**5. Conclusions**

In recent years, the need for safety signal screening has been demanded, not only for single drugs but also for drug–drug interactions. Although several methods for detecting signals of drug interaction have been reported, it is di fficult to say that these methods are used because many of them are complicated in calculation. Therefore, there were several cases [11–13] where subset analysis using the algorithm for detecting signals of single drugs (e.g., ROR [6]) was used for signal detection of drug–drug interactions before its validity was verified.

This study showed that there were many false positives in the existing subset analysis, albeit under limited conditions. Additionally, very simple modifications of the detection criteria were made to solve two problems associated with the previous subset analysis for exploring recently reported drug–drug interaction signals. This modification helped to reduce falsely detected signals found in the previous subset analysis.

Moreover, the newly proposed subset analysis is more similar to the Ω shrinkage measure model than the previous subset analysis, but the similarity with the Ω shrinkage measure model is not as high. However, the newly proposed subset analysis showed that although the detection accuracy of the drug–drug interaction signal was slightly lower than that of the Ω shrinkage measure model, the detection accuracy was su fficient. This result may also indicate the possibility of detecting signals that cannot be detected by the Ω shrinkage measure model.

**Author Contributions:** Conceptualization, Y.N.; funding acquisition, Y.N.; investigation, Y.N.; methodology, Y.N. and T.T.; project administration, Y.N. and H.T.; supervision, Y.N.; validation, Y.N. and T.T.; visualization, Y.N.; writing—original draft, Y.N.; writing—review and editing, Y.N., T.T., and H.T. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by JSPS KAKENHI gran<sup>t</sup> number 19K20731.

**Conflicts of Interest:** The authors declare no conflict of interest.
