A Multistate Modeling and Prediction of Survival Distribution Using Information on Subject's Tumor Response Over Time

• Activity Number: 128
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Shengyan Hong*+ and Iiya Lipkovich and Yan Daniel Zhao
• Companies: MedImmune and Eli Lilly and Company and Eli Lilly and Company
• Keywords: Multi-state model ; Bayesian survival analysis ; tumor response ; oncology trial

Abstract

Prediction of overall survival in oncology clinical trials remains a challenging task and documented attempts to using progression-free survival as a surrogate endpoint had been unsatisfactory. We present a novel approach to predicting OS using tumor response over time via a multi-state transition model, where at any time a patient is categorized by one of four states: tumor response (tumor size decrease by certain %), tumor progression (tumor size increase by certain %), stable (neither response or progression), and death, and the transition probabilities from one of three tumor response states to death is estimated based on proportional hazards Cox model. We developed a Bayesian version of multi-state model to handle the problem of "sparse data" in state transitions. Our methods were evaluated via simulations and illustrated with a phase III clinical trial.