All Times EDT
Keywords: Bayesian power, assurance, clinical trial design, treatment crossover
Randomized clinical trials sometimes include an option to cross over from control to active treatment to allow all subjects the possibility to experience the novel therapy under investigation. We describe a scenario where treatment crossover is allowed after the primary endpoint of progression-free survival (PFS) is met, but before a key secondary endpoint of overall survival (OS). This design allows valid evaluation of PFS; however, evaluation of OS could be impacted, especially since we are interested in estimating the treatment effect on OS in the absence of treatment crossover. We review several methods for accounting for treatment crossover to estimate the treatment effect on OS. Each of these methods has assumptions and drawbacks which are difficult to overcome, and so intention-to-treat (ITT) analysis is often considered the standard. ITT analysis simply ignores treatment crossover and therefore can underestimate the treatment effect of interest. In order to estimate this impact, we have developed a simulation program to calculate power and probability of study success incorporating treatment crossover, enrollment, dropout and interim analysis. Simulation results show that high likelihood of treatment crossover results in a large reduction in study power when conducting ITT analysis. Care needs to be taken when designing clinical trials with treatment crossover to ensure high probability of success for key secondary endpoints.