Signal detection of unexpected adverse events (AEs) in Phase 3 clinical trials is complicated by two conflicting facts: one is the high rate of false positive findings if the per comparison error rate is controlled, and the other is the high rate of false negative findings if a family wise error rate is controlled. Our proposal for detecting safety signals, intended for informal inference, is aimed at mitigating the influence of these high error rates. To illustrate the application of our method we apply it to real data. Limitations of the proposed method are also discussed.

We propose methods for (a) AE visualization to decide how to proceed with additional analyses, and (b) a method for future projections. We also propose controlling the FDR. The AEs flagged for the projected portion may be viewed as additional AEs to pay attention to in making the risk-benefit assessment.