|
Activity Number:
|
250
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 5, 2003 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #301683 |
|
Title:
|
Combining Clinical and Microarray Data for Tumor Classification
|
|
Author(s):
|
Alexander D. Smith*+ and Jaya M. Satagopan and Andrew Stephenson and Michael W. Kattan and William Gerald
|
|
Companies:
|
Memorial Sloan-Kettering Cancer Center and Memorial Sloan-Kettering Cancer Center and Memorial Sloan-Kettering Cancer Center and Memorial Sloan-Kettering Cancer Center and Memorial Sloan-Kettering Cancer Center
|
|
Address:
|
1275 York Ave., Box 44, New York, NY, 10021-6007,
|
|
Keywords:
|
microarrays ; classification ; logistic regression
|
|
Abstract:
|
A common objective of microarray analyses is the creation of a molecular model to classify samples into appropriate biological groups. As these studies are extremely expensive and laborious, this objective is only of value if the molecular classifier performs more reliably than existing clinical measures. There is increasing interest in the combining of clinical and molecular information to create a more robust classifier. Given an appropriate and reasonably large independent test set of samples, one can compare classification rates based on the clinical variable(s) alone, a molecular model alone, or a model that combines both. We propose a simple modification to a stepwise logistic regression procedure that will allow any combination of gene or clinical information, without unduly weighting either. The procedure is demonstrated with a set of 78 prostate cancer tumor samples from MSKCC. The incidence of a biochemical recurrence was used to identify the response groups. Our results demonstrate that a combined model yields slightly improved classification over models based only on clinical variables or gene expressions.
|
