ABSTRACT
In this chapter, we discuss the utilization of adaptive randomization in phase II randomized selection trials with comparison to other designs, in the setting of multiple experimental arms. A multi-arm selection trial could involve therapies with different drugs or drug combinations, or the same drug with different doses or schedules. It often serves as a screening phase that sieves out a superior treatment regimen, which could then progress to a definitive phase III trial for comparison to a standard treatment. Running a multi-arm randomized trial is more efficient than carrying out several separate single-arm trials due to less overall resources required [16], and also avoids treatmenttrial confounding due to differences in patient selection and trial conduct [9]. In the simplest single-stage design, a selection trial allocates the same predetermined sample size to each treatment arm, and selects the arm with the highest response number [15]. As a practical modification, it is common to
also monitor the treatment arms with respect to some standard response rate. For example, Yap, Pettitt and Billingham [21] propose a screened selection design by incorporating Simon’s two-stage design [14] in each experimental arm, so that an arm will be dropped if there is no evidence that the response rate is better than the null response rate. Alternatively, Estey and Thall [9] consider Bayesian stopping rule to monitor futility at multiple interim analyses. The authors present a selection trial in patients with untreated acute myeloid leukemia (AML) and abnormal karyotypes. The trial compared four experimental arms of different drug combinations, namely LD+T, LD+A, LD+T+Thal, and LD+A+Thal, where LD stands for liposomal daunorubicin, T for topotecan, A for ara-C, and Thal for Thalidomide. The primary endpoint of interest was complete remission (CR) of disease.
