Exact p-Values for Simon’s Two-Stage Designs in Clinical Trials

In a one-sided hypothesis testing problem in clinical trials, the monotonic condition of a tail probability function is fundamentally important to guarantee that the actual type I and II error rates occur at the boundary of their associated parameter spaces. Otherwise, one has to search for the actual rates over the complete parameter space, which could be very computationally intensive. This important property has been extensively studied in traditional one-stage study settings (e.g., non-inferiority or superiority between two binomial proportions), but there is very limited research for this property in a two-stage design setting, e.g., Simon’s two-stage design. In this note, we theoretically prove that the tail probability is an increasing function of the parameter in Simon’s two-stage design. This proof not only provides theoretical justification that p-value occurs at the boundary of the parameter space, but also helps to reduce the computational intensity for study design search.