Tree Models for Prediction of Survival Times from Clinical and Genomic Data

• Activity Number: 197
• Type: Topic Contributed
• Date/Time: Tuesday, August 13, 2002 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing*
• Author(s): Jennifer Pittman*+ and Mike West
• Affiliation(s): Duke University and Duke University
• Keywords: microarrays ; Bayesian trees ; classification ; survival analysis

Abstract

The success of a treatment regiment for cancer patients depends on accurate initial diagnoses and predictions of patient prognoses. These decisions require the identification of reliable prognostic factors. Clinical information, however, is often incomplete or based on subjective interpretation and, hence, can be misleading. Given the significance of RNA/protein expression in cancer pathology and recent studies of genomic data in cancer diagnosis, we are currently developing tree-based methods for predicting disease-free survival from tumor gene expression profiles and clinical information. These trees use classification rules built on Bayesian measures of association to identify genes whose tumor expression patterns classify patients into useful prognostic genotypes. This genotyping by trees and identification of genetic predictors could aid clinicians in the treatment of cancer patients. In this talk, the development of these models will be discussed, and an example of the application of such models to outcome prediction will be presented.