gms | German Medical Science

Introduction: Time-to-event endpoints are frequently applied in clinical trials measuring the patients' survival time or progression-free survival. As they are often used in the context of serious diseases, rapid evaluation of an efficient therapy has the potential of saving many patient years. Therefore, an adaptive two-stage design with the option for early stopping could be a promising way of designing such a trial.

Recently, a lot of research has been done on finding optimal adaptive designs. In the approach by Pilz et al. [1], the optimal design is determined via a numerical optimization procedure to achieve the best performance according to a chosen criterion. The optimization problem can be solved in the presence of various constraints, such as type I error rate or power. Since the performance of an optimized design cannot be improved for a given set of constraints, this approach is an attractive way of choosing design parameters.

These optimal designs are most efficient when the determined information rates are followed exactly as specified. Adhering to this principle would require constant monitoring of trial participants and performing the interim analysis right after the specified number of first-stage events is reached, which is often impractical. Operationally, it is usually much easier to plan the interim analysis for a fixed point in time. However, when initial estimates of the event rates from the planning phase turn out to be wrong, the observed information rate at the interim analysis can be markedly different from the optimal information rate of the design.

Methods: In this work, we will compare optimal adaptive designs with various other methods of designing a two-stage clinical trial for time-to-event endpoints. We will evaluate their performance in terms of type I error, power, and average sample size under a variety of different assumptions on event-rate, hazard ratio, and variations in baseline-hazard.

Results: By comparing different methods for designing a two-stage trial with time-to-event endpoints, we identify design candidates which are efficient while still being reasonably robust to misspecification of the event rate.

Discussion: If the planning assumptions on the information rate are accurate, optimal adaptive designs are an efficient and pragmatic way of designing a trial with time-to-event endpoints. However, if the assumptions turn out to be imprecise, other types of designs might be preferable.

Conclusion: The results of this simulation study helps to make an informed decision about using optimal adaptive designs in the context of survival analysis.

The authors declare that they have no competing interests.

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

This contribution has already been published: Abstract is submitted for ISCB44.