Relative efficiencies of alternative designs for randomized trials

Abstract

In conventional randomized controlled trials (RCT), one cannot estimate the influences of patient preferences on the treatment outcomes since all patients are randomized to a treatment without identifying their preferences. The two-stage design allows patients in the “choice group” to choose their desired treatment while those in the “random group” are randomized in the same way as in a RCT. The partially randomized design allows all patients to receive their preferred treatment. In the fully randomized design, although the patient preferences are identified at the first stage, all the patients are then randomized to each treatment. In this thesis, we discuss these four designs in detail with respect to their estimable effects, variances of the estimable effects, and the relative efficiency of each effect in different designs. Participants who are indifferent to the treatments (undecided participants) are included in the designs when evaluating the various effects. This thesis also shows the relationships of relative efficiency to other factors, such as the proportion of undecided participants, the relative preference rate between the two treatments, and with unequal numbers of participants being randomized to each treatment. We discuss the advantages and disadvantages of the designs under different scenarios and whether unequal randomization could improve efficiency.