Methodology of Adaptive Randomization

Abstract Simple randomization consists of allocating treatments to patients with equal probability. Attempts to improve over simple randomization stem from the desire (i) to ensure that a prespecified number of patients is enrolled in each treatment arm, (ii) to ensure balance with respect to important baseline prognostic factors across all treatment arms, or (iii) to favor (i.e., allocate with higher probability) the treatment arm that is currently faring better. These objectives can be fulfilled, respectively, by use of treatment‐adaptive randomization, covariate‐adaptive randomization, and outcome‐adaptive randomization. Treatment‐adaptive randomization can be implemented as a restricted randomization through randomly permuted blocks or as a dynamic method using a biased coin. Similarly, covariate‐adaptive randomization can use randomly permuted blocks within strata or minimization. Outcome‐adaptive randomization remains controversial because of it produces only modest gains in terms of total number of failures at the cost of increased complexity, a risk of accrual bias, and the potential for ethical concerns.