Data-Driven Extraction of Quantitative Multi-dimensional Associations of Cardiovascular Drugs and Adverse Drug Reactions

Abstract

Early detection of adverse drug reactions as a part of post-marketing surveillance is very crucial for saving a number of persons from unwanted consequences of drugs. Along with the drug, patient’s traits such as age, gender, weight, location are key factors for occurrence of adverse effects. The relationship between drug, patient attributes and adverse drug effects can be precisely represented by quantitative multi-dimensional association rules. But discovery of such rules faces the challenge of data sparsity because of the large number of possible side effects of a drug and fewer number of corresponding data records. In this paper, to address the data sparsity issue, we propose to use variable support based LPMiner technique for detecting quantitative multi-dimensional association rules. For experimental analysis, data corresponding to three cardiovascular drugs namely Rivaroxaban, Ranolazine and Alteplase has been taken from U.S. FDA Adverse Event Reporting System database. The experimental results show that based on LPMiner technique a number of association rules have been detected which went undetected in case of constant support based apriori and FP-Growth technique.