DeepCompete : A deep learning approach to competing risks in continuous time domain

An increasing number of people survive longer ages leading to a growing population of people 65 years of age or older. A large percentage of this population is afflicted with multiple acute diseases (multi-morbidity). Clinicians need new tools to quantify the relative risk of an adverse event due to each competing disease and prioritize treatment among various diseases affecting a patient. Currently available deep learning survival analysis models have limited ability to incorporate multiple risks. Also, deep learning survival analysis models in current literature work predominantly in the discrete-time domain, while all biochemical processes continuously happen in the body. In this work, we introduce a novel architecture for a continuous-time deep learning model to combat these two issues, DeepCompete, aimed at survival analysis for competing risks. Our model learns the risk of each disease in an entirely data-driven fashion without making strong assumptions about the underlying stochastic processes. Further, we demonstrate that our model has superior results compared to state of the art continuous-time statistical models for survival analysis.