MSnet: A deep neural network based on piecewise-constant proposals within Multi-State event history analysis

Aziliz Cottin, Marine Zulian, Sandrine Katsahian, Agathe Guilloux
Proceedings of the 7th Conference on Health, Inference, and Learning, PMLR 333:245-277, 2026.

Abstract

Multi-state models are essential to represent realistic disease trajectories in oncology, yet most existing survival and deep-learning approaches either rely on restrictive Markov assumptions or fail to provide subject-specific transition risks. We propose MSnet, a deep learning framework for progressive semi-Markov multi-state processes with right-censoring. MSnet models transition-specific cumulative risks as functions of sojourn time using a multi-task architecture that flexibly integrates high-dimensional clinical and omics data. Experiments on simulated data and two real-world breast cancer cohorts show that MSnet improves predictive performance while yielding clinically interpretable transition dynamics, extending deep learning–based survival analysis to more realistic, patient-centered disease processes.

Cite this Paper

BibTeX
@InProceedings{pmlr-v333-cottin26a, title = {MSnet: A deep neural network based on piecewise-constant proposals within Multi-State event history analysis}, author = {Cottin, Aziliz and Zulian, Marine and Katsahian, Sandrine and Guilloux, Agathe}, booktitle = {Proceedings of the 7th Conference on Health, Inference, and Learning}, pages = {245--277}, year = {2026}, editor = {Healey, Elizabeth and Fries, Jason and Pollard, Tom and Tang, Shengpu and Zink, Anna and Hartvigsen, Tom and Agrawal, Monica and Finlayson, Sam and Glicksberg, Benjamin and Beaulieu-Jones, Brett and Wang, Kai and Fontalvo, Daseyra and Sarker, Tasmie and Chen, Irene and Alsentzer, Emily}, volume = {333}, series = {Proceedings of Machine Learning Research}, month = {29--30 Jun}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v333/main/assets/cottin26a/cottin26a.pdf}, url = {https://proceedings.mlr.press/v333/cottin26a.html}, abstract = {Multi-state models are essential to represent realistic disease trajectories in oncology, yet most existing survival and deep-learning approaches either rely on restrictive Markov assumptions or fail to provide subject-specific transition risks. We propose MSnet, a deep learning framework for progressive semi-Markov multi-state processes with right-censoring. MSnet models transition-specific cumulative risks as functions of sojourn time using a multi-task architecture that flexibly integrates high-dimensional clinical and omics data. Experiments on simulated data and two real-world breast cancer cohorts show that MSnet improves predictive performance while yielding clinically interpretable transition dynamics, extending deep learning–based survival analysis to more realistic, patient-centered disease processes.} }
Endnote
%0 Conference Paper %T MSnet: A deep neural network based on piecewise-constant proposals within Multi-State event history analysis %A Aziliz Cottin %A Marine Zulian %A Sandrine Katsahian %A Agathe Guilloux %B Proceedings of the 7th Conference on Health, Inference, and Learning %C Proceedings of Machine Learning Research %D 2026 %E Elizabeth Healey %E Jason Fries %E Tom Pollard %E Shengpu Tang %E Anna Zink %E Tom Hartvigsen %E Monica Agrawal %E Sam Finlayson %E Benjamin Glicksberg %E Brett Beaulieu-Jones %E Kai Wang %E Daseyra Fontalvo %E Tasmie Sarker %E Irene Chen %E Emily Alsentzer %F pmlr-v333-cottin26a %I PMLR %P 245--277 %U https://proceedings.mlr.press/v333/cottin26a.html %V 333 %X Multi-state models are essential to represent realistic disease trajectories in oncology, yet most existing survival and deep-learning approaches either rely on restrictive Markov assumptions or fail to provide subject-specific transition risks. We propose MSnet, a deep learning framework for progressive semi-Markov multi-state processes with right-censoring. MSnet models transition-specific cumulative risks as functions of sojourn time using a multi-task architecture that flexibly integrates high-dimensional clinical and omics data. Experiments on simulated data and two real-world breast cancer cohorts show that MSnet improves predictive performance while yielding clinically interpretable transition dynamics, extending deep learning–based survival analysis to more realistic, patient-centered disease processes.
APA
Cottin, A., Zulian, M., Katsahian, S. & Guilloux, A.. (2026). MSnet: A deep neural network based on piecewise-constant proposals within Multi-State event history analysis. Proceedings of the 7th Conference on Health, Inference, and Learning, in Proceedings of Machine Learning Research 333:245-277 Available from https://proceedings.mlr.press/v333/cottin26a.html.

Related Material