Percentage of Survival Variation Explained by a Continuous Marker

• Activity Number: 460
• Type: Topic Contributed
• Date/Time: Wednesday, August 5, 2009 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Caixia Li*+ and Ying Lu
• Companies: University of California, San Francisco and University of California, San Francisco
• Keywords: Explained variation ; R-square ; Degree of separation ; Survival function

Abstract

A higher R2 indicates a stronger prediction of a continuous marker to a continuous outcome. Survival time is a continuous variable often with censoring. This paper uses a nonparametric R2 for conditional expected limiting survival time to evaluate predictive power of a continuous marker. Its discrete version matches the degree of separation (DOS) proposed by Lu, Jin, and Mi (2004) to compare discriminant efficiency of classification rules. Based on different cut-off values of the marker, a R2 curve or surface is proposed to create exclusive discrete risk groups. To determine if one marker predict survival better than another, a bootstrap test compares their corresponding R2 from the same cohort in a prospective study. We illustrate the method using data from the Osteoporosis & Ultrasound (OPUS) Study to assess and compare the predictive power of DXA and QUS for future fracture.