Smaller sample sizes for phase II trials based on exact tests with actual error rates by trading-off their nominal levels of significance and power

BACKGROUND: Sample sizes for single-stage phase II clinical trials in the literature are often based on exact (binomial) tests with levels of significance (alpha (alpha) <5% and power >80%). This is because there is not always a sample size where alpha and power are exactly equal to 5% and 80%, respectively. Consequently, the opportunity to trade-off small amounts of alpha and power for savings in sample sizes may be lost. METHODS: Sample-size tables are presented for single-stage phase II trials based on exact tests with actual levels of significance and power. Trade-off in small amounts of alpha and power allows the researcher to select from several possible designs with potentially smaller sample sizes compared with existing approaches. We provide SAS macro coding and an R function, which for a given treatment difference, allow researchers to examine all possible sample sizes for specified differences are provided. RESULTS: In a single-arm study with P-0 (standard treatment) 10% and P-1 (new treatment) = 20%, and specified alpha = 5% and power = 80%, the A'Hern approach yields n = 78 (exact alpha = 4.53%, power = 80.81%). However, by relaxing alpha to 5.67% and power to 77.7%, a sample size of 65 can be used (a saving of 13 patients). INTERPRETATION: The approach we describe is especially useful for trials in rare disorders, or for proof-of-concept studies, where it is important to minimise the trial duration and financial costs, particularly in single-arm cancer trials commonly associated with expensive treatment options. British Journal of Cancer (2012) 107, 1801-1809. doi:10.1038/bjc.2012.444 www.bjcancer.com (C) 2012 Cancer Research UK