Interim monitoring of bivariate responses using repeated confidence intervals

A method for interim monitoring is described for bivariate response data arising in clinical trials. This procedure centers on a sequence of rectangular joint confidence regions obtained by inverting the hypothesis testing-confidence interval relationship when the testing procedure is governed by a bivariate group sequential design described by Cook and Farewell [4]. The advantage of this type of design over other possible multivariate sequential designs is that the marginal analyses are maintained throughout the trial. This approach proves to be particularly advantageous for active control trials in which a region of therapeutic equivalence is specified for two responses related to efficacy. Two examples illustrate the design and analysis aspects of this approach. The procedure is presented in the context of normally distributed responses; however, adaptations to handle bivariate binary and bivariate failure time data are indicated. Generalizations of this procedure to handle multivariate response data are straightforward.