Sample Size Calculation in Dose Optimization Trials Using the Margin of Practical Non-Inferiority

A dose optimization trial in oncology may be performed to compare an approved dose level of a given drug with a reduced dose level, testing the hypothesis that efficacy is maintained whilst reducing side effects and consequently improving adherence and quality-of-life. This is particularly relevant with modern therapeutic agents whose mechanisms of action imply that efficacy may not necessarily be linearly related to the dose. Using a conventional non-inferiority framework leads to large sample sizes that are often unfeasible in the phase IV setting. An alternative is to use a margin of practical non-inferiority, which we define in this paper and show how it can be exploited to justify a sample size. Whilst defining the extent of the margin, researchers also pre-specify the other dimensions of interest, such as receptor occupancy and/or side effects and quality-of-life, that will be used to establish practical non-inferiority if the observed efficacy of the reduced dose level lies within the margin. The comparison of efficacy is based on the observed difference between the reduced and the approved levels, instead of the confidence interval of this difference, leading to a reduction in sample size. The reduction in precision due to the smaller sample size is compensated by formally pre-specifying the additional dimensions to the decision process, allowing a more thorough assessment of the opportunity to reduce a dose in practice, with the many advantages that this may involve.