On the difficulty to delimit disease risk hot spots

[Display omitted] ► Decision makers need clearly delimited risk zones to apply protection measures. ► In the underlying reality risk values are more likely to vary smoothly. ► Areas at risk do not correspond to the reality and are difficult to estimate. ► We apply risk partition models and clustering method for point data. ► The exact delimitation of the risk zones are quite different. Representing the health state of a region is a helpful tool to highlight spatial heterogeneity and localize high risk areas. For ease of interpretation and to determine where to apply control procedures, we need to clearly identify and delineate homogeneous regions in terms of disease risk, and in particular disease risk hot spots. However, even if practical purposes require the delineation of different risk classes, such a classification does not correspond to a reality and is thus difficult to estimate. Working with grouped data, a first natural choice is to apply disease mapping models. We apply a usual disease mapping model, producing continuous estimations of the risks that requires a post-processing classification step to obtain clearly delimited risk zones. We also apply a risk partition model that build a classification of the risk levels in a one step procedure. Working with point data, we will focus on the scan statistic clustering method. We illustrate our article with a real example concerning the bovin spongiform encephalopathy (BSE) an animal disease whose zones at risk are well known by the epidemiologists. We show that in this difficult case of a rare disease and a very heterogeneous population, the different methods provide risk zones that are globally coherent. But, related to the dichotomy between the need and the reality, the exact delimitation of the risk zones, as well as the corresponding estimated risks are quite different.