Sequential testing in clinical trials with repeated measurements

Abstract

In some clinical trials, the number of subjects on each treatment is fixed and at successive equally-spaced time points each subject yields a measurement which reflects response to treatment. A model is presented for such repeated measurements data. A sequential procedure is proposed for testing the null hypothesis of no difference in mean effect between two treatments. In this procedure, the maximum number of measurements per subject and absorption probabilities for intermediate significance tests are preassigned to yield a required approximate overall significance level. The jth significance test follows thejth measurement on each subject, and the test depends on values of model parameters estimated from all the data available so far. Simulations show that the proposed procedure yields overall significance levels close to 5% when required, for data from a variety of models.