There are two different perspectives on personalized medicine. One perspective is to optimize selecting a treatment for each individual patient among existing treatments, on average. The other perspective, the setting of this presentation, is to develop a new drug that would provide new benefit to a subgroup of patients. That is, patients in this subgroup will do better under this new treatment on average than under the Standard of Care. We will show how to correctly infer efficacy in the subgroup of patients whose biomarker value is above a certain threshold while maintaining the logical relationship among the parameters. We will also show how to find the “optimal” threshold from a new drug development perspective.