Selecting a randomization method for a multi-center clinical trial with stochastic recruitment considerations.

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Health Services Research

Background:

• Multi-center randomized controlled trials (RCTs) require careful design considerations, including sample size, center selection, and participant recruitment strategies.
• Sequential randomization methods are crucial for multi-center RCTs, with or without stratification factors.
• This study focuses on assessing randomization methods under a competitive patient recruitment policy.

Purpose of the Study:

• To systematically evaluate various randomization methods for multi-center 1:1 RCTs.
• To assess these methods within the context of a competitive patient recruitment process.
• To compare the performance of different stratification and balancing techniques.

Main Methods:

• A Poisson-gamma model was used to simulate the patient recruitment process.
• Sixteen distinct randomization methods were investigated, including unstratified, region-stratified, center-stratified, and dynamic balancing randomization (DBR).
• Monte Carlo simulations assessed statistical properties like recruitment time, treatment imbalance, and allocation efficiency.

Main Results:

• Maximum tolerated imbalance (MTI) methods (e.g., big stick, Ehrenfest urn) outperform conventional permuted block designs (PBD) in balance-randomness tradeoff.
• DBR effectively controls imbalance across trial, region, and center levels while preserving randomization.
• Increasing study centers accelerates recruitment but can increase center-level imbalances; larger block sizes or MTI thresholds improve the randomness-balance tradeoff.

Conclusions:

• The selection of an appropriate randomization method is critical for multi-center RCT design.
• Dynamic balancing randomization (DBR) emerges as a highly effective strategy for trials with competitive patient recruitment.
• Optimizing MTI thresholds within DBR is recommended for robust trial design.