Adaptive survival trials.

We present a design for adaptive survival trials, where the probability of randomization to one of two treatments is skewed away from 0.5 according to the current value of the logrank statistic. A formula mapping the logrank statistic onto [0,1] is given, which is then used to bias a coin used for randomization. Simulation evidence shows that the allocation scheme works well and offers a more ethical alternative when lifetime data are available from other patients during the recruitment period. Power is not adversely affected by the resulting unequal allocation. The usual test statistic appears to be standard normal under the proposed allocation scheme.