AUTHOR=Noguchi Yoshihiro , Ueno Anri , Otsubo Manami , Katsuno Hayato , Sugita Ikuto , Kanematsu Yuta , Yoshida Aki , Esaki Hiroki , Tachi Tomoya , Teramachi Hitomi 

TITLE=A New Search Method Using Association Rule Mining for Drug-Drug Interaction Based on Spontaneous Report System

JOURNAL=Frontiers in Pharmacology

VOLUME=Volume 9 - 2018

YEAR=2018

URL=https://www.frontiersin.org/journals/pharmacology/articles/10.3389/fphar.2018.00197

DOI=10.3389/fphar.2018.00197

ISSN=1663-9812

ABSTRACT=Background: Adverse events (AEs) can be caused not only by one drug but also by the interaction between two or more drugs. Therefore, clarifying whether an AE is due to a specific suspect drug or drug-drug interaction (DDI) is useful information for proper use of drugs. Whereas previous reports on the search for drug-induced AEs with signal detection using spontaneous reporting systems (SRSs) are numerous, reports on drug interactions are limited. This is because in methods that use “a safety signal indicator” (signal), which is frequently used in pharmacovigilance, a huge number of combinations must be prepared when signal detection is performed, and each risk index must be calculated, which makes interaction search appear unrealistic.
Objective: In this paper, we propose association rule mining (AR) using large dataset analysis as an alternative to the conventional methods. 
Methods:The data source used was the Japanese Adverse Drug Event Report (JADER) database. The combination of drugs for which the risk index is detected by the “combination risk ratio” as the target was assumed to be true data, and the accuracy of signal detection using the AR methods was valuated in terms of sensitivity, specificity, Youden's index, positive predictive value (PPV), negative predictive value (NPV), F-score and area under the receiver operating characteristic curve (AUC).
Results: Our experimental results indicate that AR has a sensitivity of 99.05%, specificity of 92.60%, Youden’s index of 0.917, PPV of 78.57%, NPV of 99.72% and F-score of 0.876 Further, the AUC obtained was 0.959.
Conclusions: If a similar calculation method that simply creates combinations from a database was used instead of AR, the number of combinations would be so enormous that it would be difficult to perform the calculations within a realistic time. However, in the AR method, the “Apriori algorithm” is used to reduce the number of calculations. Thus, the proposed method has the same detection power as the conventional methods, with the significant advantage that its calculation process is simple.