Duffy-Santner confidence intervals for the two-stage three-outcome design

Although the three-outcome phase II clinical trial design (Sargent et al., 2001) has seen increasing use, confidence interval methodology for the binary endpoint had not been explicitly defined. Typical phase II clinical trial designs with binomial endpoints use a sequential binomial confidence interval (SBCI) algorithm (Duffy and Santner, 1987). We reviewed the SBCI literature and software in order to describe the construction of confidence intervals for the three-outcome setting. Simulations confirmed that proper application of the SBCI algorithm provides appropriate coverage for the three-outcome design.