Abstract

This study on the application of Generalized Linear Mixed Model (GLMM) models to estimate disease prevalence in geographic regions is part of the SAVIAH project. The SAVIAH project (Small Area Variation in Air quality and Health) is a cooperation of centres in four countries, financed by the EEC and set up to study the relation between air pollution and the prevalence of CARA in children. The advantage of models over 'crude' estimates of prevalence is that they are able to reduce the random fluctuations in the outcome, thereby providing a clearer view on the true 'underlying' pattern of risk of disease. To make appropriate choices in the modelling process, first tests on the existence of heterogeneity and spatial autocorrelation were carried out. In the SAVIAH study area in Amsterdam heterogeneity and spatial autocorrelation of CARA prevalences in neighbourhoods were not present. Therefore relatively simple models were applied that 'smooth' values to a general mean. The differences between the models are discussed.