Sequentially adjusted randomization to force balance in controlled trials with unknown prevalence of covariates:...

In treatment outcome studies with small to medium sample sizes (n < 200), the balance of groups with regard to important factors, which sometimes occur at low prevalence, is indispensable for adequate interpretation. This study tested a method for use in clinical alcoholism research, an uncomplicated procedure for satisfactory randomization of patients to different treatments, taking into account relevant background variables.

An easily applicable modification of Efron's biased coin method for the randomization of treatments within strata of unknown but low prevalence was compared with the original approach and alternative methods by computer simulation (10,000 runs). An application example for a clinical trial in alcoholism research is given.

The sequentially adjusted randomization procedure revealed results similar to Efron's approach without the necessity for monitoring the assignment history throughout the trial. The new method was slightly superior to Efron's approach in randomizing subjects in strata with n < or = 20, whereas strata with n > 20 favoured randomization with Efron's approach. Taking into account all results from simulation, the new approach reached a proportion of acceptable balanced randomization > 95% in all stratum sizes.

The approach, a special case of the standard urn design, provides three major advantages in clinical trials: (i) it can be easily implemented in any trial without technical equipment; (ii) it works with high accuracy in trials with a priori unknown but low numbers of subjects (4 < or = n < or = 20) in prognostic relevant strata; and (iii) a deterministic assignment tendency is completely avoided, as a random process takes place throughout the assignment procedure. The modified biased coin method can be recommended as one possible strategy for special purposes, particularly in alcoholism research.