Injection of automatically selected DBpedia subjects in electronic medical records to boost hospitalization prediction

Although there are many medical standard vocabularies available, it remains challenging to properly identify domain concepts in electronic medical records. Variations in the annotations of these texts in terms of coverage and abstraction may be due to the chosen annotation methods and the knowledge graphs, and may lead to very different performances in the automated processing of these annotations. We propose a semi-supervised approach based on DBpedia to extract medical subjects from EMRs and evaluate the impact of augmenting the features used to represent EMRs with these subjects in the task of predicting hospitalization. We compare the impact of subjects selected by experts vs. by machine learning methods through feature selection. Our approach was experimented on data from the database PRIMEGE PACA that contains more than 600,000 consultations carried out by 17 general practitioners (GPs).