15	35	19	34
Lille	15	32	18	31
Limoges	16	36	20	33
Lyon	20	34	20	34
Marseille	22	34	22	34
Nantes	20	33	20	34
Nice	24	30	24	31
Paris	21	31	21	31
Strasbourg	17	35	19	34
Toulouse	21	38	21	36
Tours	17	34	19	35