When randomization is not random: Allocation bias in small sample, group sequential randomized clinical trials

Area of Science:

• Clinical Trials
• Biostatistics
• Medical Research

Background:

• Randomized controlled trials (RCTs) are crucial for establishing drug efficacy, even in rare diseases.
• Bias mitigation is essential, with regulatory bodies emphasizing the impact of bias on trial outcomes.
• Limited sample sizes and reduced blinding in rare disease trials necessitate careful bias control.

Purpose of the Study:

• To quantify the impact of allocation bias on statistical decisions in small-sample, two-arm group sequential trials.
• To evaluate how different randomization strategies and group sequential trial designs influence bias-related errors.

Main Methods:

• Simulated small-sample two-arm group sequential trials using a Blackwell-Hodges convergence strategy for allocation bias.
• Assessed Type I error and power using Lan-DeMets spending functions (Pocock, O'Brien-Fleming, Wang-Tsiatis).
• Varied factors including futility rules (binding/non-binding), interim analysis timing, number of looks, and stage-wise randomization restarting.

Main Results:

• Allocation bias substantially inflated Type I error, particularly with restrictive randomization like small block permuted designs.
• Spending more alpha early in the trial reduced Type I error inflation.
• Non-binding futility rules decreased Type I error, whereas binding futility rules increased it, especially with aggressive stopping boundaries.
• Stage-wise randomization restarting offered modest bias reduction.

Conclusions:

• Group sequential trial design choices had a limited impact on mitigating bias from predictable randomization schemes.
• For open-label trials where bias is unavoidable, employing less restrictive randomization (e.g., big stick design) or large block sizes is advised.