Développement d'un système de détection des infections associées aux soins à par tir de l'analyse de comptes-rendus d'hospitalisation

This paper describes the first version and the results obtained by a system which detects occurrences of healthcareassociated infections. The system automatically analyzes hospital discharge summaries coming from different hospitals and from different care units. The output of the system consists in stating for each document, if there is a case of healthcare-associated infection. The linguistic processor which analyzes hospital discharge summaries is a general purpose tool which has been adapted for the medical domain. It extracts textual elements that may lead to an infection suspicion. Jointly with the extraction of suspicious terms, the system performs a negation and temporal processing of texts in order to refine the extraction. We first describe the system that has been developed and give then the results of a preliminary evaluation. Mots-cles : Extraction d’information medicale, compte-rendus d’hospitalisation, infection nosocomiale, analyse syntaxique