Abstract
A recurring issue for evidence-based regulation of medicine is deciding whether to extend governmental approval from an approved use with sufficient current evidence of safety and efficacy to a novel use where such evidence is currently lacking. This “extrapolation” problem can arise in several contexts: (i) diagnosis extrapolation occurs when physicians want to use an approved drug or device to treat a new condition; (ii) patient extrapolation occurs when physicians want to use an existing drug or device to treat a new population with a given condition; (iii) dosage extrapolation occurs when physicians want to use an existing drug or device for a new duration, schedule of use, or at a new dosage; (iv) treatment extrapolation occurs when physicians want to use a new drug or device that is related to an approved counterpart. The logic of pre-approval testing, and the precautionary principal (first, do no harm), would seem to counsel prohibiting extrapolation approvals until after traditional safety and efficacy evidence exists. We reject that approach as overly conservative and instead propose a more dynamic and evolving evidence-based regime based on Bayes’ Law fundamentally, the science of learning. To apply Bayesian decision-making, one needs to (i) form a “prior” belief based on existing evidence, (ii) gather additional information, and (iii) update the prior belief. A system that allows interim periods of use can provide physicians and patients with greater treatment options while providing regulators with valuable evidence about the safety and efficacy of the proposed extrapolation. Indeed, off label drug use is legal and sometimes the medical standard of care. In contrast, a precautionary requirement conditioning all approvals on pre-existing evidence for uses that constitute just slight extrapolations along just one of these four dimensions sacrifices probable short-term health benefits at the alter of precaution. Harm is not only associated with permitting access to unsafe products but also with restricting access to beneficial products. We call for policy changes in reporting, testing, and enforcement regulations to provide a more layered and dynamic system of regulatory incentives. Our proposals are Bayesian because they force policymakers to (i) assess and acknowledge the imperfect nature of their prior beliefs regarding off-label use, (ii) gather, when cost-effective, additional information, and (iii) take action in terms of approvals, reimbursements, and enforcement based on continual updating. We aim to put Bayesianism into regulatory practice.