Regression When the Predictor may be Censored

• Activity Number: 160
• Type: Topic Contributed
• Date/Time: Monday, August 5, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Risk Analysis
• Author(s): David Oakes*+
• Companies: Univ of Rochester Medical Center
• Keywords: biomarkers ; missing data ; prediction ; survival data

Abstract

There is a voluminous literature on regression models where the response variable is subject to right censoring. Situations where a predictor variable is subject to right censoring have received much less attention. However the increasing use of biomarkers as predictors in clinical studies has focussed attention on this area. Unlike in situations with censored response variables, simply omitting observations with the censored predictors does not lead to bias in the estimation of the regression coefficient. However there may be a substantial loss of efficiency. We explore some approaches to recovery of the available information from the censored observations using both parametric and nonparametric methods and report on the results of an illustrative simulation.