An analysis of alternative classification schemes for medical atlas mapping.

Most national disease atlases adopt a classification scheme based on either the percentile distribution of rates or on the national mean. Although these schemes have a direct interpretation, they are based on the univariate statistical distribution of rates and not on their spatial distribution, and distort the underlying spatial autocorrelation in the data. If the purpose of the maps is to represent spatial patterns, alternative classification schemes might be more appropriate. This research proposes an alternative classification method that maximises spatial similarity among contiguous units in the same class interval. The method has been illustrated using selected data from the German Cancer Atlas published in 1984.