An important ethical issue that may arise when conducting conventional RCTs of drugs used off-label is how to minimize the risk involved with exposure to possibly inferior treatment. One of the major challenges involved in the implementation of PCTs of drugs used off-label is acceptance by neonatologists because of a “possible exposure to inferior treatment (placebo)” when the study drug is widely believed to be beneficial.(Roth, Adatia et al. 2006
; Short, Van Meurs et al. 2006
) One approach to minimize the risk of exposure to inferior treatment is a fixed randomization scheme that has unequal allocation.(Palmer and Rosenberger 1999
) This is similar to conventional randomization, except that instead of a 50:50 allocation, twice (or thrice (2:1 or 3:1)) as many participants are randomized to the drug arm of the study than to the placebo arm. This strategy tends to be used more for recruiting purposes, however, than for addressing the ethical issue at hand, mainly because the (fixed) allocation does not take into account evidence that is accumulating throughout the trial.
Another approach is to use a fully sequential design with equal group allocation.(Whitehead 1997
; Palmer and Rosenberger 1999
) In the commonly used group sequential design, the sample size is fixed in advance and data are monitored by a Data and Safety Monitoring Board after groups of subjects have completed the study to determine whether or not the trial should be modified (or halted) on the basis of safety or efficacy. However, with a fully sequential design, the sample size is not known in advance and data are monitored continuously. The accumulating results are evaluated to determine whether or not there is a significant benefit of treatment according to a pre-determined stopping boundary. The trial is planned by specifying a significance level, power, and type of stopping boundary (e.g., triangular boundary). The expected sample size can be calculated under various assumptions concerning the treatment effect. The expected sample size will be larger than that for a conventional fixed sample size trial under the assumption of no treatment effect, but it can be considerably smaller than that required for a conventional trial or one that uses group sequential monitoring if the treatment effect is large. In this case, the ethical issue is addressed by halting the trial as soon as a significant benefit of treatment is detected, leading to the exposure of fewer subjects to inferior treatment.
An alternative approach is to use a response adaptive design. There are two well-studied response-adaptive designs that are based on an urn model and are applicable to clinical trials that have two treatment arms (A and B) and binary outcomes (success or failure). A design that is well studied in neonates is the randomized “play-the-winner” rule. (Rosenberger 1999
) At the beginning of the study, an urn contains X
“A” balls and Y
“B” balls (typically X
). On subject recruitment, a ball is drawn and replaced. If it is an “A” ball, the subject receives treatment “A”; if it is a “B” ball, the subject receives treatment “B”. A success with treatment “A” or a failure with treatment “B” will result in adding a certain number of “A” balls to the urn, while a failure with treatment “A” or a success with treatment “B” will result in adding a certain number of “B” balls to the urn. Thus, the probability of being assigned to the (currently) superior treatment is greater than 50%.
A similar adaptive design is the randomized “drop-the-loser” rule.(Coad and Ivanova 2005
) At the start of the study, there are X
balls for both treatments “A” and “B” in the urn, and there are Z
so-called “immigration (I)” balls. The immigration balls ensure that the urn does not get emptied. When a neonate is enrolled, a ball is drawn and the subject is assigned the treatment according to the ball drawn. If the “I” ball is drawn, the ball is replaced and one additional ball of both “A” and “B” are added to the urn. If a success is observed with the “A” treatment, the “A” ball is returned to the urn; if a failure is observed with the “A” treatment, the ball is not returned to the urn. The drop-the-loser rule appears to be more promising in that there is less variable allocation and a smaller number of neonates who have treatment failures compared with the play-the-winner rule. These adaptive designs can be combined with a sequential design to further reduce the exposure to inferior treatment.(Coad and Ivanova 2005
The adaptive designs outlined here have limitations (Palmer and Rosenberger 1999
; Rosenberger 1999
). They are more applicable with treatments that are not thought to be associated with significant toxicity. Adaptive designs are only suitable when the outcome is binary and there is little to moderate delay in its ascertainment. They have the potential of suffering from accrual bias whereby subjects wish to be recruited later in the trial in order to maximize their chances of receiving the superior treatment. This is only a concern in situations where subjects have the ability to decide when to enroll and when it is impossible to mask the number of subjects that have been previously entered into the trial. Time trends in subject characteristics can also be a problem in these designs because the theory that underlies the analysis of data from these designs assumes that the response probability is constant over time. Short recruitment periods are advocated to help avoid this problem. Logistical aspects such as the randomization process are challenging but are generally not too difficult to overcome. Finally, the statistical issues involved in the design, evaluation of the design’s operating characteristics, and data analysis are relatively complex and require careful attention.
If an effective standard therapy exists and there is a need to conduct a PCT for a study drug, an alternative approach is to randomize subjects to the standard therapy plus placebo or standard therapy plus the study drug. In these “add on trials” which involves the use of standard therapy in both arms of the study, there are no ethical issues pertaining to the use of placebo because placebo is used in addition to, but not in place of, standard treatment. This, of course, will be applicable only when the treatments have different mechanisms of action, and such trials do not permit conclusions concerning monotherapy.
In a situation where there is some, but inconclusive, evidence for efficacy of an investigational drug, the use of a cross-over design, in which all participants eventually receive the investigational treatment and, therefore, its possible benefit, may be ethical.(Hyman and Shore 2000
) However, cross-over studies are applicable mainly for chronic, stable conditions and are best suited for short-term therapy.