Likelihood Based Study Designs for Time-to-Event Endpoints

TL;DR

This paper develops likelihood-based sequential study designs for time-to-event data, enabling controlled sample size projections and evidence monitoring without adjustments for multiple looks, with applications in clinical trials.

Contribution

It introduces a method for sample size projection and evidence monitoring in sequential designs for time-to-event outcomes using likelihood principles, including handling of efficacy and futility.

Findings

Sample size projections control misleading evidence probability.

Likelihood methods facilitate sequential monitoring without adjustments.

Application demonstrated in a phase II cancer trial.

Abstract

Likelihood methods for measuring statistical evidence obey the likelihood principle while maintaining bounded and well-controlled frequency properties. These methods lend themselves to sequential study designs because they measure the strength of statistical evidence in accumulating data without needing adjustments for the number of planned or unplanned examinations of data. However, sample size projections have, to date, only been developed for fixed sample size designs. In this paper, we consider sequential study designs for time-to-event outcomes assuming likelihood methods will be used to monitor the strength of statistical evidence for efficacy and futility. We develop sample size projections with the aim of controlling the probability of observing misleading evidence under the null and alternative hypotheses, and we show how efficacy and futility considerations are managed in this context. We also consider relaxing the requirement of specifying the simple alternative hypothesis in advance of the study. Finally, we end with a comparative illustration of these methods in a phase II cancer clinical trial that previously was designed within a Bayesian framework.