Apports d’une méthode de fouille de données pour la détection des cancers du sein incidents dans les données du programme de médicalisation des systèmes d’information

The objective of this work was to assess the interest of a data mining approach to detect incident breast cancer cases in medico administrative data. Data from the French casemix system (PMSI) were linked with the Isere cancer registry, which was the gold standard to define incident breast cancer. Formal Concept Analysis (FCA) was used to compute combinations of attribute values in the PMSI that could further define algorithm of detection of incident breast cancer. FCA allowed to automatically evaluate any possible combination of attribute values in terms of sensibility and Positive Predictive Value. This method can help experts in quality assessment of medico-economical databases as epidemiological tools.