A Bayesian Framework for Event Prediction in Clinical Trials with Recurrent Event Endpoints and Terminal Events

Event-driven clinical trials are designed for the observation of a certain number of events of interest to ensure enough statistical power for achieving the primary trial objective. For ongoing trial management it is important to make prediction of when the minimally required number of events may be reached with acceptable precision. We consider the situation where the primary endpoint is a recurrent event in the presence of an associated terminal event. We employ a Bayesian framework based on a joint frailty model for prediction of the timing of observing the desired number of total events. Patient enrollment and censoring of patients due to other reasons are also modeled in the Bayesian predictive framework. The proposed approach is illustrated by a simulated case study, where predictive quantities informative for trial monitoring and interim decision making are highlighted. The operating characteristics of the proposed approach are assessed in a simulation study. The prediction is presented primarily for the case of blinded interim assessment. We also compare its performance with unblinded prediction when patient treatment information is utilized, as well as with prediction by a Bayesian latent class model where patient treatment status is implicitly estimated while making event prediction.