Methods for Reclassification Calibration in the Survival Setting

• Activity Number: 595
• Type: Topic Contributed
• Date/Time: Wednesday, August 12, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Olga Demler* and Nina Paynter and Nancy Cook
• Companies: Brigham and Women's Hospital and Brigham and Women's Hospital and Brigham and Women's Hospital
• Keywords: survival analysis ; reclassification calibration ; goodness of fit ; Hosmer-Lemeshow

Abstract

Risk stratification based on prediction models can be useful to clinicians in making treatment decisions. When developing a new prediction model, risk reclassification tables can help to better understand the changes to such stratification introduced by a new model. The concept of reclassification calibration (RC) has been applied in a logistic regression framework to compare the fit of two models within important regions of risk. We extend this result to the survival setting. We apply the Greenwood-Nam-D'Agostino test to the RC table and consider various strategies for collapsing categories with small numbers of observations. We discuss the performance of the test and show that it attains 5% size for a variety of scenarios. We also demonstrate that unlike the Hosmer-Lemeshow-style approach the RC table is sensitive to the omitted important predictor.