Sensitivity of Cox Regression to Nonignorable Censoring

• Activity Number: 103
• Type: Contributed
• Date/Time: Monday, July 30, 2007 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Tao Liu*+ and Daniel F. Heitjan
• Companies: Brown University and University of Pennsylvania
• Keywords: informative censoring ; ignorability ; sensitivity analysis ; coarse-data model ; ISNI

Abstract

The validity of the partial likelihood in the Cox proportional hazards model for censored survival data is contingent on the assumption that the censoring mechanism is ignorable. Nonignorable censoring can spoil inferences made under an ignorable model. In this paper, we propose to conduct a local sensitivity analysis to address this issue, assuming that the true censoring mechanism is unknown but not far from ignorable. We express the censoring mechanism by the coarse-data selection model, and derive a profile likelihood to account for the random censoring. We show that the index of local sensitivity to nonignorability (ISNI) (Troxel et al., 2004) can be seamlessly extended to the profile likelihood, and we apply it to assess local sensitivity of the Cox regression to nonignorability. We demonstrate our method using data from a study of prognostic variables in advanced lung cancer.