Abstracts: ISEE 22nd Annual Conference, Seoul, Korea, 28 August–1 September 2010: Climate Change and Environmental Health
Heat-related deaths are largely preventable through appropriate measures. In 2004, a national heat prevention plan was developed by the French Ministry of Health. Heat alerts are activated when minimum and maximum temperatures averaged over 3 days reach city-specific thresholds.
Triggering of the warning system is crucial to ensure appropriate responses. It faces 2 issues: the choice of the temperature thresholds, and the forecasting uncertainties.
We compared 2 methods for defining the temperature thresholds in 6 French cities. The first relies on a descriptive analysis of the impact of past heat waves and expert judgment. The second uses generalized additive Poisson regression models, controlling for long-term trend, seasonality, and day of the week to determine the daily excess mortality related to temperature and the excess mortality associated to different percentiles of the distribution of the minimum and maximum temperatures averaged over 3 days. A protocol was defined with the national Weather Services to use probabilities of exceeding the temperature thresholds rather than predicted values of temperature, in order to reduce forecasting uncertainties.
In all cities but Paris, there is a good agreement between the thresholds obtained by the 2 methods, with less than 1°C differences. In Paris, the 99th percentile corresponds to thresholds of 21°C and 34°C for minimum and maximum temperature, associated to an excess mortality of 47%. The descriptive study identified thresholds of 21°C and 31°C, associated to an excess mortality of 29%. Percentiles were used to define temperature thresholds for 90 additional cities. Using the probability of exceeding the temperature thresholds rather than the temperature helps reducing the number of false warnings (eg, from 52 to 10 during summer 2009 for 96 cities).
We concluded that a simple method is sufficient to define protective temperature thresholds and that forecasting uncertainties should be taken into account.