The United States Food and Drug Administration has been considering whether to allow the use of Bayesian statistic in the design and analysis of clinical trials for several years. Happily, in my opinion, the FDA recently finalized its decision to allow, and even encourage, use of Bayesian statistics. I believe this is a good thing both because of the stated reason that Bayesian methods can enable faster and more efficient clinical trials, and also because the Bayesian point of view is, to me, a more realistic and comprehensible approach to decision science in the clinical research context. Although most of us are familiar with the traditional, frequentist approach to clinical trial data analysis, the subtleties of what the frequentist methods really tell us are sometimes lost or misinterpreted, resulting in misleading or frankly incorrect conclusions about the results of clinical trials. I believe that adoption of Bayesian methods by the clinical research field will ultimately yield the view that these methods are more closely aligned with how we really think about what we want to accomplish and will better support the types of decisions we really want to make using clinical experiments. Ultimately this may lead to better communication about trial results.
The FDA guidance document itself is fairly readable at 50 pages. In addition to the official, regulatory reasoning behind its creation, the guidance document provides a nice background on Bayesian analysis from the clinical research perspective. The last part of the document provides good detail in straight forward language regarding content that the FDA would like to see in submissions and conversations/communications that the Agency would like to have with sponsors before, during, and after trials with respect to trial design and analysis using Bayesian methods.
The document can be found on the FDA website: http://www.fda.gov/MedicalDevices/DeviceRegulationandGuidance/GuidanceDocuments/ucm071072.htm
