Disease mapping using empirical Bayes and Bayes methods on mortality statistics in the Netherlands

Abstract

To compare the occurrence of disease or death in epidemiological or health-policy research an important requirement is a valid measure. This report is about the choice of the indicator that can be used to estimate relative risk for disease or death in areas on a map. The properties of several measures are investigated using data from the project 'Regional Health Profiles'. The estimates from Bayesian and Empirical Bayesian models which were used are less affected by random fluctuations in the data than the commonly used estimators Standardized Mortality Ratio (SMR) or Comparative Mortality Figure (CMF). The Bayesian and Empirical Bayesian method model the unknown relative risks of each area jointly as a spatial stochastic process. These models can be perceived as originating from the family of 'Generalized Linear Mixed Models' (GLMM). GLMM is useful in any situation were some kind of dependency between objects exists. Different assumptions on the prior distribution or the spatial structure gave different outcomes of the estimates, as expected. Methods to test the validity of the assumptions need to be developed. Until then it can not be concluded that one method is superior to another.<br>