Adding new experimental arms to randomised clinical trials: Impact on error rates

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Experimental Design

Background:

• Multi-arm platform trials efficiently test multiple experimental treatments within a single ongoing trial.
• Assessing new treatment arms in late-phase randomized controlled trials is crucial for drug development.
• Controlling the familywise type I error rate is a key consideration in platform trial design.

Purpose of the Study:

• To clarify the calculation of familywise type I error rate when new arms are added to ongoing platform trials.
• To provide methods for calculating any-pair and all-pairs power in such scenarios.
• To derive analytical formulae for correlations between test statistics in platform trials.

Main Methods:

• Extension of Dunnett's probability for multi-arm comparisons.
• Derivation of analytical formulae for correlations between test statistics.
• Verification of analytical results through simulations.

Main Results:

• Familywise type I error rate is influenced by shared control arm data and allocation ratios.
• The number and type of pairwise comparisons are primary drivers of the error rate, more so than the timing of arm addition.
• Šidák's correction is applicable for estimating the error rate when test statistic correlations are below 0.30.

Conclusions:

• The findings facilitate the design of platform trials with deferred or added arms.
• Methods enable control of pairwise or familywise type I error rates for specific comparisons within ongoing trials.
• This research supports robust statistical planning for adaptive clinical trial designs.