ABSTRACT
Randomization provides the basis for determining causality in scientific experiments. While there are settings in medicine and public health where convincing evidence may be obtained from nonrandomized studies, there is a clear shift toward requiring the higher quality evidence obtained from well-designed randomized trials. Rapidly changing diagnostic and disease assessment tools, the use of more subjective endpoints, requirements for diverse patient populations, and modest effect sizes all tend to weaken the inference from uncontrolled experiments. In oncology, where historically a single-arm phase II trial has often been adequately informative, randomized phase II trials are increasingly common.
