Skip to Main content Skip to Navigation

Adverse drug reactions detection in clinical notes

Abstract : The Information Extraction from clinical notes provides relevant information to identify adverse side effects in post-marketing surveillance of medications (Pharmacovigilance), which is more difficult to discover by traditional medical studies since patients are taking several treatments at the same time. In recent years, data mining techniques have allowed to discover knowledge stored in big datasets, such as the clinical records collected by hospitals throughout patient's life. The goal of this work is identify adverse side effects caused by treatments. Then, we have to identify relations between medications and Adverse Drug Events (ADE) entities, which is called Adverse Drug Reaction relation. This problem is divided Named Entity Recognition (NER) and Relation Extraction tasks. Nowadays, supervised approaches based on Deep Learning and Machine Learning algorithms solve this problem in the state of the art. These supervised systems require rich features in order to learn efficient models during training, therefore, we focus on building comprehensive word representations (the input of the neural network), using character-based word representations and word representations. The proposed representation improves the performance of the baseline model, and the final model reached the performances of state of the art methods. Then we have extracted contextual information through Deep Learning models and other different features obtained from the relations, in order to identify the Adverse Drug Reaction relations. The proposed model improved the overall accuracy and the extraction of Adverse Drug Reaction compared to the baseline, indicating the effectiveness of combining Deep Learning models and extensive feature engineering.
Complete list of metadata
Contributor : Abes Star :  Contact
Submitted on : Monday, February 8, 2021 - 5:25:17 PM
Last modification on : Tuesday, February 9, 2021 - 3:18:07 AM
Long-term archiving on: : Sunday, May 9, 2021 - 8:24:15 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03135102, version 1



Edson Alejandro Florez Suarez. Adverse drug reactions detection in clinical notes. Data Structures and Algorithms [cs.DS]. Université Côte d'Azur, 2020. English. ⟨NNT : 2020COAZ4034⟩. ⟨tel-03135102⟩



Record views


Files downloads