Empirical Comparison of Continuous and Discrete-time Representations for Survival Prediction

Michael Sloma, Fayeq Syed, Mohammedreza Nemati, Kevin S. Xu
Proceedings of AAAI Spring Symposium on Survival Prediction - Algorithms, Challenges, and Applications 2021, PMLR 146:118-131, 2021.

Abstract

Survival prediction aims to predict the time of occurrence of a particular event of interest, such as the time until a patient dies. The main challenge in survival prediction is the presence of incomplete observations due to censoring. The classical formulation for survival prediction treats the survival time as a continuous outcome, which leads to a censored regression problem. Recent work has reformulated the survival prediction problem by discretizing time into a finite number of bins and then applying multi-task binary classification. While the discrete-time formulation is convenient and potentially requires less assumptions than the continuous-time approach, it also loses information by discretizing time. In this paper, we empirically investigate continuous and discrete-time representations for survival prediction to try to quantify the trade-offs between the two formulations. We find that discretizing time does not necessarily decrease prediction accuracy. Furthermore, discrete-time models can result in even more accurate predictors than continuous-time models, but the number of time bins used for discretization has a significant effect on accuracy and should thus be tuned as a hyperparameter rather than specified for convenience.

Cite this Paper

BibTeX
@InProceedings{pmlr-v146-sloma21a, title = {Empirical Comparison of Continuous and Discrete-time Representations for Survival Prediction}, author = {Sloma, Michael and Syed, Fayeq and Nemati, Mohammedreza and Xu, Kevin S.}, booktitle = {Proceedings of AAAI Spring Symposium on Survival Prediction - Algorithms, Challenges, and Applications 2021}, pages = {118--131}, year = {2021}, editor = {Greiner, Russell and Kumar, Neeraj and Gerds, Thomas Alexander and van der Schaar, Mihaela}, volume = {146}, series = {Proceedings of Machine Learning Research}, month = {22--24 Mar}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v146/sloma21a/sloma21a.pdf}, url = {https://proceedings.mlr.press/v146/sloma21a.html}, abstract = {Survival prediction aims to predict the time of occurrence of a particular event of interest, such as the time until a patient dies. The main challenge in survival prediction is the presence of incomplete observations due to censoring. The classical formulation for survival prediction treats the survival time as a continuous outcome, which leads to a censored regression problem. Recent work has reformulated the survival prediction problem by discretizing time into a finite number of bins and then applying multi-task binary classification. While the discrete-time formulation is convenient and potentially requires less assumptions than the continuous-time approach, it also loses information by discretizing time. In this paper, we empirically investigate continuous and discrete-time representations for survival prediction to try to quantify the trade-offs between the two formulations. We find that discretizing time does not necessarily decrease prediction accuracy. Furthermore, discrete-time models can result in even more accurate predictors than continuous-time models, but the number of time bins used for discretization has a significant effect on accuracy and should thus be tuned as a hyperparameter rather than specified for convenience.} }
Endnote
%0 Conference Paper %T Empirical Comparison of Continuous and Discrete-time Representations for Survival Prediction %A Michael Sloma %A Fayeq Syed %A Mohammedreza Nemati %A Kevin S. Xu %B Proceedings of AAAI Spring Symposium on Survival Prediction - Algorithms, Challenges, and Applications 2021 %C Proceedings of Machine Learning Research %D 2021 %E Russell Greiner %E Neeraj Kumar %E Thomas Alexander Gerds %E Mihaela van der Schaar %F pmlr-v146-sloma21a %I PMLR %P 118--131 %U https://proceedings.mlr.press/v146/sloma21a.html %V 146 %X Survival prediction aims to predict the time of occurrence of a particular event of interest, such as the time until a patient dies. The main challenge in survival prediction is the presence of incomplete observations due to censoring. The classical formulation for survival prediction treats the survival time as a continuous outcome, which leads to a censored regression problem. Recent work has reformulated the survival prediction problem by discretizing time into a finite number of bins and then applying multi-task binary classification. While the discrete-time formulation is convenient and potentially requires less assumptions than the continuous-time approach, it also loses information by discretizing time. In this paper, we empirically investigate continuous and discrete-time representations for survival prediction to try to quantify the trade-offs between the two formulations. We find that discretizing time does not necessarily decrease prediction accuracy. Furthermore, discrete-time models can result in even more accurate predictors than continuous-time models, but the number of time bins used for discretization has a significant effect on accuracy and should thus be tuned as a hyperparameter rather than specified for convenience.
APA
Sloma, M., Syed, F., Nemati, M. & Xu, K.S.. (2021). Empirical Comparison of Continuous and Discrete-time Representations for Survival Prediction. Proceedings of AAAI Spring Symposium on Survival Prediction - Algorithms, Challenges, and Applications 2021, in Proceedings of Machine Learning Research 146:118-131 Available from https://proceedings.mlr.press/v146/sloma21a.html.

Related Material