This Article Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Wheatley, K. Articles by Burnett, A. K. Search for Related Content PubMed PubMed Citation Articles by Wheatley, K. Articles by Burnett, A. K.

Problems With Up-Front Randomization in Clinical Trials

University of Birmingham, Birmingham, United Kingdom

Robert K. Hills, Alan K. Burnett

Cardiff University, Cardiff, United Kingdom

To the Editor:

We are concerned with the methodology used by Büchner and colleagues¹ in their recent report on a randomized trial in acute myeloid leukemia (AML). Patients younger than the age of 60 years were randomly assigned at diagnosis to receive either prolonged maintenance therapy or autologous stem-cell transplantation (SCT) after three courses of induction/consolidation chemotherapy. We have previously commented on the problems with such "up-front randomization" from a hypothetical perspective,² but hard data are now available to demonstrate the limitations of such an approach. Büchner et al report that 411 and 429 patients were assigned to maintenance and autologous SCT, respectively; overall, 70% of patients achieved complete remission (CR)—data were not given for each group separately; in the maintenance arm, 51% of those patients who achieved CR received maintenance treatment (ie, assuming a CR rate of 70% in this group, only 36% of the total randomly assigned actually complied with treatment allocation); in the autologous SCT arm, compliance was only 24%. Inspection of the survival curve (Fig 3A) shows that approximately 20% of patients in both arms died during the first 3 months (ie, during the induction/consolidation phase). These deaths represent approximately 40% of all the deaths in the study and, occurring as they do during a period where there is no possibility of maintenance and autologous SCT having differential affects, simply add a high level of unwanted noise to the comparison. An important consideration in the design of any randomized trial is to keep noncompliance to a minimum because poor compliance will severely reduce the power of the trial to detect a difference between arms, potentially leading to false-negative results. With high levels of noncompliance, there is a danger that clinically worthwhile treatment effects will be missed (there is no evidence that this is the case in the trial by Büchner et al, but see the worked example in Hills et al²) and patients will be denied effective therapies as a result of suboptimal trial design. One important way in which to minimize noncompliance is to perform the random assignment as close as logistically feasible to the point at which treatments diverge³ (ie, not at diagnosis for postremission comparisons).

Büchner et al also report the nonrandomized comparison between patients by postremission therapy actually administered, and report a significant survival advantage for maintenance (P = .005)—a result in the opposite direction to the unbiased, although underpowered, intention-to-treat analysis, which was nonsignificantly (P = .52) in favor of autologous SCT. This analysis will be potentially subject to the same selection biases as any nonrandomized comparison. In this case, there is evidence that the two groups are selected, and thus not comparable, from the different compliance rates (36% v 24% as herein).

Büchner et al have argued⁴ that up-front randomization is clinically more relevant than later randomization. We strongly disagree with this view. The question of whether to give maintenance therapy or autologous SCT after three courses of induction/consolidation chemotherapy is only relevant to patients who have successfully completed these three courses, are fit enough to receive either treatment and are capable of receiving either treatment (for example, in the case of autologous SCT have had a successful harvest). The question of which of these two postremission treatment options is best for patients who fail to achieve CR, and in many cases are dead, is clinically irrelevant, as well as being conceptually difficult to envisage. Similarly, the question is irrelevant to those patients who relapse early, die in CR during consolidation chemotherapy, have excessive toxicity precluding further treatment, or are scheduled for allogeneic SCT.

Hence, the concept of up-front randomization is seriously flawed, both clinically and statistically, and we would strongly urge that this design not be pursued in other trials.

Authors' Disclosures of Potential Conflicts of Interest

The authors indicated no potential conflicts of interest.

REFERENCES

1. Büchner T, Berdel WE, Schoch C, et al: Double induction containing either two courses or one course of high-dose cytarabine plus mitoxantrone and postremission therapy by either autologous stem-cell transplantation or by prolonged maintenance for acute myeloid leukemia. J Clin Oncol 24:2480-2489, 2006[Abstract/Free Full Text]

2. Hills RK, Richards SM, Wheatley K: Corner cutting compromises clinical trials: The inherent problems with up-front randomisation and a common standard arm. Leuk Res 27:1071-1073, 2003[CrossRef][Medline]

3. Schulz KF, Grimes DA: Sample size slippages in randomized trials: Exclusions and the lost and wayward. Lancet 358:781-785, 2002[CrossRef]

4. Büchner T, Döhner H, Ehninger G, et al: Up-front randomisation and common standard arm: A proposal for comparing AML treatment strategies between different studies. Leuk Res 26:1071-1073, 2002

This Article Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Wheatley, K. Articles by Burnett, A. K. Search for Related Content PubMed PubMed Citation Articles by Wheatley, K. Articles by Burnett, A. K.