Reducing selection bias risk to enhance RCT validity: sandwich mixed randomization outperforms permuted block design

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Epidemiology

Background:

• Random allocation is crucial for unbiased treatment assignment in randomized controlled trials (RCTs).
• Permuted block design (PBD), despite its widespread use (~75% of RCTs), suffers from predictable allocation sequences, compromising randomization integrity.
• Existing alternatives to PBD have not gained broad acceptance, highlighting the need for a simple, universally applicable method that balances treatment groups and improves unpredictability.

Purpose of the Study:

• To introduce and evaluate Sandwich Mixed Randomization (SMR), a novel method designed to enhance allocation unpredictability while maintaining treatment group balance.
• To compare the performance of SMR against fixed- and variable-sized PBD using key metrics such as treatment imbalance and allocation predictability.
• To quantify the risk of selection bias associated with PBD and SMR.

Main Methods:

• SMR integrates complete randomization with PBD within a "sandwich" framework for improved unpredictability and balance.
• Monte Carlo simulations were used to evaluate SMR in 1:1 two-arm open-label RCTs of varying sizes (n=48, 240, 1,200).
• Performance metrics included absolute group size difference, proportion of correct allocation guesses, and relative excess risk of selection bias compared to complete randomization.

Main Results:

• PBD ensures perfect balance but shows high allocation predictability (correct guesses >68%), significantly exceeding the 50% benchmark of complete randomization.
• SMR maintains balanced group sizes while reducing correct guess proportions to ~56% (small trials) and ~53% (larger trials).
• SMR reduced the risk of selection bias by >66% (small trials) and >81% (larger trials) compared to variable-sized PBD, with a consistent PBD-to-SMR risk ratio for bias exceeding 3 across all sample sizes.

Conclusions:

• The predictability of PBD, irrespective of block size, substantially elevates the risk of selection bias in RCTs.
• SMR presents a universally applicable, low-effort alternative to PBD, significantly mitigating selection bias risk across various RCT designs.
• Adopting SMR enhances randomization integrity and internal validity, reinforcing the RCT's role in generating reliable clinical evidence.