Nonparametric Estimation of Concordance Probability with Censored Data

• Activity Number: 571
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Mithat Gonen*+ and Glenn Heller and Qianxing Mo
• Companies: Memorial Sloan-Kettering Cancer Center and Memorial Sloan-Kettering Cancer Center and Memorial Sloan-Kettering Cancer Center
• Keywords: concordance ; right-censored ; risk score ; cancer staging ; Kaplan-Meier

Abstract

Concordance probability is commonly used to summarize the discriminatory power of a predictive model. Estimation with binary outcomes is straightforward using the all-pairs U statistic but its popular adaptation to censored data, the "informative-pairs" c-index, has undesirable properties. It is possible to base the estimation of concordance on Kaplan-Meier estimates when discrete groups are used as predictors. This estimator is consistent only up to the maximum observed event time. This has practical implications on reporting and interpretation. This talk investigates whether the use of extended Kaplan-Meier estimates in the estimation of concordance probability will result in lower mean-squared error when compared with the traditional Kaplan-Meier method truncated at the maximum event time. Staging of colon cancer will be used as an example throughout the presentation.