PseudoNAM: A Pseudo Value Based Interpretable Neural Additive Model for Survival Analysis

Abstract

Deep learning models have achieved the start-of-the-art per formance in survival analysis as they can handle censor ing while learning complex nonlinear hidden representa tions directly from the raw data. However, the covariate ef fects on survival probabilities are difficult to explain using deep learning models. To address this challenge, we propose PseudoNAM - an interpretable model which uses pseudo val ues to efficiently handle censoring and uses neural additive networks to capture the nonlinearity in the covariates of the survival data. In particular, PseudoNAM uses neural addi tive models to jointly learn a linear combination of neural networks corresponding to each covariate and identifies the effect of the individual covariate on the output, and thus, is inherently interpretable. We show that our PseudoNAM out puts can be used in other survival models such as random survival forests to obtain improved survival prediction per formance. Our experiments on three real-world survival anal ysis datasets demonstrate that our proposed models achieve similar or better performance (in terms of C-index and Brier scores) than the state-of-the-art survival methods. We show case that PseudoNAM provides overall feature importance scores and feature-level interpretations (covariate effect on survival risk) for survival predictions at different time points.