ABSTRACT
For early-phase cancer trials for dose Œnding, many useful designs including Bayesian dose-Œnding designs have been proposed in the literature (see, e.g., Storer, 1989, 1993, 2001; Piantadosi and Liu, 1996; Thall and Russel, 1998; Whitehead and Williamson, 1998; O’Quigley et al., 2001; Chang and Chow, 2005; Loke et al., 2006; Zhou et al., 2006). In practice, however, only two types are commonly used (Dent and Eisenhauer, 1996; Eisenhauer et al., 2000; Le Tourneau et al., 2009). These are the algorithm-based designs which follow a traditional escalation rule (TER), e.g., the “3 + 3” design as well as the model-based designs using the continual reassessment method (CRM) (see, e.g., O’Quigley et al., 1990; O’Quigley and Shen, 1996; Heyd and Carlin, 1999; O’Quigley, 2001; Babb and Rogatko, 2004; Kamp et al., 2007; Paoletti and Kramer, 2009).
