Exploring Center Effects in Randomized Studies with Censored Endpoints

• Activity Number: 72
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 4:00 PM to 5:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Richard McNally*+ and Sarah H. Kogut
• Companies: Celgene Corporation and Celgene Corporation
• Keywords: survival analysis ; multicenter randomized trials ; graphical inference ; sensitivity analysis

Abstract

Randomized oncology studies usually need a large number of centers, resulting in a low subject-to-center ratio. This is made worse for time-to-event endpoints with censoring. Inference for treatment effects by individual centers and treatment center interactions has all of the difficulties associated with small samples (eg, large confidence intervals), and the full model with center effects added is likely to have non-identifiable parameters. Current solutions (eg, pooling centers) are unsatisfactory. We propose graphics-based techniques for assessing the effect of centers on the overall treatment effect and on the statistical model. Our approach specifically examines heterogeneity across centers and its effect on the overall statistical conclusions, and can be used with any model used to analyze censored time-to-event data. We give examples and practical recommendations.