Keywords: early stopping, historical data, randomized controlled trials, design
Randomized control trial (RCT) is the gold standard of clinical studies of medical and pharmaceutical devices. The majority of research in the design of RCTs and their application has been based on the frequentist paradigm. Adaptive Bayesian clinical trials have gained increasing popularity over the years due to the significant flexibility they convey over conventional clinical trials. We present the R package bayesCT for the design and analysis of adaptive Bayesian trials for binomial, Gaussian, and time-to-event data types. Historical data offers the opportunity for dynamically decreased sample sizes, substantially reducing the cost of a trial. In our implementation, current trial data is augmented by historical data for the treatment and control groups independently, allowing immediate integration of existing, relevant data into one or more arms of a trial via the power prior or non- informative priors. The package enables early stopping for futility or success via interim analyses, allowing trials to stop or continue enrollment based on the posterior distribution of the difference between treatment and control, thus reducing patient cost and exposure to potentially inferior treatments. We use novel and efficient Monte Carlo methods for estimating Bayesian posterior probabilities, evaluation of loss to follow up, and imputation of incomplete data. Trial simulation can be carried out using parallel computing to reduce processing time. We also present a new approach to input trial parameters using the pipe operator, which makes trial design considerably more transparent than via traditional R functions. The bayesCT R package is available at https://thevaachandereng.github.io/bayesCT/ under a GPL-3 license and at CRAN.