An innovative framework for extracting adverse drug reactions of single medication and combined medications from medical transcriptions and online reviews

Abstract

Adverse drug reactions (ADRs) are unintentional and detrimental reactions arising due to normal drug usage. Identifying ADRs is vital in spheres of health and pharmacology. ADRs occur due to a single drug or a combination of multiple drugs. In the pharmaceutical industry, recognizing this type of medication interactions is viewed as a significant task. In this paper, we discuss the extraction of ADRs from combined medications (two drugs) by using medical transcripts and online reviews as the primary sources. Here, Natural Language Processing (NLP) techniques are combined with weighted association rule mining for extracting ADRs due to a single drug from medical transcripts. Single drug ADRs are also obtained from online health reviews using an ensemble classifier. These drugs along with their ADRs are used for constructing two-drug (combined medication) associated ADRs dataset. Further, by using the dataset of combined medications, the interaction of the medications and the reactions that are associated with that drug combination are predicted. In the first two phases of single drug associated ADR prediction, weighted association rule mining and ensemble classifier got an accuracy of 88%. The proposed model obtained an accuracy of 85.3%.

Keywords

• adrs
• ddis
• text mining
• weighted association rule mining
• nlp
• keybert