Modeling, building and evaluating an ontology for the automatic characterization of adverse drug effects during pharmacovigilance.

Abstract : BACKGROUND: The characterization of spontaneous reported cases is fundamental for pharmacovigilance. This task is time consuming and its reproducibility is low. OBJECTIVE: To develop a system founded on an ontology that automatically instantiates spontaneous reported cases as "known" adverse drug effects (ADE) only if the reported ADEs are described in drug compendia. METHODS: A simple ontology of drugs and their related adverse effects represented in description logics was developed from a drug database. Manual evaluation was carried out on 378 spontaneous reported cases instantiated as "known ADE of a chemical class". The initial manual characterization was reviewed by a pharmacovigilance expert to validate the generated automatic characterization. RESULTS: The ontology is composed by 57,04 concepts and 5 relations. It was successfully validated thanks to Pellet reasoner and it contains neither inconsistencies nor cycles. In this validation, 86% of the instantiated spontaneous reported cases effectively concerned notorious ADEs, whereas only 75% were initially identified manually as related to notorious ADEs. CONCLUSION: This system can assist characterization by applying a reasoning process similar to that used by experts in the search for ADEs.