Deep Survival Machines: Fully Parametric Survival Regression and\n Representation Learning for Censored Data with Competing Risks

Chirag Nagpal, Xinyu Rachel Li, Artur Dubrawski

arXiv (Cornell University) · 2020 · 21 citations · 0 references

DOIFull text

Open access

Concepts

Abstract

We describe a new approach to estimating relative risks in time-to-event\nprediction problems with censored data in a fully parametric manner. Our\napproach does not require making strong assumptions of constant proportional\nhazard of the underlying survival distribution, as required by the\nCox-proportional hazard model. By jointly learning deep nonlinear\nrepresentations of the input covariates, we demonstrate the benefits of our\napproach when used to estimate survival risks through extensive experimentation\non multiple real world datasets with different levels of censoring. We further\ndemonstrate advantages of our model in the competing risks scenario. To the\nbest of our knowledge, this is the first work involving fully parametric\nestimation of survival times with competing risks in the presence of censoring.\n