Informatics in Medicine Unlocked (Jan 2023)

Classifying referring/non-referring ADR in biomedical text using deep learning

  • Yeganeh Mohammadi,
  • Fahimeh Ghasemian,
  • Jaleh Varshosaz,
  • Mohammad Sattari

Journal volume & issue
Vol. 39
p. 101246

Abstract

Read online

An adverse drug reaction (ADR) is a harmful and unwanted disorder that occurs as a result of taking a drug at doses commonly used in humans for the prevention, diagnosis, treatment of disease, or to change physiological function. Reports have shown that about 10% of hospitalizations are due to ADRs. Reporting ADRs has become an essential part of the monitoring and evaluation activities carried out in hospitals. Still, some harmful effects are not reported by drug manufacturers and are unknown to doctors. These effects increase the number of patient hospitalizations. One of the resources that can help in this matter is the huge medical resources available on the web which contain valuable information. Most of these information sources are in the form of text, which should use natural language processing (NLP) techniques for their automatic analysis. The first step in this analysis is to identify texts related to ADR and separate them from unrelated texts. For separating related texts from the massive volume of irrelevant texts, a supervised learning classifier should be used. The purpose of this paper is to apply deep learning models to detect the side effects of drugs in texts using a binary classifier. Deep learning techniques are the state-of-the-art method for text classification. In this article, three approaches based on Recurrent Neural Network (RNN), Gated Recurrent Unit Neural Network (GRU), and Transformer are used for the ADRs identification. The models were trained and evaluated on three different datasets. The results showed that Transformer with 99.12 accuracy is the best model compared to RNN, and GRU. Also, the applied method performs significantly better than previous works.

Keywords