Piecewise exponential artificial neural networks (PEANN) for modeling hazard function with right censored data

Abstract

The hazard function plays an important role in the study of disease dynamics in survival analysis. Longer follow-up for various kinds of cancer, particularly breast cancer, has made it possible the observation of complex shapes of the hazard function of occurrence of metastasis and death. The identification of the correct hazard shape is important both for formulation and support of biological hypotheses on the mechanism underlying the disease. In this paper we propose the use of a neural network to model the shape of the hazard function in time in dependence of covariates extending the piecewise exponential model. The use of neural networks accommodates a greater flexibility in the study of the hazard shape. © 2014 Springer International Publishing Switzerland.