The Zelen design may be the best choice for a heroin-provision experiment.

Recently, the Dutch Parliament agreed upon the conduct of a randomized clinical trial on the effects on heroin provision on general health and psychosocial and criminal behavior in long-term addicts. Previous studies failed to establish the effects beyond reasonable doubt. The main reasons why previous trials failed are massive dropout or noncompliance in the control group. Designing a new heroin-provision trial, we concluded that the Zelen design provides the best guarantee for obtaining valid study results. Compared with the traditional design, the Zelen design probably reduces noncompliance and dropout considerably, thus increasing validity. Depending on the study population, the Zelen design may reduce study precision. However, in a trial aimed at badly integrated addicts, the Zelen design can be conducted without loss of precision because baseline measurements will only weakly correlate with effect measurements. The arguments favoring the Zelen design may be generalized to trials in which the experimental treatment is highly attractive to the study participants. However, the use of the Zelen design precludes blinding of participants who receive the experimental treatment. We argue that the conduct of studies that predictably tend to produce invalid results is ethically dubious. The ethical problem of studying participants without their consent can be solved by a slight modification of the Zelen design in which the sampling of a control group is postponed. Both the traditional and the Zelen design can imply ethical problems. Both designs can be ethically justifiable and should not be rejected on a priori grounds.