Sample size re-estimation for pivotal clinical trials.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Statistical Methodology

Background:

• Sample size re-estimation (SSR) designs are susceptible to inflated Type I error rates if hypothesis tests remain unchanged.
• Existing methods to control Type I error include combination tests, conditional error approaches, and conventional tests restricted to an allowable region (AR).

Purpose of the Study:

• To explore the connections between different sample size re-estimation (SSR) design approaches.
• To clarify the trade-offs in statistical power and Type I error control among these methods.
• To advocate for improved performance metrics for evaluating SSR designs.

Main Methods:

• The study analyzes the relationship between combination tests, conditional error functions, and allowable regions (AR) in SSR designs.
• It compares the statistical power of conventional tests adapted within the AR versus combination or conditional error approaches.
• Alternative step-wise adaptation rules outside the AR are considered for conventional tests.

Main Results:

• Combination tests, conditional error functions, and ARs are interconnected components of SSR designs.
• Adapting conventional tests within the AR controls Type I error but yields lower power than combination or conditional error methods.
• Step-wise adaptation rules offer an alternative when conventional tests are preferred but do not fully utilize the AR.

Conclusions:

• Controversies in comparing group sequential (GS) and SSR designs arise from misaligned performance metrics and evaluation perspectives.
• Summary metrics (median, variance, tail probabilities of sample size) alongside expectation are recommended for personalized efficiency definitions.
• Conditional metrics evaluating favorable, promising, and unfavorable interim result zones offer crucial insights for decision-making in SSR designs.