Predicting Time-to-Relapse in Breast Cancer Using Neural Networks
Abstract
We implemented neural network (MN) algorithms for analysis of censored-data in predicting time to relapse for breast cancer patients, including a generalization of the Buckley-James approach to censored linear regression, and the methods proposed by Faraggi and Simon and Liestol et al. The data set available included 236 women with node negative breast cancer treated with surgery only. In Cox models HER-2/neu amplification, tumor size, treatment center, and age were univariately associated with outcome, but only HER-2/neu and treatment center significant in multivariate analyses. The recursive partitioning method selected HER-2/neu as the strongest predictor, and divided the non-amplified group by treatment center, and the amplified group by nuclear grade.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1997
- Accession Number
- ADA346651
Entities
People
- Jonathan D. Buckley
Organizations
- University of Southern California