|
Activity Number:
|
384
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 15, 2002 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
General Methodology
|
| Abstract - #301834 |
|
Title:
|
Analysis of Several Correlated Nonparametric ROC Curves in the Presence of Missing Data
|
|
Author(s):
|
Alicia Toledano*+
|
|
Affiliation(s):
|
Brown University
|
|
Address:
|
167 Angell Street, 2nd Floor, Box G-H, Providence, Rhode Island, 02912, USA
|
|
Keywords:
|
categorical data ; structural components ; U-statistics
|
|
Abstract:
|
This paper discusses relationships among methods for estimating elements of the variance-covariance matrix for a vector of correlated areas under nonparametric ROC curves computed from ordered categorical data, extends available methods to accommodate data arising from an incomplete factorial design with additional missingness, and explores the impact of retaining certain terms appearing in the expression for the asymptotic variance-covariance matrix that are often dropped in estimators thereof. These terms relate to the probability of ties, which for categorical data does not approach zero as the sample size approaches infinity. The work was motivated by a study that evaluated the accuracy of a new diagnostic imaging procedure, with images reviewed by several radiologists on three different workstations. We review the complex design of this study, and complications that arose due to missing data and particular patterns of responses made by some of the radiologists. The method used to analyze the resulting data is developed as an extension of previously reported methods. A simulation study provides further insight into the properties of the proposed estimator.
|
- The address information is for the authors that have a + after their name.
- Authors who are presenting talks have a * after their name.
Back to the full JSM 2002 program |
