[Critical analysis of a clustering method applied to the study of geographic distribution of cancer]

Cancer registration permits to elaborate atlases that describe the spatial distribution of incidence within the registration area. The division of this area in small units is usually an administrative one, leading a wide range of unit sizes. Hence, results of indirect standardization are often difficult to interpret, due to the large variability of the observed cases. We used Huel's method (1986), which is based on the grouping of contiguous geographical zones that have the same incidence level. Here are displayed the results for cancer incidence within the "département de l'Isère" (France). We focus on method mechanisms and on the impact of the choice of the risk level.