Editors' ChoiceBiostatistics

"Drop-the-Loser" Designs—An Alternative Design for Phase III Studies

See allHide authors and affiliations

Science Translational Medicine  02 Jun 2010:
Vol. 2, Issue 34, pp. 34ec88
DOI: 10.1126/scitranslmed.3001315

The ultimate prize in clinical research is a phase III clinical trial in which the treatment under investigation is shown to improve clinical outcomes as compared with those of a placebo treatment. Given the huge weight placed on the results of a phase III trial for regulatory approval and changes in clinical practice, a sound study design is of critical importance. All too often, however, at the end of a phase II study it is difficult to know exactly how to design the phase III trial. One particularly thorny aspect is the choice of the appropriate dose of the drug. This is because surrogate markers are often used as the measured outcome in phase II trials and the sample sizes are small, making conclusions about the safety of various doses difficult. Another way to address this issue is called “drop-the-loser” trial design. In these drop-the-loser designs, several doses are investigated in phase III trials, and the ineffective or unsafe doses may be dropped at an interim analysis point in the trial. How to appropriately set the efficacy and futility criteria used to evaluate trial data before the end of the study—also classified as interim analysis stopping rules—has been hard to work out. Recently, Chen et al. offered a solution to this issue. These authors have now described how to compute the futility and efficacy boundaries. They investigated a variety of approaches, including joint monitoring, in which the interim decision is computed for all doses simultaneously, and marginal monitoring, in which the type I error for each dose is assigned at the start of the study and each dose group is monitored separately. They show that the strengths of marginal monitoring are outweighed by the computer-intensive nature of the calculations. Thus, marginal monitoring, although slightly more conservative, provides a computationally feasible alternative. Chen et al. also compare the drop-the-loser designs with the traditional single-dose–versus-placebo clinical trial. As expected, when there is a clearly superior dose, phase II studies will likely reveal this, and a traditional clinical trial will have a greater ability to find a treatment effect when it exists. But in other situations—such as when phase II clinical trial data are ambiguous regarding the optimal dose for the phase III study—drop-the-loser designs offer a useful alternative.

Y. H. J. Chen et al., Some drop-the-loser designs for monitoring multiple doses. Statistics in Medicine 20 May 2010 (10.1002/sim.3958). [Abstract]

Navigate This Article