Since in health care systems the amount of data is continuously growing, data mining techniques can be applied to analyse these large collections and gain interesting insights. However, some critical issues should be properly addressed. For example, data collections on patient treatments are usually characterized by an inherent sparseness and variable distribution, due to the large variety of possible treatments performed by patients affected by a given disease. To effectively extract

more »

... g knowledge from such collections, we present a framework coupling a clustering and a classification algorithm. The clustering approach is named multiple-level because we apply the clustering algorithm in a multiple-level fashion, by focusing on different dataset portions and locally identifying groups of patients with similar profile and examination history. The classification algorithm is used to characterize the discovered clusters. This paper also describes future research issues and possible developments of the proposed framework.