Two-Phase, Three-Stage Adaptive Designs in Clinical Trials

In clinical investigator initiated clinical trials, we frequently encounter the situation where it is very difficult to estimate the effect size and the clinically meaningful difference between the treatment and control groups. In this paper we explore various two-phase, three-stage adaptive designs which can be applied to this situation. The first phase determines whether the trial should proceed or not. If the decision is to proceed, then the sample size is re-estimated. The second phase consists of two stages, but the sample size is not re-estimated. We introduce hybrid and alpha-split designs, adding to two existing adaptive designs: Bauer-Köhne design and Lehmacher-Wassmer design. Main findings are: 1) the differences in the overall powers and the average sample number (ASN)s among these designs are small, except for the design which includes O’Brien-Fleming boundaries and the alpha-split design, 2) the two-phase, three-stage design suffers a relative loss of power by 15% but the ASN is less than 50%, as compared to the single stage design under the optimal condition, 3) two-phase, three-stage design compares with the three-stage group sequential design. We conclude that the design can be a candidate when there is no useful information on the effect size.