Clinical Intervention Prediction and Understanding with Deep Neural Networks

Harini Suresh, Nathan Hunt, Alistair Johnson, Leo Anthony Celi, Peter Szolovits, Marzyeh Ghassemi
Proceedings of the 2nd Machine Learning for Healthcare Conference, PMLR 68:322-337, 2017.

Abstract

Real-time prediction of clinical interventions remains a challenge within intensive care units (ICUs). This task is complicated by data sources that are sparse, noisy, heterogeneous and outcomes that are imbalanced. In this work, we integrate data across many ICU sources — vitals, labs, notes, demographics — and focus on learning rich representations of this data to predict onset and weaning of multiple invasive interventions. In particular, we compare both long short-term memory networks (LSTM) and convolutional neural networks (CNN) for prediction of five intervention tasks: invasive ventilation, non-invasive ventilation, vasopressors, colloid boluses, and crystalloid boluses. Our predictions are done in a forward-facing manner after a six hour gap time to support clinically actionable planning. We achieve state-of-the-art results on these predictive tasks using deep architectures. Further, we explore the use of feature occlusion to interpret LSTM models, and compare this to the interpretability gained from examining inputs that maximally activate CNN outputs. We show that our models are able to significantly outperform baselines for intervention prediction, and provide insight into model learning.

Cite this Paper


BibTeX
@InProceedings{pmlr-v68-suresh17a, title = {Clinical Intervention Prediction and Understanding with Deep Neural Networks}, author = {Suresh, Harini and Hunt, Nathan and Johnson, Alistair and Celi, Leo Anthony and Szolovits, Peter and Ghassemi, Marzyeh}, booktitle = {Proceedings of the 2nd Machine Learning for Healthcare Conference}, pages = {322--337}, year = {2017}, editor = {Doshi-Velez, Finale and Fackler, Jim and Kale, David and Ranganath, Rajesh and Wallace, Byron and Wiens, Jenna}, volume = {68}, series = {Proceedings of Machine Learning Research}, month = {18--19 Aug}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v68/suresh17a/suresh17a.pdf}, url = {https://proceedings.mlr.press/v68/suresh17a.html}, abstract = {Real-time prediction of clinical interventions remains a challenge within intensive care units (ICUs). This task is complicated by data sources that are sparse, noisy, heterogeneous and outcomes that are imbalanced. In this work, we integrate data across many ICU sources — vitals, labs, notes, demographics — and focus on learning rich representations of this data to predict onset and weaning of multiple invasive interventions. In particular, we compare both long short-term memory networks (LSTM) and convolutional neural networks (CNN) for prediction of five intervention tasks: invasive ventilation, non-invasive ventilation, vasopressors, colloid boluses, and crystalloid boluses. Our predictions are done in a forward-facing manner after a six hour gap time to support clinically actionable planning. We achieve state-of-the-art results on these predictive tasks using deep architectures. Further, we explore the use of feature occlusion to interpret LSTM models, and compare this to the interpretability gained from examining inputs that maximally activate CNN outputs. We show that our models are able to significantly outperform baselines for intervention prediction, and provide insight into model learning.} }
Endnote
%0 Conference Paper %T Clinical Intervention Prediction and Understanding with Deep Neural Networks %A Harini Suresh %A Nathan Hunt %A Alistair Johnson %A Leo Anthony Celi %A Peter Szolovits %A Marzyeh Ghassemi %B Proceedings of the 2nd Machine Learning for Healthcare Conference %C Proceedings of Machine Learning Research %D 2017 %E Finale Doshi-Velez %E Jim Fackler %E David Kale %E Rajesh Ranganath %E Byron Wallace %E Jenna Wiens %F pmlr-v68-suresh17a %I PMLR %P 322--337 %U https://proceedings.mlr.press/v68/suresh17a.html %V 68 %X Real-time prediction of clinical interventions remains a challenge within intensive care units (ICUs). This task is complicated by data sources that are sparse, noisy, heterogeneous and outcomes that are imbalanced. In this work, we integrate data across many ICU sources — vitals, labs, notes, demographics — and focus on learning rich representations of this data to predict onset and weaning of multiple invasive interventions. In particular, we compare both long short-term memory networks (LSTM) and convolutional neural networks (CNN) for prediction of five intervention tasks: invasive ventilation, non-invasive ventilation, vasopressors, colloid boluses, and crystalloid boluses. Our predictions are done in a forward-facing manner after a six hour gap time to support clinically actionable planning. We achieve state-of-the-art results on these predictive tasks using deep architectures. Further, we explore the use of feature occlusion to interpret LSTM models, and compare this to the interpretability gained from examining inputs that maximally activate CNN outputs. We show that our models are able to significantly outperform baselines for intervention prediction, and provide insight into model learning.
APA
Suresh, H., Hunt, N., Johnson, A., Celi, L.A., Szolovits, P. & Ghassemi, M.. (2017). Clinical Intervention Prediction and Understanding with Deep Neural Networks. Proceedings of the 2nd Machine Learning for Healthcare Conference, in Proceedings of Machine Learning Research 68:322-337 Available from https://proceedings.mlr.press/v68/suresh17a.html.

Related Material