Statistical innovations in clinical trial design with a focus on drug combinations, factorials, and other multiple therapy issues

Statistical methods in clinical research tend to become entrenched. Innovations threaten the status quo. The "right way" becomes frozen in lore. This is so even when the "right way" is not best. "Statistical significance" and the associated requirement of "high power" is an example. This attitude is an impediment to efficient design. Willingness to address some design issues with moderate power enables building highly informative and highly efficient clinical trials. This article considers several types of clinical trials, including dose-finding, combinations, and factorial designs. Bayesian adaptive methods are used to show that trials can be made more efficient and more informative. Surprisingly, the approach is consistent with many attitudes of the widely regarded "Father of Modern Statistics," R.A. Fisher. Fisher was anti-Bayesian in rejecting its subjective interpretations. But Fisher and Bayes come to the same conclusion in many applied matters. Fisher invented factorial design. Its principal attraction for him was enabling addressing two or more questions with a single experiment. He complained about attitudes that hindered progress: "No aphorism is more frequently repeated in connection with field trials [and clinical trials], than that we must ask Nature few questions, or, ideally, one question at a time… this view is wholly mistaken." Fisher's primary analysis required modeling and making assumptions. For example, his first analysis in a factorial setting assumed no interactions among the factors. He investigated possibilities of interactions but he did not see the need for doing so with high power.