Minimal Sufficient Balance Randomization and Site-Specific Covariate and Group Size Imbalance in Multicenter Acute Stroke Trials

BACKGROUND: Preservation of treatment allocation randomness, achievement of treatment group size balance, and balance on prognostic baseline covariates are desirable properties of optimal randomization schemes. Previous studies have demonstrated the accuracy of covariate-adaptive randomizations, such as minimal sufficient balance (MSB) randomization, for achieving covariate balance in acute stroke trials at the end of the trial. This study evaluates the performance of covariate-adaptive randomization techniques against simple and block randomization in minimizing site-specific treatment group imbalance in multicenter acute stroke trials. METHODS: Monte Carlo simulations were used to evaluate the performance of stratified and unstratified versions of MSB, common scale MSB, and common scale group-size MSB (CSSize-MSB), against permuted block and simple randomization designs, for achieving balance across baseline covariates and sites. Simulation conditions investigated include the number of sites (3, 6, or 20 sites), enrollment per site (equal or unequal enrollment across sites), number and distribution of baseline covariates (sex, age, National Institutes of Health Stroke Scale score, large vessel occlusion status), and sample size (n=250, 600, 1000, 3000). The probability of observing statistically significant imbalance on any baseline covariate, proportion of biased allocations, and overall and site-specific group allocation ratio at interims and end of enrollment were used to evaluate the performance of the randomization schemes. RESULTS: The average probability of observing imbalance on any of the baseline covariates for the simple randomization, permuted block, common scale MSB, common scale group-size MSB, and MSB were 21%, 21%, 0%, 2%, and 2%, respectively, at n=600 with 20 study sites. Although site-specific treatment allocation imbalance was improved under MSB algorithms, imbalance at low-enrolling sites persisted, regardless of the randomization scheme. Treatment allocation randomness and treatment-control group balance were preserved in high-volume sites under MSB. CONCLUSIONS: Although site-specific treatment group imbalance persisted in low-enrolling sites, regardless of the randomization technique adopted, the overall randomness of treatment allocation and balance of covariates were preserved with MSB algorithms. Logistical considerations and oversight to minimize low enrollment across sites are recommended before onboarding sites in multicenter acute stroke trials.