Application of Disease Progression and Time-to-Event Models to Heart Failure and Large-Tumor Oncology Data

• Activity Number: 573
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Shu Yang*+ and David James and Ramesh Sarangapani
• Companies: Novartis and Novartis and Novartis
• Keywords: time-to-event ; joint-models ; extended Cox proportional hazards

Abstract

Unlike baseline-only risk models that ignore changes in patient status throughout the follow up period, we incorporate disease progression in modeling the risk of morbidity and mortality outcomes in a non-confirmatory setting. In particular, the disease process is captured (albeit imperfectly) through one of more relevant biological markers and linked to a time-to-event process using two different approaches: a joint-model where subject-specific random effects are assumed to be shared by the disease and event processes (Tsiatis and Davidian (2001, 2004), Wang and Taylor (2001), Xu and Zeger (2001).), and a time-varying Cox extended model (Cox (1972), Kalbfleisch and Prentice (2002), Thurneau and Grambsch (2000).). We compare the two modeling approaches in terms of predictive accuracy in the context of the large-scale heart failure study Val-HeFT (Cohn(1999)), where time-varying biomarkers (e.g., B-type natriuretic peptides, internal left ventricle diastolic volume, etc), and in a few oncology studies where tumor-size was linked to outcomes. In addition, we study the gains and drawbacks of the use of longitudinal markers in time-to-event models for trials across diseases.