SemiCompDNN
Overview
Prognostication for individuals with lung cancer is a complex task, often relying on the use of risk factors and health events spanning their entire life course. One challenge is that an individual’s disease course involves non-terminal (e.g., disease progression) and terminal (e.g., death) events, which form semi-competing relationships. By semi-competing, we mean that the occurrence of a non-terminal event is subject to the occurrence of a terminal event.
Following developments in the prediction of time-to-event outcomes with neural networks, deep learning has become a key area of focus for the development of risk prediction methods in survival analysis. However, limited work has been done to predict multi-state or competing risk outcomes, let alone semi-competing outcomes. To address this, we propose a novel neural expectation-maximization algorithm, in which we hope to bridge the gap between classical statistical approaches and machine learning. Our algorithm allows us to estimate the non-parametric baseline hazards of each state transition, risk functions of our predictors, and the degree of dependence among different transitions by utilizing a multi-task deep neural network with transition-specific sub-architectures. As deep learning can recover non-linear risk scores, we test our method by simulating risk surfaces of varying complexity.