Time Series Methods in the Epidemiological Studies Regarding Air Pollution

Abstract

The time series methods in the epidemiological studies on air pollution are reviewed, illustrated by means of an autoregressive Poisson regression which was employed in the APHEA and EMECAM Projects. A listing is provided of the variations in the daily number of deaths of people over age 70 (all causes, CIE-9:001-799) in Barcelona, 1991-1995, with the average variations in the daily smog pollution levels. A Poisson regression is used insofar as the dependent random variable presumably follows such a probability distribution. As variables possibly leading to confusion, the impact of weather variables (daily temperature and relative humidity averages), seasonal, tendency-related behaviors and day of the year on the death rate are taken into account (all estimated on a determinist basis), in addition to any other variable which behaves in a way that it can be related to the dependent variable (i.e. flu epidemics). The relationship between the death rate and the confusing-causing variables is modeled on a non-linear basis, and the foreseeable lag times are also taken into account (i.e. by using explicative variable time lags). However, due to control not being perfect, it has been decided to opt for estimating an autoregressive Poisson model (adding in some different explicative variables time giving rise to a lag in the death rate) offsetting the residual autocorrelation. The main advantage of the method of analysis described above is that of making it possible to control confusing variables from a determinist standpoint with a software to which all of the groups taking part in this Project had access. This also affords the possibility of using this method in a set, standardized manner, facilitating the comparison of results and making an objective point analysis possible.