We read with interest the article by Marc Buyse and Damien McEntegart (ACT, May 2004, pp. 36-40). To try to achieve a more balanced discussion of the recent European regulatory guidance in question, we would like to comment on several of the points that they raised.

"The CPMP's position discouraging dynamic allocation techniques is unfair." Regulatory advice applies to all companies (large and small, European-based or not) and so is not particularly directed at one sector of the pharmaceutical industry rather than another. The regulatory guidance in question applies to all potential applicants. In addition, an extensive public consultation process is part of the process of developing European regulatory guidance of this type. So it is difficult to see in what sense it is "unfair."

To say that the guidance does not cite any references to support its views is an irrelevant argument. The document cites no references at all--either for aspects that Buyse and McEntegart may agree with or for points that they do not agree with. Absence of references is normal in these regulatory guidance documents. If Buyse and McEntegart's criticism were to be "fair," then on this basis they would also have to reject all the advice written in regulatory guidance with which they do agree. However, they do not dismiss advice in this way. On the contrary, to support their viewpoint, they use other regulatory guidance that also cites no references.

Buyse and McEntegart quote references that support the use of minimization but they do not quote other references that are more conservative. The correspondence between Taves¹ and Atkinson,^2,3 for example, shows that the scientific community is not of one mind regarding the use of covariate-adaptive randomization procedures. Also Rosenberger and Lachin⁴ cautiously state that "very little is known about its theoretical properties." This is a substantial point. The direct theoretical link between randomization and methods of statistical analysis has provided a solid foundation for reliable conclusions from clinical trial work for many years.

We have all been taught to beware of simple apparently random allocation methods that may mislead us. If we use minimization methods then we rely, in whole or in part, on potentially fallible judgements rather than theoretical certainty. It is true that studies using permuted block randomization suffer from similar problems when analyzed without incorporating the blocking factor in the analysis. This indicates a deficiency of permuted blocks rather than a strength of minimization. However, the potential influence of blocking is relatively easy to understand and the correction of the analysis is generally easier to implement than is the case for minimized designs.

The CONSORT statement⁵ is not principally concerned with trial design and conduct but rather with trial reporting. We wholly agree that the method of assigning treatments to patients should be clearly described in a study report. If trialists have used minimization, then we would agree that this should be explicitly described (including details of exactly what form of minimization has been used) in the study report.

Buyse and McEntegart's reference to the editorial by Treasure and MacRae in the British Medical Journal⁶ says that it drew no adverse response. In fact, for example, Atkinson commented "Despite this encomium, the comparison given here shows that this particular scheme is inadmissible."² Furthermore, recent papers presented at the Society for Clinical Trials and International Society for Clinical Biostatistics meeting in London in 2003^7-9 have begun to describe how blinding may be compromised with either minimization or more general covariate-adaptive procedures.

Buyse and McEntegart claim that the CPMP Guidance and the International Conference on Harmonisation Guidance, ICH E9, "Statistical Principles for Clinical Trials"¹⁰ conflict with each other. We do not understand why this is the case. ICH E9 does not specifically support the use of dynamic allocation--it simply describes the fact that it can be done and advises on ways to make it more acceptable if you do plan to use it. The final sentence in section 2.3.2 of that document points out the possible complexity of logistics and the potential difficulties with the analysis. These rather neutral comments are in stark contrast to other areas of ICH E9 where much stronger language is used to actively support approaches to clinical trials that the guidance favors.

Dangers of minimization
On the point about logistics and practical complexity, when reviewing applications from sponsor companies, we have often observed that the use of minimization may result in more harm than good. In the context of confirmatory Phase III studies, we have rarely seen the need to use any allocation procedure more complex than simple stratified randomization with permuted blocks. However, in circumstances where sponsors have chosen to use minimization, we have seen situations where programming algorithms have been incorrect, the choice of factors to include in the minimization algorithm has been poorly thought out or where telephone systems or Web-based systems have proved unreliable. Such examples occur across large and small companies and so our general advice to all companies would be to avoid such a procedure since it seems to add no benefit but has potential difficulties.

Finally, we should point out that the Points to Consider document also discourages the use of too many covariates and models that become too complex. Dynamic allocation procedures are said to be beneficial when the number of strata becomes large. However, since the guidance argues for including stratification factors in the statistical model, and it discourages a large number of factors in that model, then only a few stratification factors should be used. Hence, the guidance is being perfectly consistent when it questions the place of covariate adaptive randomization schemes.

References
1. D.R. Taves, "Faulty Assumptions in Atkinson's Criteria for Clinical Trial Design," Journal of the Royal Statistical Society Series A, 167, 179-180 (2004).
2. A.C. Atkinson, "The Comparison of Designs for Sequential Clinical Trials with Covariate Information," Journal of the Royal Statistical Society Series A, 165, 349-373 (2002).
3. A.C. Atkinson, Author response, Journal of the Royal Statistical Society Series A, 167, 180-181 (2002).
4. W.F. Rosenberger and J.M. Lachin, Randomisation in Clinical trials. Theory and Practice (New York, Wiley, 2002).
5. D.G. Altman et al., CONSORT Group (Consolidated Standards of Reporting Trials) "The Revised CONSORT Statement for Reporting Randomised Trials: Explanation and Elaboration," Annals of Internal Medicine, 134, 663-694 (2001).
6. T. Treasure and K.D. MacRae, "Minimisation: The Platinum Standard for Trials?," British Medical Journal, 317, 362-363 (1998).
7. R. Hills, R. Gray, K. Wheatley, "High Probability of Guessing Next Treatment Allocation with Minimisation by Clinician (abstract)," Controlled Clinical Trials, 24, 70S (2003).
8. G. McPherson, M. Campbell, D. Elbourne, "Minimisation: Predictability Versus Balance (abstract)," Controlled Clinical Trials, 24, 133S (2003).
9. N. Scott and G. McPherson, "Minimisation in Multicentre Clinical Trials (abstract)," Controlled Clinical Trials, 24, 175S (2003).
10. International Conference on Harmonisation, E9 document, Guidance on Statistical Principles for Clinical Trials.

Simon Day,* PhD, is with Medicines and Healthcare products Regulatory Agency, Room 13-205, Market Towers, 1 Nine Elms Lane, London SW8 5NQ, UK, e-mail simon.day@mhra.gsi.gov.uk. Jean-Marie Grouin, PhD, is at Université de Rouen, rue Lavoisier, 76821 Mont Saint-Aignan, France. John A. Lewis, DSc, is at Silverton Lodge, Silverton Lane, Rothbury, Northumberland, NE65 7RJ, UK.

*To whom all correspondence should be addressed.