New Zealand Statistical Association 2024 Conference

Thomas Lumley

University of Auckland

Speeding up design in Bayesian platform trials

Bayesian platform trials are now becoming popular. Although these trials have results and stopping guidelines specified in Bayesian terms, sponsors (in industry or the public sector) often want to know operating characteristics such as probability of finding an effective treatment if there is one, probability of falsely declaring a treatment effective, and probability of continuing for longer than a specified time period with no results. Estimating these characteristics requires simulation. I will present ways to speed up this simulation in R and Stan using approximations to the posterior and using ratio estimation.