How to estimate exposure when studying the temperature-mortality relationship? A case study of the Paris area.

Time series studies assessing the effect of temperature on mortality generally use temperatures measured by a single weather station. In the Paris region, there is a substantial measurement network, and a variety of exposure indicators created from multiple stations can be tested. The aim of this study is to test the influence of exposure indicators on the temperature-mortality relationship in the Paris region. The relationship between temperature and non-accidental mortality was assessed based on a time series analysis using Poisson regression and a generalised additive model. Twenty-five stations in Paris and its three neighbouring departments were used to create four exposure indicators. These indicators were (1) the temperature recorded by one reference station, (2) a simple average of the temperatures of all stations, (3) an average weighted on the departmental population and (4) a classification of the stations based on land use and an average weighted on the population in each class. The relative risks and the Akaike criteria were similar for all the exposure indicators. The estimated temperature-mortality relationship therefore did not appear to be significantly affected by the indicator used, regardless of study zone (departments or region) or age group. The increase in temperatures from the 90(th) to the 99(th) percentile of the temperature distribution led to a significant increase in mortality over 75 years (RR = 1.10 [95% CI, 1.07; 1.14]). Conversely, the decrease in temperature between the 10(th) and 1(st) percentile had a significant effect on the mortality under 75 years (RR = 1.04 [95% CI, 1.01; 1.06]). In the Paris area, there is no added value in taking multiple climatic stations into account when estimating exposure in time series studies. Methods to better represent the subtle temperature variations in densely populated areas in epidemiological studies are needed.