Content area
Abstract
Since the "double-consent randomized design" is the prerandomization mechanism that is currently of major interest, this paper will focus on problems of analysis in studies using this design; some scientific, ethical, and practical concerns with its use; and the experience with the design to date. Analysis of Prerandomized Studies The analysis of studies in which patients are prerandomized presents some special problems. Because patients are randomly assigned to treatment before giving consent, there will be some proportion of randomized patients on each arm who refuse the assigned treatment. If, instead of excluding the patients who refused, we include them with those who were assigned to treatment B, the observed-response rate for group 15 will then be 34 per cent (0.54 X0.20 + 0.46 x 0.50). Since we can never be sure that no such selection bias is operating, the only appropriate analysis is the comparison of patients randomized to one treatment with those randomized to the other. the dilution of the treatment effect can be compensated for by inflating the sample size. [...]it may be more correct to view this inflation factor as an upper boundary in assessing the potential inefficiency of prerandomized studies.) It may seem anomalous to attempt to increase accrual by introducing a study design that requires more patients than the conventional design.