[Short-term modeling of the effect of air pollution on health. Example: SO2 and total mortality, Paris 1987-1999]

Since 1990, many epidemiological time series studies have provided evidence that ambient air pollution levels have adverse health effects. The ERPURS study (Evaluation des Risques de la Pollution Urbaine pour la Santé) has permitted to quantify this impact in the Paris region. This study was based on an ecological time series approach. We present, step by step, the method used, illustrated by an example: association between SO2 levels and total mortality (excluding external causes), 1987-1990. Mortality modelling has taken trend into account by a linear term, seasons by trigonometrics functions sum, day of the week effects by 6 dummy variables, temperature peak by a dummy variable, influenza epidemics by appropriate variables, mean temperature by linear and quadratic terms, relative humidity by a linear term. SO2 1 day lag was introduced in the model by a linear term. The central issue is the control of seasonal variations and long term trend. An inadequate control can lead to some spurious results. The relationship between mortality and weather variables is generally nonlinear. The use of statistical and graphical diagnostics, are necessary at each step. Time series analysis are important tools to study short term relationship between air pollutants and health indicators. The method applied in the ERPURS study is only one of the possible approaches. Whatever the method used, it is important to understand the underlying process of the data and to control for confounding factors with the appropriate method for the temporal structure of the data.