Survival Analysis via Density Estimation

Hiroki Yanagisawa, Shunta Akiyama

Abstract: This paper introduces a novel framework for survival analysis by reinterpreting it as a form of density estimation. Our algorithm post-processes density estimation outputs to derive survival functions, enabling the application of any density estimation model to effectively estimate survival functions. This approach broadens the toolkit for survival analysis and enhances the flexibility and applicability of existing techniques. Our framework is versatile enough to handle various survival analysis scenarios, including competing risk models for multiple event types. It can also address dependent censoring when prior knowledge of the dependency between event time and censoring time is available in the form of a copula. In the absence of such information, our framework can estimate the upper and lower bounds of survival functions, accounting for the associated uncertainty.

Lay Summary: Our paper introduces a novel algorithm for survival analysis, a regression task that estimates the probability distribution of a target label from a feature vector. A key aspect of this analysis is that some target labels are censored, meaning only a lower bound of the true target label is available. Survival analysis is crucial in fields such as healthcare and engineering, where accurate predictions can have significant impacts. We propose transforming survival analysis into a multiclass classification problem, specifically through density estimation. This transformation allows for the use of advanced models like LightGBM, which is known for its speed and enhanced prediction quality. Furthermore, we demonstrate that if the density estimates are close to the true values, the resulting survival predictions will be accurate, ensuring reliable results.

Link To Code: https://github.com/CyberAgentAILab/cenreg

Primary Area: General Machine Learning->Supervised Learning

Keywords: survival analysis, time-to-event analysis, competing risks, density estimation

Submission Number: 11297