Detecting Potential Serious Adverse Drug Reactions Using Sequential Pattern Mining Method

Adverse drug reactions (ADRs), defined as noxious and unintended responses to drugs at normal doses, cause hospitalization and death all around the world, and cost billions of dollars every year. Different techniques have been explored to discover potential ADRs using materials from various sources, such as spontaneous reporting systems, electronic medical records and clinical related social media data. In this paper, we proposed a framework that combines sequential data mining method Prefix-Span and classic disproportionality based method Proportional Report Ratio (PRR) to detect potential serious ADRs, taking temporal information among clinical events into consideration. And Naranjo algorithm, a questionnaire designed to determine the likelihood of causal relationships between drugs and reactions, is introduced to verify the validity of the results we obtain. In the end, we got 127 pair-wise frequent patterns of drugs and reactions, and 52% of the drugs overlap with that of the Naranjo algorithm. And the comparison of those two approaches shows that detecting potential ADRs based on disease category may increase the accuracy and creditability of the results.