Adverse Event Report Classification in Social and Medico-Social Sector

Abstract

Adverse event (AE) analysis is one of the most important missions in French social and medico-social centers, which enables targeted prevention measures to avoid recurrences. However, social and medico-social centers are struggling to conduct an efficient analysis of AEs due to the abysmal quality of event reports. In this paper, we propose an automated classification solution to predict the fact of an AE from the event description to ease the decision-making task of center managers. This solution explores different text similarity methods based on a dedicated terminology of synonyms for the social and medico-social sector in being able to propose the most appropriate facts. The approach has the advantage of being fast with an accurate prediction since it achieved an accuracy up to 88%, when it is tested on 388 real AE reports.