Sources of bias for single-arm phase II cancer clinical trials

A phase II trial is conducted to investigate if an experimental therapy is efficacious enough to proceed to a large-scale phase III trial or not. In spite of the fast progress in design and analysis methods, single-arm two-stage design is still the most popular for phase II cancer clinical trials. In this review article, we discuss two design and analysis methods that are popularly used for phase II clinical trials, but that can cause serious bias. One is about using the sample proportion as the estimator of the true response rate from single-arm two-stage trials. For a two-stage design with a futility interim test, the sample proportion is negatively biased by ignoring the two-stage design. The other is about the design and analysis of single-arm phase II trials for patient populations consisting of multiple sub-populations with different response rates. In this case, a standard design method is to project the prevalence of each subpopulation and select a standard two-stage design based on the expected response rate for the whole population. By using an unstratified statistical testing in this case, the standard analysis method can be seriously biased if the observed prevalence is very different from the projected one. In this paper, we review appropriate design and analysis methods that are proposed to avoid these sources of bias.