In this study, we have described and compared two classes of models - the Generalized Linear Models (GLM) and the Generalized Additive Models (GAM) _ which can be used to evaluate the association between atmospheric pollution and morbidity and mortality counts. We consider GAM as an alternative for modeling unspecific nonlinear relationships, and show that this class of models constitute a good option to represent both seasonality and the relation between the outcome and the meteorological variables. As an illustration, we evaluate the association between mortality in elderly people and air pollution in São Paulo, Brazil for the period ranging from 1994 to 1997. The two modeling classes have produced coherent results, but the most sophisticated models were more powerful to detect significant effects. Associations between mortality and the levels of CO, SO2 and, in to a lesser extent, PM10 were observed.
Linear models; Generalized additive models; Air pollution; Morbidity; Mortality; Aged
