|
Activity Number:
|
264
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 10, 2004 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #300811 |
|
Title:
|
Statistical Considerations in the Development and Validation of Prognostic Profiles Derived from Gene Expression Microarray Data
|
|
Author(s):
|
Lisa McShane*+ and Lara Lusa and Michael Radmacher and Joanna H. Shih and George Wright and Richard Simon
|
|
Companies:
|
National Cancer Institute and Istituto Nazionale Tumori of Milan and Kenyon College and National Cancer Institute and National Cancer Institute and National Institutes of Health
|
|
Address:
|
6130 Executive Blvd, Room 8126, Bethesda, MD, 20892-7434,
|
|
Keywords:
|
model validation ; microarray ; cross-validation ; prognostic models ; expression profiles
|
|
Abstract:
|
DNA microarray analysis has been used to develop classifiers based on expression profiles that predict disease outcome for cancer patients. Often some preliminary validation of the predictive power of a profile-based classifier is carried out using the same dataset that was used to derive the classifier. Techniques such as cross-validation can be used in this setting to assess predictive power, and if applied correctly, will result in a nearly unbiased estimate of predictive accuracy of a classifier. However, some investigators conducting these gene expression profiling studies have applied cross-validation techniques incorrectly and have obtained highly biased estimates of predictive accuracy. Also, some cross-validated estimates of predictive accuracy can have very high variance. Alternatively, some investigators have tried to validate predictors by testing the statistical significance of the association between cross-validated predicted classes and disease outcomes. We will discuss the limitations of these various approaches and make recommendations for proper validation strategies.
|
