An evaluation of spatio-temporal models for the estimation of the mortality relative risk from breast cancer in Galicia, Spain.

Abstract

Background: Disease mapping is now a big focus of interest in the area of Public Health, and the geographical distribution of a disease has an important role in understanding its origin or its causes. The purpose of this work is to review and evaluate different techniques to map the mortality risk of a disease in small geographical areas.

Methods: Three different methods have been studied. The first one is a classical approach consisting of mapping SMRs, which are maximum likelihood estimates of the relative risk under a Poisson model of death counts. In a second step we consider Poisson and negative binomial regression to fit the rates and finally we use a Bayesian approach that assumes a hierarchical model where the death counts follow a Poisson distribution conditioned by the prior information. These methods have been applied to the study of geographical variation in female breast cancer mortality from 1976 to 1999 in the districts of Galicia, Spain.

Results: Mapping the SMRs using the first method has important drawbacks and there are difficulties to distinguish the mortality pattern. With the second method we achieved some improvements. The Bayesian methodology produces smoother maps with a clear mortality pattern.

Discussion: These methods are powerful tools for identifying areas with elevated risk. The Bayesian methodology has many advantages over the other methods that had been analysed in this work.