Balanced two-stage designs for phase II clinical trials

Abstract

Oncology phase II clinical trials are often designed using Simon's two-stage designs (optimal and 'minimax') for independent observations. Simon's designs do not include the use of correlated observations, and do not balance the sample sizes of the two stages. In these designs, the sample sizes of the two stages can be highly unequal. In certain circumstances, an alternative design option that balances the sample sizes is desirable. To develop a two-stage phase II design that balances the sample sizes of the first and the second stages, while controlling for type I and type II error rates. We simulated designs based on response rates under various null and alternative hypotheses, type I and type II error constraints, and the degree of correlation in the case of correlated data. For correlated data, Sargent's method is adopted to account for the loss of information due to intra-person correlation. Design characteristics for different parameter settings were generated using balanced design method, separately for independent and correlated data. Results were evaluated and compared to the optimal and minimax design. For correlated data, designs were produced only for trials with half of the participants having one observation and the other half having two. Also, the degree of intra-person correlation was fixed at three levels. The balanced design provides an additional choice for two-stage phase II trials when the investigators would like to monitor the trial near a study's halfway point. Meanwhile, its total sample sizes are comparable with Simon's designs.