Online Program

Enrollment, Events Prediction, and Statistical Power Prediction for Event-Driven Trials (302942)

Keywords: Time-to-event, Poisson-Gamma, Poisson process, probability of success

For event-driven trials—for instance oncology and cardiovascular trials—it is crucial to predict highly correlated patient enrollment, events accrual (especially if it is the primary endpoint) so the trial can be planned and powered properly (i.e., to secure with a given probability the targeted number of events/subjects), study duration, major milestones, and statistical metrics. In addition, it is critical to know the probability of success of the clinical study during the operation process. The Poisson-Gamma model has been of increasing interest, and has been widely accepted in patient enrollment modeling recently. This presentation describes the event prediction based on the exponential, or Weibull, distribution of time-to-events (with potential type 2 censoring) that follows the Poisson-Gamma patient arrival and staggered site initiation. The study power for the log rank test is predicted based on the potential number of endpoints accrued at any interim stage. The probability of success of the clinical study is calculated as the probability of the study power above a pre-defined threshold at any interim look.