The sequential parallel design consists of two phases of a clinical trial, an initial phase in which patients are randomized to placebo and drug and a second phase in which placebo non-responders are randomized to placebo or drug. The efficiency of this design compared to the conventional two arm trial is examined for both binary and for continuous efficacy data. In the continuous data case, we propose using seemingly unrelated regression to combine the inference over the two phases of the trial. The results of our examinations suggest that for both binary and continuous efficacy data, a reduction of the sample size around 20% is possible over the conventional design. The trade off between smaller sample size versus longer duration of trial is also discussed within the context of an antidepressant clinical trial.
Return to Biostatistics Working Group