|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Originally published as JCO Early Release 10.1200/JCO.2005.10.956 on February 7 2005 © 2005 American Society of Clinical Oncology.
Randomized Phase II Trials: What Does Randomization Gain?University of Pittsburgh, Pittsburgh, PA In this issue of the Journal of Clinical Oncology, Dark et al1 present the results of a trial of two intravenous schedules of a liposomal topoisomerase I inhibitor. The design they used, which was proposed by Simon et al,2 called for concurrent randomization between the schedules in what the authors describe as a "pick the winner statistical design." They carefully spelled out the details of the design and followed it closely, representing a successful application of a randomized phase II trial. Because this trial design may not be familiar to some investigators, we believed that an additional discussion of the properties of this and other randomized phase II designs would be useful. Generally, the randomization for phase II trials is between two regimens, tumor response is the end point of interest, and the goal is to determine whether one experimental regimen should be brought forward for further testing. The subsequent trial would usually be a randomized phase III trial designed to compare the experimental regimen to a standard regimen, with a more traditional end point such as survival. Randomized phase II designs can also be used with more than two regimens; multiregimen trials are sometimes discussed in the context of selection trials.3 In this discussion, I will assume that only two regimens are being evaluated. In the abstract of their article, Dark et al1 refer to their design as a pick the winner design. In fact, the first part of such a design is to perform the equivalent of a two-stage, single-arm, phase II trial, such as the Simon4 design, for each regimen. The benefit of this approach is that there is randomization between the two treatments, allowing a greater degree of comparability than if there were two independent phase II trials in a seemingly similar population of patients. The pick the winner role only occurs if both regimens are efficacious. However, this trial design is not as simple as it seems and is a multistep process. Step 1 is to identify two experimental regimens that are to be evaluated in a phase II setting. In Dark et al, 1 the experimental regimens were two regimens involving the same therapies but with different schedules. Step 2 is to determine the historical response rate (R0), determine a response rate that would be sufficiently high to warrant further study of a regimen (R1), and identify a design for a two-stage, single-arm, phase II study that will have a small probability (usually P = .05) of concluding that a regimen merits further study when the response rate for the regimen is no better than the historical rate (R0), but a high probability (usually P = .80 or .90) of correctly concluding that the regimen merits further study when the true response rate is R1 (the power). The design will specify the number of patients to be accrued at each stage, the number of responses required to continue the study at the end of the first stage, and the number of responses required at the end of the trial to conclude that the regimen merits further study. Step 3 is as follows. When the trial opens, patients are randomly assigned to regimen A or B until each regimen has the desired number of patients for the first stage. In the Dark et al1 study, this number was 20. When response information has been obtained for all the patients entered onto the first stage, it is determined for each regimen whether enough responses have been observed to continue to a second stage (in the Dark et al trial, this number was one response). If both regimens fail to meet this criterion, the trial is stopped, and neither regimen has sufficient activity to warrant further study. If one regimen fails to meet this criterion, accrual to that regimen is stopped, but accrual is continued to the second regimen. If both regimens meet the criterion for continuation, patients are randomly assigned to both arms until the trial is completed. Step 4 is as follows. At the end of the study, if less than a certain predetermined number of responses are observed for each regimen, neither regimen merits further evaluation. If one regimen has at least this certain number of responses, then that regimen merits further study, which was the case in the Dark et al1 trial. In such a case, there is no pick the winner aspect to this design, and the conclusions are those that could be obtained from an individual phase II trial. If both regimens have at least the certain number of responses but, for practical reasons, only one regimen can be carried to further testing, the winner is chosen by integrating information such as response rates, toxicity, cost, and quality of life. There is no formal comparison of response rates, and no claim can be made that one regimen is more efficacious than the other because both regimens have passed the test for efficacy. This approach could also be used in a setting in which both regimens would be carried forward if their response rates were sufficient to warrant further study. A variation on the design described by Dark et al1 calls for randomization between an experimental regimen and a standard regimen. In this design, the experimental regimen is evaluated in the usual manner for a phase II trial, as if there were no concurrently randomized arm. If the patients randomly assigned to the standard arm have response rates similar to those reported historically, this would make one more confident that the rates observed in the phase II trial were not caused by a biased sample. Conversely, if the response rates observed in concurrently randomly assigned control patients are inconsistent with historical control data, it would be a warning that the results observed for the experimental regimen should be viewed with caution. Again, there is no formal comparison of the two regimens. Johnson et al5 presented the results of a randomized phase II design that called for a formal comparison of regimens at the conclusion of the trial. This design is used less frequently because it requires roughly four times as many patients as a single-arm trial and twice as many patients as a trial using the approach of Dark et al1 to achieve the same power. For example, the design for the Johnson et al trial5 required 33 patients for each of three arms but was powered for a huge difference in response rates (27% response rate for controls v 52% for the combined regimens) for the comparison of the control regimen to an experimental regimen at a power of 0.80. In fact, the design for the Johnson et al1 trial was more complex than a randomized phase II design for response because there were two experimental treatment arms (different doses of bevacizumab) and a control arm, there were two primary end points (response and time to progression), and the power was based on finding a difference between the combined experimental regimens (66 patients) and the control regimen. Both end points required large differences (27% response rate for controls v 52% for the combined regimens, and a doubling of the median time to progression for the combined regimens v the controls). When the trial was completed, neither of these differences was attained, and there were no statistically significant differences observed, although a secondary analysis revealed a borderline improvement in time to progression for the higher dose of bevacizumab compared with the control regimen. The authors did not claim that the results were statistically significant and reported the results as suggestive; a validation study is currently underway. Randomized phase II designs have the attractive property of reducing selection bias. The comparative randomized phase II trial has the additional property that its primary analysis might identify a statistically significant difference between response rates for two regimens. This would not obviate the need for a subsequent randomized trial to assess more clinically relevant outcomes, such as survival and quality of life, but would allow one to be confident that the treatment carried forward was the more active treatment for response. However, the comparative randomized phase II design has a potential drawback because it is often powered to detect a substantial benefit and may, therefore, fail to identify an experimental regimen with only moderate (but possibly clinically relevant) improvement over standard therapy. A second potential benefit of the randomized design is that the randomized cohort allows for unbiased comparisons in secondary analyses. However, the potential for false positives increases as the number of secondary analyses increases. If secondary analyses are performed, they should be reported with extreme caution and clearly identified as exploratory, particularly if end points such as progression-free or overall survival are compared. Findings from these analyses cannot replace a properly designed phase III trial. Author's Disclosures of Potential Conflicts of Interest The author indicated no potential conflicts of interest. REFERENCES
1. Dark GG, Calvert AH, Grimshaw R, et al: Randomized trial of two intravenous schedules of the topoisomerase I inhibitor liposomal lurtotecan in women with relapsed epithelial ovarian cancer: A trial of the National Cancer Institute of Canada Clinical Trials Group. J Clin Oncol 23:1859-1866, 2005 2. Simon R, Wittes RE, Ellenberg SS: Randomized phase II clinical trials. Cancer Treat Rep 69:1375-1381, 1985[Medline] 3. Steinberg SM, Venzon DJ: Early selection in a randomized phase II clinical trial. Stat Med 21:1711-1726, 2002[CrossRef][Medline] 4. Simon R: Optimal two-stage designs for phase II clinical trials. Control Clin Trials 10:1-10, 1989[Medline]
5. Johnson DH, Fehrenbacher L, Novotny WF, et al: Randomized phase II trial comparing bevacizumab plus carboplatin and paclitaxel with carboplatin and paclitaxel alone in previously untreated locally advanced or metastatic non-small-cell lung cancer. J Clin Oncol 22:2184-2191, 2004
Related Article
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2005 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
