|
Activity Number:
|
26
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 3, 2008 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
| Abstract - #302563 |
|
Title:
|
Evaluation of Prediction in Models of Composite Endpoints and Their Components
|
|
Author(s):
|
Robert Glynn*+ and Bernard Rosner
|
|
Companies:
|
Brigham & Women's Hospital and Channing Laboratory
|
|
Address:
|
Division of Preventive Medicine, Boston, MA, 02215,
|
|
Keywords:
|
prediction ; C-statistic ; competing risks ; survival
|
|
Abstract:
|
Composite endpoints are commonly used in both observational and randomized studies for a broad perspective on the impact of exposures of interest. In the setting of time to occurrence of outcomes, one can use methods of competing risks survival analysis to evaluate possible heterogeneity of associations of exposures with different components of the composite endpoint. We present two alternative approaches to quantify discrimination in such competing risks models. One relative approach considers likelihood ratios from nested models that assume common vs. distinct associations of exposures on different components of the endpoint. A second approach generalizes the C-statistic proposed by Harrell to the setting of composite endpoints. Approaches are illustrated with data on a composite endpoint including first occurrence of arterial or venous thrombosis in a prospective cohort study.
|
