Detecting Medications and Adverse Drug Events in Clinical Notes Using Recurrent Neural Networks

[edit]

Xi Yang, Jiang Bian, Yonghui Wu ;
Proceedings of the 1st International Workshop on Medication and Adverse Drug Event Detection, PMLR 90:1-6, 2018.

Abstract

Early detection of Adverse Drug Events (ADEs) from Electronic Health Records (EHRs) is an important, challenging task to support pharmacovigilance and drug safety surveillance. The authors present a Recurrent Neural Network (RNN)-based system to detect medication name and its attributes (dosage, frequency, route, duration), as well as mentions of ADEs, Indications, other signs and symptoms from clinical notes. We developed an RNN-based Named Entity Recognition (NER) system implemented using Long-Short Term Memory (LSTM). Two NER models, RNN-1 and RNN-2, were developed using different training strategies. Both of the two models only utilized pretrained word embeddings provided by organizers without any extra feature engineering. The RNN-2 model achieved a top 3 performance (F1-score of 0.8233) for sub-task 1, demonstrating the efficiency of RNN for clinical NER tasks.

Related Material