What is Zelen's design?BMJ 1998; 316 doi: https://doi.org/10.1136/bmj.316.7131.606 (Published 14 February 1998) Cite this as: BMJ 1998;316:606
- a National Primary Care Research and Development Centre, Centre for Health Economics, University of York, York YO1 5DD
- b National Primary Care Research and Development Centre, University of Manchester, Manchester M13 9PL
- Correspondence to: Dr Torgerson.
This is the fourth of an occasional series on the methods of randomised controlled trials
When patients do not receive their preferred treatment in randomised trials there may be difficulties with patient recruitment and scientific problems with bias.1 For example, bias may occur when patients are aware of a new treatment not available to them and comply poorly with the standard treatment.
Zelen's design can address these difficulties 2 3 by randomising patients before consent to participate has been sought. Two types of the design exist: double and single consent. In the double consent version patients are initially offered the treatment to which they were randomised; however, if they decline the randomised treatment, they can then be offered alternative therapies—including the experimental treatment. In the single consent version only patients offered the experimental treatment are told there is an alternative treatment (the control) available. Patients randomised to the control treatment are not allowed the experimental treatment (although they are given unhindered access to any usual treatment facilities). Analysis is undertaken with patients retaining their original assignment.
Zelen's design has been much discussed and for most therapeutic trials is probably unethical. Occasionally, however, it has been chosen on ethical grounds. For example, in a trial of extracorporeal membrane oxygenation for infants with pulmonary hypertension Zelen's design was used as it was considered preferable not to raise false hopes among half the parents that there was a novel treatment available for their child only to have it denied them through the randomisation.4
Zelen's design may be particularly useful for evaluating population based interventions such as screening, where it is important to estimate the effects on a whole population. However, if the presence of the trial is known to the non-screened group this may artificially induce changes in that group which may influence the results (a Hawthorne effect). For example, in a randomised trial of bone density screening5 the non-screened group were not contacted at baseline as this might have artificially increased their use of hormone replacement therapy. Had the trialists not used Zelen's design the investigators could not have been sure of the full unbiased impact of screening on uptake of hormone replacement therapy.
If bias due to patients knowing they are in the “usual care” group is to be avoided patients usually need to be followed up for key events at a distance so as not to alert them to the study. For example, in a randomised trial of colorectal cancer screening cancer events for both groups of patients were ascertained through medical records and a cancer registry.6 By using Zelen's design in screening trials it is possible to achieve more accurate estimates of population outcomes such as cancer reduction 6 7 compared with the conventional trial designs.
There are obvious ethical problems in using Zelen's design to randomise patients without their consent8 (though treatment consent is always sought). For some interventions, however, such as screening, this may be the only practical design. For example, if all patients in the colorectal cancer screening trials had been screened but only a random half had been offered intervention, there would have been an ethical dilemma of not offering further investigation and treatment to control patients who appeared to be at high risk.
Zelen's design can have other disadvantages. If the trial requires intrusive data collection or monitoring then Zelen's design as control patients will be aware of the study. Given that intrusive data collection is not feasible, it may not be possible to use restrictive inclusive or exclusive patient recruitment criteria. Furthermore, if many patients refuse their original treatment, this will lead to a reduction in study power. Both these factors will lead to the need for a large sample size.9