Subset Analysis for Screening Drug–Drug Interaction Signal Using Pharmacovigilance Database

Many patients require multi-drug combinations, and adverse event profiles reflect not only the effects of individual drugs but also drug–drug interactions. Although there are several algorithms for detecting drug–drug interaction signals, a simple analysis model is required for early detection of adverse events. Recently, there have been reports of detecting signals of drug–drug interactions using subset analysis, but appropriate detection criterion may not have been used. In this study, we presented and verified an appropriate criterion. The data source used was the Japanese Adverse Drug Event Report (JADER) database; “hypothetical” true data were generated through a combination of signals detected by three detection algorithms. The accuracy of the signal detection of the analytic model under investigation was verified using indicators used in machine learning. The newly proposed subset analysis confirmed that the signal detection was improved, compared with signal detection in the previous subset analysis, on the basis of the indicators of Accuracy (0.584 to 0.809), Precision (= Positive predictive value; PPV) (0.302 to 0.596), Specificity (0.583 to 0.878), Youden’s index (0.170 to 0.465), F-measure (0.399 to 0.592), and Negative predictive value (NPV) (0.821 to 0.874). The previous subset analysis detected many false drug–drug interaction signals. Although the newly proposed subset analysis provides slightly lower detection accuracy for drug–drug interaction signals compared to signals compared to the Ω shrinkage measure model, the criteria used in the newly subset analysis significantly reduced the amount of falsely detected signals found in the previous subset analysis.