Abstract
The objective of this work is to develop and study dynamic patient-state models and patient-state representations that are predictive of a wide range of future events in the electronic health records (EHRs). One challenge to overcome when building predictive EHRs representations is the complexity of multivariate clinical event time-series and their short and long-term dependencies. We address this challenge by proposing a new neural memory module called Multi-scale Temporal Memory (MTM) linking events in a distant past with the current prediction time. Through a novel mechanism implemented in MTM, information about previous events on different time-scales is compiled and read on-the-fly for prediction through memory contents. We demonstrate the efficacy of MTM by combining it with different patient state summarization methods that cover different temporal aspects of patient states. We show that the combined approach is 4.6% more accurate than the best result among the baseline approaches and it is 16% more accurate than prediction solely through hidden states of LSTM.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Bajor, J.M., Lasko, T.A.: Predicting medications from diagnostic codes with recurrent neural networks. In: ICLR (2017)
Bengio, Y., et al.: A neural probabilistic language model. J. Mach. Learn. Res. 3(Feb), 1137–1155 (2003)
Choi, E., et al.: Doctor AI: predicting clinical events via recurrent neural networks. In: Machine Learning for Healthcare Conference, pp. 301–318 (2016)
Choi, E., et al.: Multi-layer representation learning for medical concepts. In: The 22nd International Conference on Knowledge Discovery and Data Mining (2016)
Choi, E., et al.: Retain: an interpretable predictive model for healthcare using reverse time attention mechanism. In: Neural Information Processing Systems (2016)
Choi, E., et al.: Using recurrent neural network models for early detection of heart failure onset. J. Am. Med. Inform. Assoc. 24, 361–370 (2017)
Esteban, C., et al.: Predicting clinical events by combining static and dynamic information using RNN. In: International Conference on Healthcare Informatics (ICHI) (2016)
Hauskrecht, M., et al.: Outlier detection for patient monitoring and alerting. J. Biomed. Inform. 46(1), 47–55 (2013)
Hauskrecht, M., et al.: Outlier-based detection of unusual patient-management actions: an ICU study. J. Biomed. Inform. 64, 211–221 (2016)
Hochreiter, S., Schmidhuber, J.: Long short-term memory. Neural Comput. 9, 1735–1780 (1997)
Hochreiter, S., et al.: Gradient flow in recurrent nets: the difficulty of learning long-term dependencies (2001)
Johnson, A.E., et al.: MIMIC-III, a freely accessible critical care database. Sci. Data 3, 160035 (2016)
Kingma, D.P., Ba, J.: Adam: a method for stochastic optimization. arXiv:1412.6980 (2014)
Laub, P.J., et al.: Hawkes processes. arXiv preprint arXiv:1507.02822 (2015)
Lee, J.M., Hauskrecht, M.: Recent-context-aware LSTM-based clinical time-series prediction. In: In Proceedings of AI in Medicine Europe (AIME) (2019)
Lee, J.M., Hauskrecht, M.: Clinical event time-series modeling with periodic events. In: The Thirty-Third International Flairs Conference. AAAI (2020)
Liu, S., Hauskrecht, M.: Nonparametric regressive point processes based on conditional Gaussian processes. In: Neural Information Processing Systems (2019)
Liu, Z., Hauskrecht, M.: Clinical time series prediction: toward a hierarchical dynamical system framework. Artif. Intell. Med. 65(1), 5–18 (2015)
Liu, Z., Hauskrecht, M.: A regularized linear dynamical system framework for multivariate time series analysis. In: Twenty-Ninth AAAI Conference on Artificial Intelligence, pp. 1798–1804 (2015)
Liu, Z., Wu, L., Hauskrecht, M.: Modeling clinical time series using gaussian process sequences. In: Proceedings of the 2013 SIAM International Conference on Data Mining, pp. 623–631. SIAM (2013)
Malakouti, S., Hauskrecht, M.: Hierarchical adaptive multi-task learning framework for patient diagnoses and diagnostic category classification. In: International Conference on Bioinformatics and Biomedicine (BIBM) (2019)
Mei, H., Eisner, J.M.: The neural hawkes process: a neurally self-modulating multivariate point process. In: Neural Information Processing Systems (2017)
Mikolov, T., et al.: Distributed representations of words and phrases and their compositionality. In: Neural Information Processing Systems (2013)
Miotto, R., et al.: Deep patient: an unsupervised representation to predict the future of patients from the electronic health records. Sci. Rep. 6, 1–10 (2016)
Nemati, S., et al.: An interpretable machine learning model for accurate prediction of sepsis in the ICU. Crit. Care Med. 46(4), 547–553 (2018)
Pascanu, R., et al.: On the difficulty of training recurrent neural networks. In: International Conference on Machine Learning, pp. 1310–1318 (2013)
Rasmussen, J.G.: Lecture notes: temporal point processes and the conditional intensity function. arXiv preprint arXiv:1806.00221 (2018)
Wang, F., et al.: Composite distance metric integration by leveraging multiple experts’ inputs and its application in patient similarity assessment. Stat. Anal. Data Min.: ASA Data Sci. J. 5(1), 54–69 (2012)
Yu, K., et al.: Monitoring ICU mortality risk with a long short-term memory recurrent neural network. In: Pacific Symposium on Biocomputing. World Scientific (2020)
Acknowledgement
The work in this paper was supported by NIH grant R01GM088224. The content of the paper is solely the responsibility of the authors and does not necessarily represent the official views of NIH.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
A Examples of Top Past Events Predictive of Target Events
A Examples of Top Past Events Predictive of Target Events
Table 4 shows top past events predictive of target events for the different temporal ranges (Distant, Intermediate, and Recent past) as identified by our methods. For example, the top predictive events for amiodarone (treats irregular heartbeat such as tachycardia) include metoprolol and diltiazem. Both of these are used to treat high blood pressure and heart issues. Similarly, past events predictive of diltiazem and labetalol (medications treating high blood pressure) include clinical events that are related to high blood pressure and heart function: digoxin, metoprolol, hydralazine, and nicardipine. Finally, the top past events predicting vasopressin (a medication treating a low blood pressure) include norepinephrine and phenylephrine that are also used to treat low blood pressure.
Rights and permissions
Copyright information
© 2020 Springer Nature Switzerland AG
About this paper
Cite this paper
Lee, J.M., Hauskrecht, M. (2020). Multi-scale Temporal Memory for Clinical Event Time-Series Prediction. In: Michalowski, M., Moskovitch, R. (eds) Artificial Intelligence in Medicine. AIME 2020. Lecture Notes in Computer Science(), vol 12299. Springer, Cham. https://doi.org/10.1007/978-3-030-59137-3_28
Download citation
DOI: https://doi.org/10.1007/978-3-030-59137-3_28
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-59136-6
Online ISBN: 978-3-030-59137-3
eBook Packages: Computer ScienceComputer Science (R0)