An Empirical Study of Representation Learning for Reinforcement Learning in Healthcare

Taylor W. Killian, Haoran Zhang, Jayakumar Subramanian, Mehdi Fatemi, Marzyeh Ghassemi
Proceedings of the Machine Learning for Health NeurIPS Workshop, PMLR 136:139-160, 2020.

Abstract

Reinforcement Learning (RL) has recently been applied to sequential estimation and prediction problems identifying and developing hypothetical treatment strategies for septic patients, with a particular focus on offline learning with observational data. In practice, successful RL relies on informative latent states derived from sequential observations to develop optimal treatment strategies. To date, how best to construct such states in a healthcare setting is an open question. In this paper, we perform an empirical study of several information encoding architectures using data from septic patients in the MIMIC-III dataset to form representations of a patient state. We evaluate the impact of representation dimension, correlations with established acuity scores, and the treatment policies derived from them. We find that sequentially formed state representations facilitate effective policy learning in batch settings, validating a more thoughtful approach to representation learning that remains faithful to the sequential and partial nature of healthcare data.

Cite this Paper


BibTeX
@InProceedings{pmlr-v136-killian20a, title = {An Empirical Study of Representation Learning for Reinforcement Learning in Healthcare}, author = {Killian, Taylor W. and Zhang, Haoran and Subramanian, Jayakumar and Fatemi, Mehdi and Ghassemi, Marzyeh}, booktitle = {Proceedings of the Machine Learning for Health NeurIPS Workshop}, pages = {139--160}, year = {2020}, editor = {Emily Alsentzer and Matthew B. A. McDermott and Fabian Falck and Suproteem K. Sarkar and Subhrajit Roy and Stephanie L. Hyland}, volume = {136}, series = {Proceedings of Machine Learning Research}, month = {11 Dec}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v136/killian20a/killian20a.pdf}, url = {http://proceedings.mlr.press/v136/killian20a.html}, abstract = {Reinforcement Learning (RL) has recently been applied to sequential estimation and prediction problems identifying and developing hypothetical treatment strategies for septic patients, with a particular focus on offline learning with observational data. In practice, successful RL relies on informative latent states derived from sequential observations to develop optimal treatment strategies. To date, how best to construct such states in a healthcare setting is an open question. In this paper, we perform an empirical study of several information encoding architectures using data from septic patients in the MIMIC-III dataset to form representations of a patient state. We evaluate the impact of representation dimension, correlations with established acuity scores, and the treatment policies derived from them. We find that sequentially formed state representations facilitate effective policy learning in batch settings, validating a more thoughtful approach to representation learning that remains faithful to the sequential and partial nature of healthcare data.} }
Endnote
%0 Conference Paper %T An Empirical Study of Representation Learning for Reinforcement Learning in Healthcare %A Taylor W. Killian %A Haoran Zhang %A Jayakumar Subramanian %A Mehdi Fatemi %A Marzyeh Ghassemi %B Proceedings of the Machine Learning for Health NeurIPS Workshop %C Proceedings of Machine Learning Research %D 2020 %E Emily Alsentzer %E Matthew B. A. McDermott %E Fabian Falck %E Suproteem K. Sarkar %E Subhrajit Roy %E Stephanie L. Hyland %F pmlr-v136-killian20a %I PMLR %P 139--160 %U http://proceedings.mlr.press/v136/killian20a.html %V 136 %X Reinforcement Learning (RL) has recently been applied to sequential estimation and prediction problems identifying and developing hypothetical treatment strategies for septic patients, with a particular focus on offline learning with observational data. In practice, successful RL relies on informative latent states derived from sequential observations to develop optimal treatment strategies. To date, how best to construct such states in a healthcare setting is an open question. In this paper, we perform an empirical study of several information encoding architectures using data from septic patients in the MIMIC-III dataset to form representations of a patient state. We evaluate the impact of representation dimension, correlations with established acuity scores, and the treatment policies derived from them. We find that sequentially formed state representations facilitate effective policy learning in batch settings, validating a more thoughtful approach to representation learning that remains faithful to the sequential and partial nature of healthcare data.
APA
Killian, T.W., Zhang, H., Subramanian, J., Fatemi, M. & Ghassemi, M.. (2020). An Empirical Study of Representation Learning for Reinforcement Learning in Healthcare. Proceedings of the Machine Learning for Health NeurIPS Workshop, in Proceedings of Machine Learning Research 136:139-160 Available from http://proceedings.mlr.press/v136/killian20a.html.

Related Material