Survival Data Analysis using Neural Networks

Abstract

This thesis introduces a new approach in survival data analysis i.e., Cox regression using neural networks. It is implemented on a data set of 575 patients which was provided by the CRC trials at Birmingham University. In the first part of the project, the standard Cox regression method was implemented. The objective of Cox regression is to model the probability functions of the patients based on a fundamental hypothesis which is described in the thesis. First the method was implemented on synthetic data in order to check its performance. Then it was implemented on the real data, and the results were compared with the ones found by the statisticians at the CRC trials. In the second part the new approach was introduced: Cox regression using neural networks. It was again applied first to synthetic data, and then to the real data. The results obtained were compared with the ones found by the implementation of standard Cox method. In the third and last part, the cumulative baseline hazard function was estimated. After applying the method to synthetic data, as before, it was implemented on the real data, and the cumulative hazard function and survival probability were also estimated. The implementation was done using both approaches and the results were compared. The conclusions derived from the results obtained are discussed at the end of the thesis.