Bayesian predictive probability design: theory and practical application in a prospective study

Abstract

In an experiment-based prospective study aiming to determine the efficiency of a treatment, the time by which it becomes clear whether a therapy is effective or not is critical. This applies specifically to clinical trials and refers to the same extent to both successful and futile therapies. This study seeks to answer the question when there is enough evidence allowing the trial to be finalised. The key is to find enough statistical signals, working on the smallest possible sample, to make a judgment whether to extend, continue or terminate the study. The Bayesian predictive design allows drawing conclusions about the prognosis of a study considering the actual results. The article provides a theoretical background and presents a practical perspective, addressing the statistical properties and technical aspects of conducting a trial based on a predictive design. Additionally, the sensitivity of the design to the choice of prior distribution is discussed.