Bayesian methods are becoming increasingly popular in the statistical design and analysis related to drug and medical device development. In early phases, they permit borrowing from auxiliary data and expert opinion in order to reduce trial sample sizes and save valuable development time. Their ability to deliver exact probability statements about the likelihood of safety, efficacy, and other key quantities makes them especially well-suited to predictions of late phase success, facilitating Go/No-Go and portfolio management decisions. However, Bayesian methods do require the user to specify a prior distribution, as a starting point for the analysis. In this webinar, we review the role and meaning of prior distributions, and how they can be elicited from past data and expert opinion. We then show this information can then be utilized to power a Bayesian clinical trial using a new Shiny app written in the popular R language. The app computes the design’s Type I error and power under both the elicitee’s informative prior, and a noninformative, reference prior. Other applications and exemplification will also be included.
Key learnings:
- Appreciation of the role of the prior distribution in Bayesian analysis, how it can affect the results, and how it can be readily derived
- Understanding of how informative prior distributions drive the Type I error and power of Bayesian adaptive clinical trial designs, and how this process is viewed by regulators
- Illustration of a new R Shiny app to carry out these two functions, and how it informs the proper sample sizes to use in Phase I, II, and III studies
