Cluster analysis with constraints: its use with breast cancer mortality rates in Argentina.

One approach to analyse geographic variations of a disease within a country is through mortality rates in administrative areas. In some geographical areas the rates can be unstable due to their low population. Thus, the statistical analysis of the rate could be not significant due to low population, not because of the real value. An alternative approach to this problem is to set a minimum level of population that enables valid statistical comparisons with the national rate to be made. It is decided to apply an algorithm that groups neighbouring geographical units that reach the minimum population. In the resulting regions, mortality rates are calculated, and the geographic patterns are analysed through the Moran's I coefficient of spatial clustering. This paper presents an application of this approach to mortality rates from breast cancer by Argentine departments, political administrative units into which a province is divided. Applying this procedure, 217 regions were obtained. In two of the regions the rates were significantly higher than the national rate, while in 40 regions they were significantly lower. Significant spatial grouping, reflected by a Moran I coefficient of O-47, was observed.

Duke Scholars

Author Daniel Wojdyla