CoChat
Sign Up
Cited by 2ECG Monitoring and AnalysisArtificial Intelligence in HealthcareMachine Learning in Healthcare

Artificial Intelligence-Enabled ECG Analysis to Predict Incident Heart Failure

Open access
Khurshid, Shaan, Friedman, Samuel F., Kany, Shinwan, D’Souza, Valentina, Roshandelpoor, Athar, Kosyakovsky, Leah B., Chitsazan, Mandana, Cunningham, Jonathan W. +13 more·2026·Circulation: Heart Failure
DOI 10.1161/circheartfailure.125.013927PMC PMC12997051
Open this paper in CoChat — free account

Abstract

BACKGROUND: ECG-based artificial intelligence may enable efficient prediction of incident heart failure (HF) risk to facilitate preventive efforts. Prior models are proprietary, with modest or inconsistent accuracy. We sought to develop and validate a generalizable and publicly available convolutional neural network to predict incident HF using the 12-lead ECG waveform (ECG-to-HF [ECG2HF]). METHODS: We developed ECG2HF in 94 636 patients receiving longitudinal ambulatory care at Massachusetts General Hospital (MGH), and validated it in 3 test sets: MGH, Brigham and Women’s Hospital (BWH), and Beth Israel Deaconess Medical Center (BIDMC), among 93 868 individuals aged 30 to 79 years without HF. HF events at 10 years were identified using a validated electronic health record-based natural language processing model. Discrimination was quantified using the area under the receiver operating characteristic curve. We then compared discrimination and net reclassification (at <10%, 10% to 20%, ≥20% 10-year risk categories) using ECG2HF versus the 15-component Pooled Cohorts Equations to Prevent HF score. RESULTS: The test sets comprised MGH (13 954 individuals, 441 events, age 57±13 years, 48% women), BWH (54 396 individuals, 1809 events, age 57±13 years, 55% women), and BIDMC (25 457 individuals, 901 events, age 57±13 years, 53% women). Over 10 years, the cumulative risk of HF was 4.6% (95% CI, 4.1–5.0) in MGH, 5.0% (4.8–5.2) in BWH, and 4.4% (4.1–4.7) in BIDMC. ECG2HF discriminated 10-year incident HF in each test set (area under the receiver operating characteristic curve: MGH 0.86 [0.84–0.87]; BWH 0.85 [0.84–0.86]; BIDMC 0.84 [0.83–0.86]). Compared with the Pooled Cohorts Equations to Prevent HF, ECG2HF provided favorable discrimination (improvement in area under the receiver operating characteristic curve MGH/BWH 0.061 [0.025–0.097]; BIDMC 0.038 [−0.0096 to 0.086]) and net reclassification (NRI MGH/BWH 0.16 [0.077–0.24]; BIDMC 0.23 [0.10–0.35]) of 10-year HF risk. CONCLUSIONS: ECG2HF is a publicly available 12-lead ECG-based artificial intelligence model that discriminates the risk of future HF with favorable and consistent performance across 3 large health care samples from the northeastern United States. ECG2HF may enable efficient prioritization of high-risk individuals for HF-related preventive measures.

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1161/circheartfailure.125.013927

Cites

38 of 38 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • At present, AI-ECG is better interpreted as a tool for risk enrichment and follow-up prioritization in selected screening settings than as a stand-alone mandate for structured surveillance or immediate pathway escalation.

    “At present, AI-ECG is better interpreted as a tool for risk enrichment and follow-up prioritization in selected screening settings than as a stand-alone mandate for structured surveillance or immediate pathway escalation.12,19,44,45”

    From AI-ECG for Echocardiography Triage in Structural Heart Disease: Evidence, Implementation, and Future Directions (2026) · cites this paper as [45]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Cites

Flagged first, then by citations

Research works printed but not linked: 4

  1. [1]Disease and Injury Incidence and Prevalence Collaborators. Global, regional, and national incidence, prevalence, and years lived with disability for 354 diseases and injuries for 195 countries and territories. Lancet. 10; 392: 1789– 1858. PMC6227754 30496104
  2. [34]Snoek J, Larochelle H, Adams RP. Practical Bayesian optimization of machine learning algorithms [Internet]. arXiv [stat.ML]. 2012;Available from: http://arxiv.org/abs/1206.2944 · arXiv
  3. [36]Liu L, Jiang H, He P, Chen W, Liu X, Gao J, Han J. On the variance of the adaptive learning rate and beyond [Internet]. arXiv [cs.LG]. 2019;Available from: http://arxiv.org/abs/1908.03265 · arXiv
  4. [37]Luo J, Wang Y, Xu R, Liu G, Wang X, Gong Y. Channel drop out: A simple way to prevent CNN from overfitting in motor imagery based BCI. In: Communications in Computer and Information Science. Singapore: Springer Nature Singapore; 2021. p. 443– 452.

Cited by

Sorted by citations

Related

Authors
Khurshid, S., Friedman, S. F., Kany, S., D’Souza, V., Roshandelpoor, A., Kosyakovsky, L. B., Chitsazan, M., Cunningham, J. W., Singh, P., Lau, E. S., Pipilas, D., Al-Alusi, M. A., Rämö, J. T., Pirruccello, J. P., Reeder, C., Waks, J. W., Lubitz, S. A., Philippakis, A. A., Maddah, M., Ellinor, P. T., & Ho, J. E.
Affiliations
  • Broad Institute (US)
  • Data Sciences Platform, Broad Institute of MIT and Harvard, Cambridge, MA (S.F.F., C.R., A.A.P., M.M.).
  • Telemachus and Irene Demoulas Family Foundation Center for Cardiac Arrhythmias, Heart and Vascular Institute, Mass General Brigham, Boston (S. Khurshid, V.D.S., P.S., D.P., M.A.A.-A., S.A.L., P.T.E.).
  • Beth Israel Deaconess Medical Center (US)
  • Mass General Brigham (US)
  • Cardiovascular Disease Initiative, Broad Institute of MIT and Harvard, Cambridge, MA (S. Khurshid, S. Kany, J.W.C., E.S.L., D.P., M.A.A.-A., J.T.R., S.A.L., P.T.E., J.E.H.).
  • University of California, San Francisco (US)
Corresponding author
  • Khurshid, Shaan
  • Kany, Shinwan
  • Ellinor, Patrick T.
  • Ho, Jennifer E.
ORCID
Venue
Circulation: Heart Failure
Publisher
Ovid Technologies (Wolters Kluwer Health)
Published
2026-04
Volume
19
Issue
4
Pages
e013927-e013927
ISSN
  • 1941-3289 (ISSN-L)
  • 1941-3297
DOI
10.1161/circheartfailure.125.013927
PMC
PMC12997051
OpenAlex
W7131126599
Cited by
2
References
38
Citations by year
2026: 2
FWCI
9.99
Type
journal article
Language
en
Open access
Locations
  • Circulation Heart Failure (publishedVersion)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion)
Funders
National Heart, Lung, and Blood Institute (R01HL140224, R01HL134893, K24HL105780, K24HL153669, K08HL159346, K23HL169839, R01HL168889, R01HL092577, R01HL160003, T32HL160522, T32HL007208, K23HL168163, K23HL159243)
Topics
  • Health Sciences › Medicine › Cardiology and Cardiovascular Medicine › ECG Monitoring and Analysis
  • Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
  • Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
Keywords
heart failure prediction, 12-lead ECG, convolutional neural networks, incident heart failure risk, net reclassification improvement, model generalization, cardiovascular risk prediction, outpatients, 10-year risk
MeSH terms
Intelligent Systems, Convolutional Neural Networks, Prediction Algorithms, Adult, Aged, Artificial Intelligence, Electrocardiography, Female, Heart Failure, Humans, Male, Middle Aged, Predictive Value of Tests, Risk Factors, Reproducibility of Results, Incidence, Neural Networks, Computer, Risk Assessment, Electronic Health Records

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review