IDI Analysis for Competing-Risk Censored Outcomes

• Activity Number: 654
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Risk Analysis
• Author(s): Haiwen Shi*+ and Yu Cheng
• Companies: University of Pittsburgh and University of Pittsburgh
• Keywords: Competing risk ; Area under the receiver operating characteristic curve ; Bivariate survival function ; Bivariate cumulative incidence function ; Integrated discrimination improvement ; Improvement in the area under the curve

Abstract

In this paper we investigate a complementary criterion to the Area Under the Receiver Operating Characteristic curve for evaluating the added value of new markers over conventional predictors for an event outcome. Uno et al. (2009) considered the highly cited criterion Integrated Discrimination Improvement (IDI) by Pencina et al. (2008), and proposed an inverse weighting estimator for IDI based on conditional distributions of the difference in predicted risks with and without new markers, given that subjects have or have not developed the event. We propose alternative estimators for the conditional distributions using the Dabrowska estimator for bivariate survival functions (Dabrowska, 1988). Furthermore, we extend Uno's estimators to a competing risk setting where the outcome of interest is subject to competing-risk censoring. We consider fitting regression models based on the cumulative incidence function (CIF) for the primary event. Parallel estimators are also proposed based on the bivariate CIF estimators that were developed in Cheng et al. (2007). Both estimators perform well in simulation studies and in an application to a breast cancer dataset.