George Y.H. Chi (FDA, USA)
Clinical Decision Rules and Multiple Endpoints
In clinical drug trials, one often observes a lack of a clear decision rule that is used to assess the effect of the drug based on the final outcome of the trial. The actual decision rule used may be either ad-hoc or post-hoc. Even in cases where a decision rule was defined, there may be a lack of optimality from the clinical perspective, or a lack of proper statistical support structure from the statistical perspective in the decision rule. These may lead to serious inflation of type I error, either assessible or unassessible. This lack of a clear decision rule, or lack of optimality in the decision rule is essentially related to the problem of multiple endpoints. The purpose of this presentation is to illustrate some of these problems with practical examples, and to propose that a rational way of dealing with the multiple endpoints problem is to define clinical decision rules with proper statistical support structures that will provide the necessary basis for making valid statistical inference.