Adverse drug event detection and extraction from open data: A deep learning approach. Issue 1 (January 2020)

• Record Type: Journal Article
• Authors: Fan, Brandon; Fan, Weiguo; Smith, Carly; Garner, Harold "Skip"
• Is Part Of: Information processing & management. Volume 57:Issue 1(2020:Jan.)
• Publication Date: 2020-01
• Subjects: Information extraction -- Deep learning -- Pharmacovigilance -- Drug side effects -- Open data -- BERT -- Natural language processing; Information storage and retrieval systems -- Periodicals; Information science -- Periodicals; Systèmes d'information -- Périodiques; Sciences de l'information -- Périodiques; Information science; Information storage and retrieval systems; Periodicals; 658.4038
• DOI: 10.1016/j.ipm.2019.102131
• Languages: English
• ISSNs: 0306-4573

Abstract

Highlights: Previous pharmacovigilance research fails to accurately discover drug side effects. We introduce a novel adverse drug event extraction algorithm using deep learning. We introduce the use of novel contextual word and sentence embeddings. Results show that our model outperforms current pharmacovigilance models. This model can be applied to a wide variety of information extraction tasks. Abstract: Drug prescription is a task that doctors face daily with each patient. However, when prescribing drugs, doctors must be conscious of all potential drug side effects. In fact, according to the U.S. Department of Health and Human Services, adverse drug events (ADEs), or harmful side effects, account for 1/3 of total hospital admissions each year. The goal of this research is to utilize novel deep learning methods for accurate detection and identification of professionally unreported drug side effects using widely available public data (open data). Utilizing a manually-labelled dataset of 10, 000 reviews gathered from WebMD and Drugs.com, this research proposes a deep learning-based approach utilizing Bidirectional Encoder Representations from Transformers (BERT) based models for ADE detection and extraction and compares results to standard deep learning models and current state-of-the-art extraction models. By utilizing a hybrid of transfer learning from pre-trained BERT representations and sentence embeddings, the proposed model achieves an AUC score of 0.94 for ADE detection and an F1 score of 0.97 for ADE extraction. Previous state of the art deep learning approach achieves an AUC of 0.85 in ADE detection and an F1 of 0.82 in ADE extraction on our dataset of review texts. The results show that a BERT-based model achieves new state-of-the-art results on both the ADE detection and extraction task. This approach can be applied to multiple healthcare and information extraction tasks and used to help solve the problem that doctors face when prescribing drugs. Overall, this research introduces a novel dataset utilizing social media health forum data and shows the viability and capability of using deep learning techniques in ADE detection and extraction as well as information extraction as a whole. The model proposed in this paper achieves state-of-the-art results and can be applied to multiple other healthcare and information extraction tasks including medical entity extraction and entity recognition.