Dipole: Diagnosis prediction in healthcare via attention-based bidirectional recurrent neural networks

Fenglong Ma, Radha Chitta, Jing Zhou, Quanzeng You, Tong Sun, Jing Gao

Research output: Chapter in Book/Report/Conference proceedingConference contribution

81 Citations (Scopus)

Abstract

Predicting the future health information of patients from the historical Electronic Health Records (EHR) is a core research task in the development of personalized healthcare. Patient EHR data consist of sequences of visits over time, where each visit contains multiple medical codes, including diagnosis, medication, and procedure codes. The most important challenges for this task are to model the temporality and high dimensionality of sequential EHR data and to interpret the prediction results. Existing work solves this problem by employing recurrent neural networks (RNNs) to model EHR data and utilizing simple attention mechanism to interpret the results. However, RNN-based approaches suffer from the problem that the performance of RNNs drops when the length of sequences is large, and the relationships between subsequent visits are ignored by current RNN-based approaches. To address these issues, we propose Dipole, an end-to-end, simple and robust model for predicting patients' future health information. Dipole employs bidirectional recurrent neural networks to remember all the information of both the past visits and the future visits, and it introduces three attention mechanisms to measure the relationships of different visits for the prediction. With the attention mechanisms, Dipole can interpret the prediction results effectively. Dipole also allows us to interpret the learned medical code representations which are confirmed positively by medical experts. Experimental results on two real worldEHR datasets show that the proposed Dipole can significantly improve the prediction accuracy compared with the state-of-the-art diagnosis prediction approaches and provide clinically meaningful interpretation.

Original languageEnglish (US)
Title of host publicationKDD 2017 - Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
PublisherAssociation for Computing Machinery
Pages1903-1911
Number of pages9
ISBN (Electronic)9781450348874
DOIs
StatePublished - Aug 13 2017
Event23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2017 - Halifax, Canada
Duration: Aug 13 2017Aug 17 2017

Publication series

NameProceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
VolumePart F129685

Other

Other23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2017
CountryCanada
CityHalifax
Period8/13/178/17/17

Fingerprint

Recurrent neural networks
Health

All Science Journal Classification (ASJC) codes

  • Software
  • Information Systems

Cite this

Ma, F., Chitta, R., Zhou, J., You, Q., Sun, T., & Gao, J. (2017). Dipole: Diagnosis prediction in healthcare via attention-based bidirectional recurrent neural networks. In KDD 2017 - Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 1903-1911). (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining; Vol. Part F129685). Association for Computing Machinery. https://doi.org/10.1145/3097983.3098088
Ma, Fenglong ; Chitta, Radha ; Zhou, Jing ; You, Quanzeng ; Sun, Tong ; Gao, Jing. / Dipole : Diagnosis prediction in healthcare via attention-based bidirectional recurrent neural networks. KDD 2017 - Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery, 2017. pp. 1903-1911 (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining).
@inproceedings{1e480522a9334a05bd1b908fd8733d29,
title = "Dipole: Diagnosis prediction in healthcare via attention-based bidirectional recurrent neural networks",
abstract = "Predicting the future health information of patients from the historical Electronic Health Records (EHR) is a core research task in the development of personalized healthcare. Patient EHR data consist of sequences of visits over time, where each visit contains multiple medical codes, including diagnosis, medication, and procedure codes. The most important challenges for this task are to model the temporality and high dimensionality of sequential EHR data and to interpret the prediction results. Existing work solves this problem by employing recurrent neural networks (RNNs) to model EHR data and utilizing simple attention mechanism to interpret the results. However, RNN-based approaches suffer from the problem that the performance of RNNs drops when the length of sequences is large, and the relationships between subsequent visits are ignored by current RNN-based approaches. To address these issues, we propose Dipole, an end-to-end, simple and robust model for predicting patients' future health information. Dipole employs bidirectional recurrent neural networks to remember all the information of both the past visits and the future visits, and it introduces three attention mechanisms to measure the relationships of different visits for the prediction. With the attention mechanisms, Dipole can interpret the prediction results effectively. Dipole also allows us to interpret the learned medical code representations which are confirmed positively by medical experts. Experimental results on two real worldEHR datasets show that the proposed Dipole can significantly improve the prediction accuracy compared with the state-of-the-art diagnosis prediction approaches and provide clinically meaningful interpretation.",
author = "Fenglong Ma and Radha Chitta and Jing Zhou and Quanzeng You and Tong Sun and Jing Gao",
year = "2017",
month = "8",
day = "13",
doi = "10.1145/3097983.3098088",
language = "English (US)",
series = "Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining",
publisher = "Association for Computing Machinery",
pages = "1903--1911",
booktitle = "KDD 2017 - Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining",

}

Ma, F, Chitta, R, Zhou, J, You, Q, Sun, T & Gao, J 2017, Dipole: Diagnosis prediction in healthcare via attention-based bidirectional recurrent neural networks. in KDD 2017 - Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, vol. Part F129685, Association for Computing Machinery, pp. 1903-1911, 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2017, Halifax, Canada, 8/13/17. https://doi.org/10.1145/3097983.3098088

Dipole : Diagnosis prediction in healthcare via attention-based bidirectional recurrent neural networks. / Ma, Fenglong; Chitta, Radha; Zhou, Jing; You, Quanzeng; Sun, Tong; Gao, Jing.

KDD 2017 - Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery, 2017. p. 1903-1911 (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining; Vol. Part F129685).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - Dipole

T2 - Diagnosis prediction in healthcare via attention-based bidirectional recurrent neural networks

AU - Ma, Fenglong

AU - Chitta, Radha

AU - Zhou, Jing

AU - You, Quanzeng

AU - Sun, Tong

AU - Gao, Jing

PY - 2017/8/13

Y1 - 2017/8/13

N2 - Predicting the future health information of patients from the historical Electronic Health Records (EHR) is a core research task in the development of personalized healthcare. Patient EHR data consist of sequences of visits over time, where each visit contains multiple medical codes, including diagnosis, medication, and procedure codes. The most important challenges for this task are to model the temporality and high dimensionality of sequential EHR data and to interpret the prediction results. Existing work solves this problem by employing recurrent neural networks (RNNs) to model EHR data and utilizing simple attention mechanism to interpret the results. However, RNN-based approaches suffer from the problem that the performance of RNNs drops when the length of sequences is large, and the relationships between subsequent visits are ignored by current RNN-based approaches. To address these issues, we propose Dipole, an end-to-end, simple and robust model for predicting patients' future health information. Dipole employs bidirectional recurrent neural networks to remember all the information of both the past visits and the future visits, and it introduces three attention mechanisms to measure the relationships of different visits for the prediction. With the attention mechanisms, Dipole can interpret the prediction results effectively. Dipole also allows us to interpret the learned medical code representations which are confirmed positively by medical experts. Experimental results on two real worldEHR datasets show that the proposed Dipole can significantly improve the prediction accuracy compared with the state-of-the-art diagnosis prediction approaches and provide clinically meaningful interpretation.

AB - Predicting the future health information of patients from the historical Electronic Health Records (EHR) is a core research task in the development of personalized healthcare. Patient EHR data consist of sequences of visits over time, where each visit contains multiple medical codes, including diagnosis, medication, and procedure codes. The most important challenges for this task are to model the temporality and high dimensionality of sequential EHR data and to interpret the prediction results. Existing work solves this problem by employing recurrent neural networks (RNNs) to model EHR data and utilizing simple attention mechanism to interpret the results. However, RNN-based approaches suffer from the problem that the performance of RNNs drops when the length of sequences is large, and the relationships between subsequent visits are ignored by current RNN-based approaches. To address these issues, we propose Dipole, an end-to-end, simple and robust model for predicting patients' future health information. Dipole employs bidirectional recurrent neural networks to remember all the information of both the past visits and the future visits, and it introduces three attention mechanisms to measure the relationships of different visits for the prediction. With the attention mechanisms, Dipole can interpret the prediction results effectively. Dipole also allows us to interpret the learned medical code representations which are confirmed positively by medical experts. Experimental results on two real worldEHR datasets show that the proposed Dipole can significantly improve the prediction accuracy compared with the state-of-the-art diagnosis prediction approaches and provide clinically meaningful interpretation.

UR - http://www.scopus.com/inward/record.url?scp=85029071016&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85029071016&partnerID=8YFLogxK

U2 - 10.1145/3097983.3098088

DO - 10.1145/3097983.3098088

M3 - Conference contribution

AN - SCOPUS:85029071016

T3 - Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

SP - 1903

EP - 1911

BT - KDD 2017 - Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

PB - Association for Computing Machinery

ER -

Ma F, Chitta R, Zhou J, You Q, Sun T, Gao J. Dipole: Diagnosis prediction in healthcare via attention-based bidirectional recurrent neural networks. In KDD 2017 - Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery. 2017. p. 1903-1911. (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining). https://doi.org/10.1145/3097983.3098088