Keywords: medical devices, Bayesian models, historical controls, least burdensome
Two-arm randomized controlled trials may not be necessary or appropriate for all evaluations of device safety and efficacy. For example, surgically implanted heart valves have a long history of approval via one-arm trials in which device performance is compared to an objective performance criterion (OPC). The OPC is a set of adverse event rates to which a new device’s adverse event rates are compared. The methodological challenges of setting and using OPCs include incorporating diverse evidence sources, reflecting statistical uncertainty, adjusting for patient characteristics, and extending to new devices and patient populations. In this talk, I outline features of an ideal OPC and propose hierarchical Bayesian models for setting and updating OPCs. I demonstrate these ideas in the example of transcatheter aortic valve replacement (TAVR), a relatively new alternative to surgical aortic valve replacement.