Online Program Home
My Program

Abstract Details

Activity Number: 449 - Evaluating Risk Predictions for Use in Decision-Making
Type: Invited
Date/Time: Wednesday, July 31, 2019 : 8:30 AM to 10:20 AM
Sponsor: ENAR
Abstract #300644 Presentation
Title: A General Framework for Using the Overall Concordance Statistic to Assess the Discriminatory Ability of Risk Predictions
Author(s): Li Cheung* and Qing Pan and Barry Graubard
Companies: National Cancer Institute and George Washington University and National Cancer Institute
Keywords: survival analysis; area under the ROC curve; interval censoring; complex survey

Harrell’s concordance (C) index is a widely used non-parametric measure of discrimination ability. However, the estimate of C depends on study-specific characteristics, such as the length of follow-up, the censoring distribution, and the population representativeness of the data. In addition to bias issues, even for simple random samples, some commonly used softwares for estimating C provide overly conservative variance estimates. Building upon previously published work, we propose a modified concordance index that measures overall discriminatory ability up to a time tau and that converges to a censoring-independent quantity. This modified C-index can be applied to interval- or right-censored event time data or to complex survey data, allowing for more meaningful comparison of concordance statistics estimated from different data sets. We derive closed-form variance for the modified C-index and for the difference between 2 correlated C indices, using Taylor linearization methods that can be easily implemented with available software. Results from our simulation studies suggest that the modified C-index and proposed variance methods will perform well in finite samples.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program