Online Program

Predictive Statistical Modeling of Clinical Trial Operation (302957)

 View Presentation

Keywords: patient enrollment, prediction, Poisson process, event modeling, risk-based monitoring, Bayesian technique

The increasing complexity of late-stage clinical trials requires developing innovative statistical techniques that account for stochasticity and hierarchic structure of trial operation. In this talk, the novel statistical techniques for data-driven predictive modeling of different stages of clinical trial operation are discussed. A new approach that uses evolving stochastic processes is proposed. It assumes each patient in the trial generates some follow-up evolving process (e.g., the process of future visits and associated events). The statistical technique for evaluating predictive distributions for evolving processes is developed. It allows deriving closed-form solutions for many practical scenarios, thus does not require Monte Carlo simulation. This technique is applied to predictive modeling of patients’ enrollment and follow-up processes, the number of events in event-driven trials and stopping time, associated costs, and different triggers for risk-based site monitoring: detecting unusual behavior and predicting site performance—non-enrolling sites; enrollment bounds; number of CRF, AE, etc. A few case studies are considered.