Solving the problem of small-population-based areas for the analysis of rare diseases by clustering with constraints methods.

The aim of this study was to find a sensible fusion of small geographical areas into, as far as possible, homogeneous larger regions with the necessary minimal population size according to 14 indicators of socioeconomic development, which is known to be indirectly related to cancer incidence. The starting point was the minimal population size which could still provide an estimation of a statistically significantly lower rate relative to the national average. Being aware of the heterogeneity and complexity of cancer etiology, the problem was studied step by step: regionalization was obtained according to selected socioeconomic indicators with different numbers of regions (from 60 to 32). With the best-obtained regionalization into 32 regions by clustering with constraints methods, zero values were reduced from 112 to 6, while almost the same variance of most cancers was retained.