| Abstract: | Suppose that a clinical trial has been carried out using a sequential design. Any analysis that neglects the monitoring of the trial is potentially biased, and this holds for analyses of responses not directly used to determine whether the study should be stopped. Although valid methods exist to deal with the responses actually used in the sequential design, little has been written about secondary analyses of further responses. In this paper, two approaches to secondary analyses are explored. One is a conditional approach that avoids the problems of bias, the other an unconditional approach that allows for the sequential nature of the trial. The two methods are illustrated by application to three examples, each concerning a different type of response variable. |
