gms | German Medical Science

Group-sequential designs are commonly used for clinical trials to allow early stopping for efficacy or futility. While the design of a single-stage randomized trial is guided by a target power for an alternative hypothesis of interest, the addition of interim analyses is driven by technical choices that are less understandable for clinicians. For example, the commonly used Lan-DeMets methodology requires specification of the timing of analyses and error spending functions. Since the rationale and effect of these technical choices is often unclear, the operating characteristics of the final design are explored under various values of the parameter of interest, and the design is then adjusted until desired properties are obtained.

In this work we develop methods for constructing designs that achieve the desired operating characteristics without the need to specify error spending functions or the timing of analyses. Specifically, we consider designing a study for the mean difference δ of a normally distributed outcome with known variance. The null hypothesis H₀: δ=δ₀ is tested versus Ha: δ=δ_A, with power π at a significance level α. The interim analyses are designed so that for a pre-specified sequence δ_Ak the study stops for efficacy at stage k with probability π if δ=δ_Ak. If stopping for futility is also considered, then the requirement to stop for futility at stage k with probability π_F if δ=δ_0k for pre-specified sequence δ_0k can also be added.

We show that under some monotonicity restrictions, such designs exist for any choice of the timing of interim analyses. Specific designs can be selected by imposing additional optimality requirements, such as minimizing the expected sample size under the target alternative δ_A, or the average sample size under a weighted selection of the alternatives.

The utility of the proposed methods is demonstrated via a simulation study and illustrated examples of prior trials. The methods are implemented in the R package gsDesignOC.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.