Backgrounds: In epidemiology, standardized Incidence Ratio (SIR) can have large variance and it is then difficult to distinguish random fluctuations from real spatial variations when describing spatial variations in the rate of cancer. In this context, hierarchical model produce smoothed relative risks estimations helpful for solving this problem. The main advantage of these methods is to combine information of each geographical area with that obtained from prior assumption on the similarity between geographical sub-units. Nevertheless different assumptions produce different geographical maps of incidence of cancer, and the purpose of the present study was the development of a strategy to choose the most satisfactory description of the incidence of digestive cancer in a French department.
Methods: The strategy to choose the most satisfactory geographical map depends on the following criteria: variability between geographical sub-units, auto-correlation, and variability within geographical sub-unit. These criteria have been estimated from observed data for each site of cancer.
Results: This strategy was applied to digestive tract cancers diagnosed between 1976 and 1997 in the department of Côte-d'Or, France. High-risk areas were often detected in the urban zone of the department, but without autocorrelation in most cases.
Conclusion: This strategy permitted to describe cancers in very small areas, avoiding to a large extent the danger of focusing on falsely positive high-risk areas.