Online Program

Latent Supervised Learning for Survival Data

Michael Rene Kosorok, University of North Carolina at Chapel Hill 
*Susan Wei, UNC Chapel Hill 

Keywords: Censoring, Classification and Clustering, Cox Model, Inverse Probability of Censoring Weight, Proportional Hazards, Random Forest, Statistical Learning, Sieve Maximum Likelihood Estimation, Sliced Inverse Regression, Survival Analysis

Consider right-censored survival data with covariates. Given two subgroups, a basic task in survival analysis is to determine whether the survival experience between the two groups is different. Less studied are problems whereby the subgroup labels are unknown \emph{a priori}. The focus of this paper is on one such problem -- the discovery of subgroups in the covariate space that differ in survival. Existing methods utilize either survival data alone or covariate data alone. An integrated approach is proposed here which borrows techniques from latent supervised learning, a recent development in the field of machine learning. Theoretical properties of the methodology are studied. A comparison of the proposed methodology to other competitors in various simulations settings is conducted as well. Finally the applicability of the method is demonstrated in some real data examples.