|
Activity Number:
|
606
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 6, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #304042 |
|
Title:
|
Nonparametric Estimation of Time-Dependent Predictive Accuracy Curve
|
|
Author(s):
|
Paramita Saha*+ and Patrick Heagerty
|
|
Companies:
|
University of Washington and University of Washington
|
|
Address:
|
F-600 Health Sciences Building, Seattle, WA, 98195,
|
|
Keywords:
|
sensitivity ; specificity ; time-dependent ROC ; ROC ; AUC
|
|
Abstract:
|
A major biomedical goal for developing predictive survival model is to accurately distinguish between incident cases at t from the controls surviving beyond t. Extensions of standard binary classification measures like time-dependent True & False Positives & ROC curve have become popular in this context. AUC curve has been introduced as a measure of discrimination of the marker throughout the entire study period. However, the existing AUC curve estimators are not estimated directly; they are derived from ROC curve via numerical integration. We propose a direct, nonparametric estimator of the time-dependent AUC curve without estimating the ROC curve first. The proposed method extends nonparametric AUC estimator arising from a binary data context & possesses desirable asymptotic properties. An overall measure of concordance is proposed. Time-dependent marker can also be accommodated.
|
