Deep Learning for Identification of Adverse Drug Reaction Relations

The extraction of relations from clinical notes provides relevant information to identify the side effects of medications in postmarketing surveillance. Nowadays, systems based on supervised learning solve relation extraction in clinical records, which requires rich features to learn effective models from the training data. Named Entity Recognition (NER) systems identify medical concepts in clinical notes like Medications and Indications. This work aims to establish if an adverse side effect was caused by taking a specific medication. For this, it is necessary to identify the Adverse Drug Reaction relation, which is the relation between the medical concepts Medications and Adverse Drug Events (ADE). For this purpose, contextual information is extracted via Deep Learning models and other different features were obtained from the relations. The proposed model improves the overall accuracy and the extraction of Adverse relations of the baseline, indicating the effectiveness of combining Deep Learning models and extensive feature engineering.