Statistical inference and the design of clinical trials

According to the likelihood principle of statistics, a decision to stop or otherwise alter a clinical trial can be made on the basis of accumulating information without losing the ability to draw inferences from the results of the trial. In particular, balanced, randomized designs are not necessary. The probability that a particular treatment is the best among those in the trial can be calculated after each patient response, and may suggest that the treatment should be used predominantly in the next stage of the trial. Doing so results in more effective treatment of the patients in the trial while sacrifacing some of the information on the other treatments. Not doing so results in equal information on the treatments but sacrifices effective treatment. There are many trials in which compromise is possible between these antagonistic consequences.