TO THE EDITOR: In a recent article published in Journal of Clinical Oncology, Korn and Freidlin compared outcome-adaptive randomization with designs using 1:1 and 2:1 fixed-ratio randomization (2:1 means that the probability of randomization to the experimental arm is 2/3 and to the standard arm is 1/3). The authors found no benefit in using outcome-adaptive randomization compared with fixed-ratio randomization and recommended the latter. This finding is important and interesting because clinical trial designs would be greatly simplified by adopting the straightforward fixed-ratio randomization approach rather than the logistically more involved outcome-adaptive randomization method. The results of Korn and Freidlin rely heavily on simulation studies that focus on a specific outcome-adaptive randomization method and a limited number of simulation settings. Consequently, one may question the generality of their conclusions. From a theoretic point of view, we provide additional justification and insights into comparisons between outcome-adaptive and fixed-ratio randomization. Assuming a similar setup as that used by Korn and Freidlin, we considered a comparison between a standard treatment and an experimental treatment with an immediately ascertainable binary end point. We tested the difference in the response rates between the two treatments on the basis of the one-sided normal approximation test with a type I error rate of .1 and a type II error rate of .1, and we evaluated the performance of the adaptive and fixed randomization designs on the basis of the expected number of nonresponders (denoted as nAR and nFR for adaptive and fixed randomization, respectively) and the probability that a patient would be a responder. It is known that the optimal outcome-adaptive design that minimizes the expected number of nonresponders should target the allocation ratio of pe : ps between the experimental arm and the standard arm, where pe and ps are the response rates of the experimental and standard treatments, respectively. On the basis of this result, we obtained the minimal expected number of nonresponders for outcome-adaptive randomization designs using the following equation: