Sample size recalculation in three-stage clinical trials and its evaluation

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Statistical Study Design

Background:

• Determining adequate sample size in clinical trials is complex due to uncertainties in effect size and nuisance parameters.
• Sample size recalculation, involving interim analyses to adapt future sample sizes, is a key strategy to manage this uncertainty.
• Existing research primarily focuses on sample size recalculation within two-stage trial frameworks.

Purpose of the Study:

• To extend sample size recalculation methodologies to three-stage clinical trials.
• To adapt existing two-stage recalculation rules for application in three-stage designs.
• To evaluate the performance of sample size recalculation in three-stage trials using a conditional performance score.

Main Methods:

• Modified existing sample size recalculation rules from two-stage trials for three-stage trial application.
• Applied the conditional performance score to evaluate recalculation rules in three-stage trials.
• Conducted a simulation study comparing two-stage and three-stage group sequential designs, with and without recalculation.

Main Results:

• Three-stage designs demonstrated improved statistical power and reduced expected sample size compared to two-stage designs.
• The advantages of applying sample size recalculation rules in three-stage trials were found to be contingent on the specific performance measures used.
• Observed a notable favorable effect in power and expected sample size when using three-stage designs versus two-stage designs.

Conclusions:

• Sample size recalculation is a viable and applicable method within three-stage clinical trial designs.
• The practical benefits of sample size recalculation in three-stage trials are dependent on the prioritization of specific trial characteristics by study applicants.
• The utility of recalculation hinges on whether optimizing power, sample size, or other performance metrics is the primary goal.