Perhaps nowhere is the need for agreement on a routine and robust method of evaluating statistical significance greater than in regulatory science. Statististicians in charge of drug and device approvals require methods that are straightforward to explain and implement, so that applicants feel they are being treated equitably. While a large number of guidance documents on the use of Bayesian methods in regulatory science have been produced by FDA and similar groups in Europe, most of this advice to date has been in the domain of medical devices, where historical data is often available for the construction of reliable prior distributions. Drug and biologics regulators are increasingly willing to use Bayesian methods, but agreement on an approach for their routine implementation remains elusive. Challenging issues in multiplicity also arise (as when a manufacturer seeks approval only for a particular subgroup of patients, or to extend a drug or device's label to pediatric use), and merely switching one's outlook from frequentist to Bayesian does not eliminate the problem. In this session, we will hear from 3 leading researchers in this field, all of whom understand the points of view of both the regulator and the regulated, and who will speculate on the future of Bayesian methods in this important area where P-values currently dominate.
