Investigating maximum power losses in survival studies with nonstratified randomization.

In studies with nonstratified randomization, imbalances may occur which lead to power loss. We investigate how much power may be lost at worst, that is, with small but nonnegligible probability. The situation considered is a two-treatment trial with censored survival data, and three to five strata. We also derive asymptotic results, as the sample size and/or number of strata increase. In addition, we consider how much a projected sample size needs to be increased to be confident (at a certain probability level) that the intended power will be retained, conditionally on the unbalanced configuration.