Prediction of mortality from 12-lead electrocardiogram voltage data using a deep neural network

Sushravya Raghunath; Alvaro E. Ulloa Cerna; Linyuan Jing; David P. vanMaanen; Joshua Stough; Dustin N. Hartzel; Joseph B. Leader; H. Lester Kirchner; Martin C. Stumpe; Ashraf Hafez; Arun Nemani; Tanner Carbonati; Kipp W. Johnson; Katelyn Young; Christopher W. Good; John M. Pfeifer; Aalpen A. Patel; Brian P. Delisle; Amro Alsaid; Dominik Beer; Christopher M. Haggerty; Brandon K. Fornwalt

doi:10.1038/s41591-020-0870-z

Prediction of mortality from 12-lead electrocardiogram voltage data using a deep neural network

Sushravya Raghunath, Alvaro E. Ulloa Cerna, Linyuan Jing, David P. vanMaanen, Joshua Stough, Dustin N. Hartzel, Joseph B. Leader, H. Lester Kirchner, Martin C. Stumpe, Ashraf Hafez, Arun Nemani, Tanner Carbonati, Kipp W. Johnson, Katelyn Young, Christopher W. Good, John M. Pfeifer, Aalpen A. Patel, Brian P. Delisle, Amro Alsaid, Dominik BeerChristopher M. Haggerty, Brandon K. Fornwalt

Physiology

Research output: Contribution to journal › Article › peer-review

90 Scopus citations

Abstract

The electrocardiogram (ECG) is a widely used medical test, consisting of voltage versus time traces collected from surface recordings over the heart¹. Here we hypothesized that a deep neural network (DNN) can predict an important future clinical event, 1-year all-cause mortality, from ECG voltage–time traces. By using ECGs collected over a 34-year period in a large regional health system, we trained a DNN with 1,169,662 12-lead resting ECGs obtained from 253,397 patients, in which 99,371 events occurred. The model achieved an area under the curve (AUC) of 0.88 on a held-out test set of 168,914 patients, in which 14,207 events occurred. Even within the large subset of patients (n = 45,285) with ECGs interpreted as ‘normal’ by a physician, the performance of the model in predicting 1-year mortality remained high (AUC = 0.85). A blinded survey of cardiologists demonstrated that many of the discriminating features of these normal ECGs were not apparent to expert reviewers. Finally, a Cox proportional-hazard model revealed a hazard ratio of 9.5 (P < 0.005) for the two predicted groups (dead versus alive 1 year after ECG) over a 25-year follow-up period. These results show that deep learning can add substantial prognostic information to the interpretation of 12-lead resting ECGs, even in cases that are interpreted as normal by physicians.

Original language	English
Pages (from-to)	886-891
Number of pages	6
Journal	Nature Medicine
Volume	26
Issue number	6
DOIs	https://doi.org/10.1038/s41591-020-0870-z
State	Published - Jun 1 2020

Bibliographical note

Funding Information:
This work was supported in part by funding from the Pennsylvania Department of Health (SAP 4100070267), an American Heart Association Competitive Catalyst Award (17CCRG33700289), the Geisinger Health Plan and Clinic, and Tempus. The content of this article does not reflect the views of the funding sources. Geisinger receives funding from Tempus for ongoing development of predictive modeling technology and commercialization. Tempus and Geisinger have jointly applied for a patent related to the work. None of the Geisinger authors has ownership interest in any of the intellectual property resulting from the partnership. A.N., T.C., A.H., M.C.S. and K.W.J. are employees of Tempus.

Publisher Copyright:
© 2020, The Author(s), under exclusive licence to Springer Nature America, Inc.

ASJC Scopus subject areas

Biochemistry, Genetics and Molecular Biology (all)

Access to Document

10.1038/s41591-020-0870-z

Cite this

Raghunath, S., Ulloa Cerna, A. E., Jing, L., vanMaanen, D. P., Stough, J., Hartzel, D. N., Leader, J. B., Kirchner, H. L., Stumpe, M. C., Hafez, A., Nemani, A., Carbonati, T., Johnson, K. W., Young, K., Good, C. W., Pfeifer, J. M., Patel, A. A., Delisle, B. P., Alsaid, A., ... Fornwalt, B. K. (2020). Prediction of mortality from 12-lead electrocardiogram voltage data using a deep neural network. Nature Medicine, 26(6), 886-891. https://doi.org/10.1038/s41591-020-0870-z

@article{e9bed0dc794c42beaf69b8e1a1e457c0,

title = "Prediction of mortality from 12-lead electrocardiogram voltage data using a deep neural network",

abstract = "The electrocardiogram (ECG) is a widely used medical test, consisting of voltage versus time traces collected from surface recordings over the heart1. Here we hypothesized that a deep neural network (DNN) can predict an important future clinical event, 1-year all-cause mortality, from ECG voltage–time traces. By using ECGs collected over a 34-year period in a large regional health system, we trained a DNN with 1,169,662 12-lead resting ECGs obtained from 253,397 patients, in which 99,371 events occurred. The model achieved an area under the curve (AUC) of 0.88 on a held-out test set of 168,914 patients, in which 14,207 events occurred. Even within the large subset of patients (n = 45,285) with ECGs interpreted as {\textquoteleft}normal{\textquoteright} by a physician, the performance of the model in predicting 1-year mortality remained high (AUC = 0.85). A blinded survey of cardiologists demonstrated that many of the discriminating features of these normal ECGs were not apparent to expert reviewers. Finally, a Cox proportional-hazard model revealed a hazard ratio of 9.5 (P < 0.005) for the two predicted groups (dead versus alive 1 year after ECG) over a 25-year follow-up period. These results show that deep learning can add substantial prognostic information to the interpretation of 12-lead resting ECGs, even in cases that are interpreted as normal by physicians.",

author = "Sushravya Raghunath and {Ulloa Cerna}, {Alvaro E.} and Linyuan Jing and vanMaanen, {David P.} and Joshua Stough and Hartzel, {Dustin N.} and Leader, {Joseph B.} and Kirchner, {H. Lester} and Stumpe, {Martin C.} and Ashraf Hafez and Arun Nemani and Tanner Carbonati and Johnson, {Kipp W.} and Katelyn Young and Good, {Christopher W.} and Pfeifer, {John M.} and Patel, {Aalpen A.} and Delisle, {Brian P.} and Amro Alsaid and Dominik Beer and Haggerty, {Christopher M.} and Fornwalt, {Brandon K.}",

note = "Funding Information: This work was supported in part by funding from the Pennsylvania Department of Health (SAP 4100070267), an American Heart Association Competitive Catalyst Award (17CCRG33700289), the Geisinger Health Plan and Clinic, and Tempus. The content of this article does not reflect the views of the funding sources. Geisinger receives funding from Tempus for ongoing development of predictive modeling technology and commercialization. Tempus and Geisinger have jointly applied for a patent related to the work. None of the Geisinger authors has ownership interest in any of the intellectual property resulting from the partnership. A.N., T.C., A.H., M.C.S. and K.W.J. are employees of Tempus. Publisher Copyright: {\textcopyright} 2020, The Author(s), under exclusive licence to Springer Nature America, Inc.",

year = "2020",

month = jun,

day = "1",

doi = "10.1038/s41591-020-0870-z",

language = "English",

volume = "26",

pages = "886--891",

number = "6",

}

Raghunath, S, Ulloa Cerna, AE, Jing, L, vanMaanen, DP, Stough, J, Hartzel, DN, Leader, JB, Kirchner, HL, Stumpe, MC, Hafez, A, Nemani, A, Carbonati, T, Johnson, KW, Young, K, Good, CW, Pfeifer, JM, Patel, AA, Delisle, BP, Alsaid, A, Beer, D, Haggerty, CM & Fornwalt, BK 2020, 'Prediction of mortality from 12-lead electrocardiogram voltage data using a deep neural network', Nature Medicine, vol. 26, no. 6, pp. 886-891. https://doi.org/10.1038/s41591-020-0870-z

TY - JOUR

T1 - Prediction of mortality from 12-lead electrocardiogram voltage data using a deep neural network

AU - Raghunath, Sushravya

AU - Ulloa Cerna, Alvaro E.

AU - Jing, Linyuan

AU - vanMaanen, David P.

AU - Stough, Joshua

AU - Hartzel, Dustin N.

AU - Leader, Joseph B.

AU - Kirchner, H. Lester

AU - Stumpe, Martin C.

AU - Hafez, Ashraf

AU - Nemani, Arun

AU - Carbonati, Tanner

AU - Johnson, Kipp W.

AU - Young, Katelyn

AU - Good, Christopher W.

AU - Pfeifer, John M.

AU - Patel, Aalpen A.

AU - Delisle, Brian P.

AU - Alsaid, Amro

AU - Beer, Dominik

AU - Haggerty, Christopher M.

AU - Fornwalt, Brandon K.

N1 - Funding Information: This work was supported in part by funding from the Pennsylvania Department of Health (SAP 4100070267), an American Heart Association Competitive Catalyst Award (17CCRG33700289), the Geisinger Health Plan and Clinic, and Tempus. The content of this article does not reflect the views of the funding sources. Geisinger receives funding from Tempus for ongoing development of predictive modeling technology and commercialization. Tempus and Geisinger have jointly applied for a patent related to the work. None of the Geisinger authors has ownership interest in any of the intellectual property resulting from the partnership. A.N., T.C., A.H., M.C.S. and K.W.J. are employees of Tempus. Publisher Copyright: © 2020, The Author(s), under exclusive licence to Springer Nature America, Inc.

PY - 2020/6/1

Y1 - 2020/6/1

N2 - The electrocardiogram (ECG) is a widely used medical test, consisting of voltage versus time traces collected from surface recordings over the heart1. Here we hypothesized that a deep neural network (DNN) can predict an important future clinical event, 1-year all-cause mortality, from ECG voltage–time traces. By using ECGs collected over a 34-year period in a large regional health system, we trained a DNN with 1,169,662 12-lead resting ECGs obtained from 253,397 patients, in which 99,371 events occurred. The model achieved an area under the curve (AUC) of 0.88 on a held-out test set of 168,914 patients, in which 14,207 events occurred. Even within the large subset of patients (n = 45,285) with ECGs interpreted as ‘normal’ by a physician, the performance of the model in predicting 1-year mortality remained high (AUC = 0.85). A blinded survey of cardiologists demonstrated that many of the discriminating features of these normal ECGs were not apparent to expert reviewers. Finally, a Cox proportional-hazard model revealed a hazard ratio of 9.5 (P < 0.005) for the two predicted groups (dead versus alive 1 year after ECG) over a 25-year follow-up period. These results show that deep learning can add substantial prognostic information to the interpretation of 12-lead resting ECGs, even in cases that are interpreted as normal by physicians.

AB - The electrocardiogram (ECG) is a widely used medical test, consisting of voltage versus time traces collected from surface recordings over the heart1. Here we hypothesized that a deep neural network (DNN) can predict an important future clinical event, 1-year all-cause mortality, from ECG voltage–time traces. By using ECGs collected over a 34-year period in a large regional health system, we trained a DNN with 1,169,662 12-lead resting ECGs obtained from 253,397 patients, in which 99,371 events occurred. The model achieved an area under the curve (AUC) of 0.88 on a held-out test set of 168,914 patients, in which 14,207 events occurred. Even within the large subset of patients (n = 45,285) with ECGs interpreted as ‘normal’ by a physician, the performance of the model in predicting 1-year mortality remained high (AUC = 0.85). A blinded survey of cardiologists demonstrated that many of the discriminating features of these normal ECGs were not apparent to expert reviewers. Finally, a Cox proportional-hazard model revealed a hazard ratio of 9.5 (P < 0.005) for the two predicted groups (dead versus alive 1 year after ECG) over a 25-year follow-up period. These results show that deep learning can add substantial prognostic information to the interpretation of 12-lead resting ECGs, even in cases that are interpreted as normal by physicians.

UR - http://www.scopus.com/inward/record.url?scp=85084483732&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85084483732&partnerID=8YFLogxK

U2 - 10.1038/s41591-020-0870-z

DO - 10.1038/s41591-020-0870-z

M3 - Article

C2 - 32393799

AN - SCOPUS:85084483732

VL - 26

SP - 886

EP - 891

IS - 6

ER -

Prediction of mortality from 12-lead electrocardiogram voltage data using a deep neural network

Abstract

Bibliographical note

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this