Pith. sign in

REVIEW 3 major objections 5 minor 53 references

A single 24-hour ECG recording can predict a patient's risk of being diagnosed with heart failure within five years, according to a deep learning model that reads the full day-long signal.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-03 14:57 UTC pith:WNBUFXAM

load-bearing objection A strong first demonstration of 24-hour Holter ECG for HF risk, but the incident-HF label likely captures some prevalent disease—the paper admits as much—so the 0.80 AUROC should be read as partly detection, not pure prediction. the 3 major comments →

arxiv 2601.00014 v2 pith:WNBUFXAM submitted 2025-12-20 eess.SP cs.AIcs.LG

Modeling Day-Long ECG Signals to Predict Heart Failure Risk with Explainable AI

Eran Zvuloni , Ronit Almog , Michael Glikson , Shany Brimer Biton , Ilan Green , Izhar Laufer , Offer Amir , Joachim A. Behar This is my paper
classification eess.SP cs.AIcs.LG
keywords heart failure prediction24-hour Holter ECGdeep learningsingle-lead ECGrisk stratificationexplainabilitycircadian variationincident heart failure
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper claims that a deep learning model processing an entire 24-hour single-lead Holter ECG can predict whether a patient will be diagnosed with heart failure within five years, achieving an AUROC of 0.80. The day-long recording matters: a model restricted to 30-second clips scored 0.77, and the authors argue the gap shows that paroxysmal arrhythmias and circadian patterns carry predictive information. If correct, routine Holter exams done for other indications could opportunistically flag high-risk patients for preventive follow-up, such as BNP testing or echocardiography, without extra cost or invasive procedures. The model also stratifies prognosis: patients flagged as high-risk have fourfold higher odds of death and twofold higher odds of hospitalization or death than lower-risk patients.

Core claim

DeepHHF, trained and validated on the Technion-Leumit Holter ECG dataset (57,575 recordings from 40,174 patients), learns to predict incident heart failure within five years from raw 24-hour single-lead ECG. The full-recording model reaches AUROC 0.80 on the held-out test set, outperforming both a 30-second-window encoder (0.77) and the PCP-HF clinical score (0.74), and achieves AUROC 0.81 in a zero-shot external cohort. Explainability via gradient attention rollout shows the model concentrates on daytime hours and on ectopic beats—premature ventricular contractions and supraventricular ectopy—consistent with known arrhythmia burdens preceding heart failure. The authors conclude that day-lon

What carries the argument

A two-stage deep learning architecture: first, a convolutional encoder (based on EnCodec blocks) is trained on randomly sampled 30-second windows to extract compact latent features; second, the frozen encoder converts 720 fixed-interval windows spanning the full 24-hour recording into a sequence, which a transformer sequential head integrates into a single HF risk score. Gradient attention rollout then traces which parts of the recording drove the prediction. This machinery enables the model to use the entire day-long signal rather than a short snapshot, and the transformer's sequential integration is what the authors credit for capturing circadian and paroxysmal information.

Load-bearing premise

The load-bearing assumption is that the first documented ICD-9 heart failure code in the electronic medical record within five years is a valid and timely proxy for incident heart failure; if many patients already have undocumented or prevalent heart failure at the time of the Holter recording, the model may be detecting existing disease rather than predicting future onset.

What would settle it

Apply DeepHHF to a cohort in which heart failure status at the time of the Holter is independently adjudicated by symptoms, BNP level, and echocardiography. If the AUROC stays near 0.80 among patients confirmed free of heart failure at baseline, prediction is validated; if performance mostly disappears when patients with baseline HF medications or echo evidence are excluded, the result reflects detection bias. Additionally, compute AUROC only for recordings 4–5 years before the documented diagnosis: if it falls to near 0.5, the five-year-horizon claim collapses.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Routine 24-hour Holter recordings, ordered for palpitations or syncope, can be repurposed to estimate five-year heart failure risk without additional tests.
  • DeepHHF outperforms the guideline-recommended PCP-HF clinical score on the test set, and adding clinical variables pushes AUROC to 0.82.
  • High-risk patients identified by the model have fourfold odds of all-cause mortality and twofold odds of hospitalization or death versus low/moderate-risk patients.
  • Zero-shot performance on an external cohort (AUROC 0.81) suggests the model transfers beyond the training health system.
  • Using DeepHHF to select patients for preventive interventions reduces the number needed to screen to prevent one major cardiovascular hospitalization, from 61 in the overall test set to 21 in the high-risk subgroup.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A testable extension of this work is whether a compact summary of a Holter recording (for example, hourly ectopic beat counts and heart rate variability) can approach the same performance as the full signal, which would make the approach feasible on wearable patches with limited storage.
  • If undiagnosed prevalent heart failure partly drives the 0.80 AUROC, then a truly asymptomatic screening population would likely show lower performance; the clinical value depends on whether the model detects subclinical disease before symptoms, which this retrospective design cannot fully separate.
  • The daytime attention peak (8 AM–3 PM) could be a circadian signature; a direct experiment would be to retrain the model on nighttime-only segments and measure the AUROC drop, isolating the timing contribution from overall arrhythmia burden.
  • Because the endpoint relies on ICD-9 codes, linking the model to echocardiogram-confirmed heart failure with ejection fraction subtypes would test whether DeepHHF generalizes across HFpEF and HFrEF, which the authors aim to cover.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper presents DeepHHF, a deep-learning model that takes a full 24-hour single-lead Holter ECG recording as input and outputs a five-year heart-failure risk score. The model is trained and evaluated on the Technion-Leumit Holter ECG (TLHE) dataset, a large retrospective Israeli HMO cohort, with a temporally held-out test set (January–April 2018), patient-level split, and an external zero-shot cohort from Rambam. The authors report AUROC 0.80 on the internal test set, outperforming a 30-second-window encoder (0.77) and the PCP-HF clinical score (0.74), with AUROC 0.81 on the external cohort. Additional analyses include subgroup performance by time-to-diagnosis, Kaplan-Meier survival curves, risk-group stratification, and gradient-attention-rollout explainability. The paper's central claim is that five-year incident heart-failure risk is learnable from day-long single-lead ECG, with the full 24-hour recording adding value over short segments.

Significance. If the central claim holds, this is a substantial contribution: it is, to my knowledge, the first demonstration that raw 24-hour single-lead Holter ECG predicts incident heart failure at five years, and it is supported by several strong methodological components — a large real-world dataset, a temporally held-out test set with complete follow-up, patient-level stratification, bootstrapped confidence intervals, zero-shot external validation, and extensive label-verification analyses. The publicly released model and reproducible architecture details are also strengths. However, the manuscript has three load-bearing concerns that must be addressed before the claim can be accepted at face value: the endpoint may include prevalent but undocumented heart failure, the PCP-HF comparison is not on the same test subset, and the training-set labeling may include censored non-HF examples. These are fixable with additional analyses or re-analysis, and the underlying dataset and model are sufficiently valuable that the paper merits major revision rather than rejection.

major comments (3)
  1. [Methods — Class definition; Discussion paragraph] The paper's own Discussion concedes that 'undiagnosed prevalent HF cases with active symptoms and treatments at the time of Holter recording could influence the ECG data and potentially inflate model performance.' This is load-bearing for the claim of incident-risk prediction. The exclusion criteria only remove patients with a pre-Holter documented HF diagnosis; label-verification analyses (prescriptions, echocardiography) establish that documented diagnoses are real but cannot establish absence of HF at Holter time. The argument that AUROC remains consistent within 0–2 years does not rule out inflation, because diagnoses delayed by up to two years would still fall in that interval. Please provide a sensitivity analysis that excludes patients with HF medications, loop diuretics, or abnormal echocardiography before the Holter, or otherwise demonstrate that the 0.80 AUROC is not substantia
  2. [Figure 3b; Methods — PCP-HF score computation] The PCP-HF AUROC is computed on the 1,917 test-set examinations for which all PCP-HF covariates were available, while the DeepHHF AUROC is computed on the full 4,461-examination test set. These are different patient subsets, so the reported significant difference (p<0.05) is confounded by case mix. The manuscript should either re-evaluate DeepHHF on the same 1,917-exam subset or evaluate PCP-HF on the full test set using imputation (e.g., multiple imputation or mean/median imputation with sensitivity analysis). Without this, the claim that DeepHHF outperforms PCP-HF is not supported by the displayed analysis.
  3. [Methods — Dataset split and preprocessing; Results — Study cohort] The text states that the test set (January–April 2018) ensures complete 5-year follow-up, but the training and validation sets appear to include Holter recordings from 2010 through June 2023, while EMR data end in April 2024. If any training/validation recording was made after April 2019, a 5-year follow-up window is not available, and labeling such recordings as non-HF because no diagnosis appears before the data cutoff creates censored negative labels. Please clarify whether training/validation recordings were restricted to those with at least five years of observable follow-up; if not, the model training used systematically mislabeled negatives, and the analysis should be repeated with a training set restricted to complete observations or with a time-to-event formulation.
minor comments (5)
  1. [Keywords; Discussion] The keyword 'circardian' appears to be a typo for 'circadian.'
  2. [Discussion, paragraph 8] The text refers to 'Figure 1a' when discussing subgroup performance by time-to-diagnosis; the correct reference appears to be Figure 4a. Please check all cross-references.
  3. [Performance measures and statistical analysis] The 95% CI for AUROC is obtained by bootstrapping 250 positive and 250 negative examples per iteration, which is not the usual full test-set bootstrap. Please report CIs at the full test-set level and, if appropriate, use DeLong's test or paired bootstrap for classifier comparison.
  4. [External Validation] The external cohort AUROC of 0.81 is reported without a confidence interval. Given only 29 positive cases, the CI is likely wide; please include it or at least an exact binomial CI.
  5. [Methods — Deep learning model] The sentence 'The optimization attempted to maximize the validation set AUROC score' is clear, but the hyperparameter list is long and the final selected values are not reported. Providing the final hyperparameters in a table or Extended Data would aid reproducibility.

Circularity Check

0 steps flagged

No significant circularity: supervised prediction evaluated on a disjoint held-out test set and an external zero-shot cohort; the admitted label-timing issue is a validity limitation, not an input-output identity.

full rationale

The derivation chain is a standard supervised learning pipeline, not a circular one. The endpoint (first documented HF ICD-9 code within five years) is defined from the EMR and is independent of the ECG input; the model is fitted to training labels, tuned on a validation set, and evaluated on a disjoint test set fixed as the first four months of 2018, with patient-specific stratification to prevent leakage (Methods: Dataset split and preprocessing). The AUROC 0.80 is therefore a held-out accuracy measure, and the external Rambam cohort (150 Holters, zero-shot) provides an independent check. The PCP-HF comparison (AUROC 0.74) and the 30-second-encoder comparison (0.77) are computed on the same test set, so the 0.80 claim does not reduce to a fit. The paper's own Discussion transparently flags the main threat: 'undiagnosed prevalent HF cases with active symptoms and treatments at the time of Holter recording could influence the ECG data and potentially inflate model performance.' This is a label-timing validity concern (detection vs. prediction), not the kind of definitional identity or fitted-input-renamed-as-prediction that constitutes circularity; the model output is not defined in terms of the EMR label, and the held-out/external evaluation protocol is standard non-circular practice. Minor self-citations (refs 9, 48, 49: AF-risk context and PhysioZoo beat detection) are instrumental and non-load-bearing, and the encoder is adopted from the external EnCodec architecture with stated assumptions. No uniqueness theorem is imported, and no Eq. reduces to its own input by construction. Per the hard rules, no circular step can be exhibited, so the appropriate score is low.

Axiom & Free-Parameter Ledger

2 free parameters · 5 axioms · 0 invented entities

The central claim rests primarily on EMR-label validity and on the assumption that the model is predicting future disease rather than detecting undocumented prevalent disease. No new physical entities or scientific constants are introduced; the free parameters are standard ML hyperparameters and decision thresholds.

free parameters (2)
  • Model hyperparameters (sequence elements, learning rates, dropout, number of filters) = e.g., sequence length 720, learning rates 1e-3 and 5e-5, tuned via Optuna on validation AUROC
    These are standard ML hyperparameters selected by Bayesian search on the validation set. They affect the reported metrics but are not derived from first principles.
  • Risk-group probability thresholds = specificity 70% (moderate risk) and 90% (high risk)
    Chosen by the authors to define low/moderate/high-risk groups. Threshold choices affect odds-ratio and Kaplan-Meier analyses, though not the primary AUROC.
axioms (5)
  • domain assumption First documented ICD-9 HF code in the EMR within five years is a valid incident heart failure endpoint.
    Used in Methods 'Class definition'. The paper performs label-verification analyses (medication prescriptions, echocardiograms) to support this, but it remains an administrative diagnosis with possible delays and misclassification.
  • domain assumption Recordings with no HF code within five years are truly non-HF.
    Methods 'Class definition'. Undiagnosed or undocumented prevalent HF would contaminate the negative class and could inflate AUROC; the authors acknowledge this risk in the Discussion.
  • domain assumption Holter referral for clinical indications (syncope, palpitations, known arrhythmia) does not invalidate the predictive relationship.
    Methods 'Technion-Leumit Holter ECG dataset elaboration'; Discussion 'inherent selection bias'. The model is developed on a referred population, so generalizability to the general population is an assumption.
  • standard math Bootstrapped t-tests, logrank tests, k-means, and silhouette analysis are appropriate for the stated inferences.
    Used in 'Performance measures and statistical analysis' and 'Model explainability analysis'. Standard tools, but the t-test between bootstrapped AUROC distributions is a heuristic comparison.
  • domain assumption The external Rambam cohort’s 3-year follow-up and different event definition are comparable to the internal 5-year endpoint.
    Methods 'Rambam Holter databank cohort for external validation'. The external cohort uses 150 patients and 29 HF events with follow-up through November 2021; the shorter horizon may undercount events.

pith-pipeline@v1.3.0-alltime-deepseek · 21672 in / 10569 out tokens · 106275 ms · 2026-08-03T14:57:18.207525+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Modeling Day-Long ECG Signals to Predict Heart Failure Risk with Explainable AI." pith.science (2026). https://pith.science/paper/WNBUFXAM

@misc{pith2026260100014,
  author       = {Pith},
  title        = {Pith review of: Modeling Day-Long ECG Signals to Predict Heart Failure Risk with Explainable AI},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WNBUFXAM}},
  note         = {Machine review of arXiv:2601.00014}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Heart failure (HF) affects 11.8% of adults aged 65 and older, reducing quality of life and longevity. Preventing HF can reduce morbidity and mortality. We hypothesized that artificial intelligence (AI) applied to 24-hour single-lead electrocardiogram (ECG) data could predict the risk of HF within five years. To research this, the Technion-Leumit Holter ECG (TLHE) dataset, including 69,663 recordings from 47,729 patients, collected over 20 years was used. Our deep learning model, DeepHHF, trained on 24-hour ECG recordings, achieved an area under the receiver operating characteristic curve of 0.80 that outperformed a model using 30-second segments and a clinical score. High-risk individuals identified by DeepHHF had a two-fold chance of hospitalization or death incidents. Explainability analysis showed DeepHHF focused on arrhythmias and heart abnormalities. This study highlights the feasibility of deep learning to model 24-hour continuous ECG data, capturing paroxysmal events essential for reliable risk prediction. Artificial intelligence applied to single-lead Holter ECG is non-invasive, inexpensive, and widely accessible, making it a promising tool for HF risk prediction.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

53 extracted references · 26 canonical work pages · 3 internal anchors

  1. [1]

    European Journal of Heart Failure22(8), 1342–1356 (2020) https://doi.org/10.1002/ejhf.1858

    Groenewegen, A., Rutten, F.H., Mosterd, A., Hoes, A.W.: Epidemiology of heart failure. European Journal of Heart Failure22(8), 1342–1356 (2020) https://doi.org/10.1002/ejhf.1858

    doi:10.1002/ejhf.1858 2020

  2. [2]

    The Lancet393(10175), 1034–1044 (2019) https://doi.org/10.1016/S0140-6736(18)31808-7

    Rossignol, P., Hernandez, A.F., Solomon, S.D., Zannad, F.: Heart failure drug treatment. The Lancet393(10175), 1034–1044 (2019) https://doi.org/10.1016/S0140-6736(18)31808-7

    work page doi:10.1016/s0140-6736(18)31808-7 2019

  3. [3]

    Circulation145(18), 895–1032 (2022) https://doi.org/10.1161/CIR.0000000000001063

    Heidenreich, P.A., Bozkurt, B., Aguilar, D., Allen, L.A., Byun, J.J., Colvin, M.M., Deswal, A., Drazner, M.H., Dunlay, S.M., Evers, L.R., Fang, J.C., Fedson, S.E., Fonarow, G.C., Hayek, S.S., Hernandez, A.F., Khazanie, P., Kittleson, M.M., Lee, C.S., Link, M.S., Milano, C.A., Nnacheta, L.C., Sandhu, A.T., Stevenson, L.W., Vardeny, O., Vest, A.R., Yancy, C...

    2022

  4. [4]

    Heart & Lung31(4), 262–270 (2002) https://doi.org/10.1067/mhl.2002

    Bennett, J.A., Riegel, B., Bittner, V., Nichols, J.: Validity and reliability of the NYHA classes for measuring research outcomes in patients with cardiac disease. Heart & Lung31(4), 262–270 (2002) https://doi.org/10.1067/mhl.2002. 124554

    work page doi:10.1067/mhl.2002 2002

  5. [5]

    Journal of Cardiac Failure27(4), 387–413 (2021) https://doi.org/10.1016/J.CARDF AIL.2021.01.022

    Bozkurt, B., Coats, A.J., Tsutsui, H., Abdelhamid, M., Adamopoulos, S., Albert, N., Anker, S.D., Atherton, J., B¨ ohm, M., Butler, J., Drazner, M.H., Felker, G.M., Filippatos, G., Fonarow, G.C., Fiuzat, M., Gomez-Mesa, J.E., Heidenreich, P., Imamura, T., Januzzi, J., Jankowska, E.A., Khazanie, P., Kinugawa, K., Lam, C.S.P., Matsue, Y., Metra, M., Ohtani, ...

    2021

  6. [6]

    Circulation135(10), 146–603 (2017) https://doi.org/10.1161/CIR.0000000000000485/ ASSET/065F105B-352E-48D3-AC2B-2F0A69AE4BE0/ASSETS/CIR.0000000000000485.FP.PNG

    Benjamin, E.J., Blaha, M.J., Chiuve, S.E., Cushman, M., Das, S.R., Deo, R., De Ferranti, S.D., Floyd, J., Fornage, M., Gillespie, C., Isasi, C.R., Jim’nez, M.C., Jordan, L.C., Judd, S.E., Lackland, D., Lichtman, J.H., Lisabeth, L., Liu, S., Longenecker, C.T., MacKey, R.H., Matsushita, K., Mozaffarian, D., Mussolino, M.E., Nasir, K., Neumar, R.W., Palaniap...

    2017

  7. [7]

    European Heart Journal44(Supplement 2), 655–892 (2023) https://doi.org/10.1093/eurheartj/ehad655.892

    Wang, H., Gao, C., Guignard-Duff, M., Cole, C., Hall, C., Larman, M., Baruah, R., Gao, H., Mamza, J.B., Lang, C.C., Mordi, I.: Importance of early diagnosis and treatment of heart failure across the spectrum of ejection fraction. European Heart Journal44(Supplement 2), 655–892 (2023) https://doi.org/10.1093/eurheartj/ehad655.892

    work page doi:10.1093/eurheartj/ehad655.892 2023

  8. [8]

    Nature Communications11(1), 1760–1760 (2020) https://doi.org/10.1038/ s41467-020-15432-4

    Ribeiro, A.H., Ribeiro, M.H., Paix˜ ao, G.M.M., Oliveira, D.M., Gomes, P.R., Canazart, J.A., Ferreira, M.P.S., Andersson, C.R., Macfarlane, P.W., Meira, W., Sch¨ on, T.B., Ribeiro, A.L.P.: Automatic diagnosis of the 12-lead ECG using a deep neural network. Nature Communications11(1), 1760–1760 (2020) https://doi.org/10.1038/ s41467-020-15432-4

    2020

  9. [9]

    European Heart Journal - Digital Health2(4), 576–585 (2021) https://doi.org/10.1093/EHJDH/ZTAB071 14

    Biton, S., Gendelman, S., Ribeiro, A.H., Miana, G., Moreira, C., Ribeiro, A.L.P., Behar, J.A.: Atrial fibrillation risk prediction from the 12-lead electrocardiogram using digital biomarkers and deep representation learning. European Heart Journal - Digital Health2(4), 576–585 (2021) https://doi.org/10.1093/EHJDH/ZTAB071 14

    work page doi:10.1093/ehjdh/ztab071 2021

  10. [10]

    Nature Medicine29, 1804–1813 (2023) https://doi.org/10.1038/s41591-023-02396-3

    Al-Zaiti, S.S., Martin-Gill, C., Z` egre-Hemsey, J.K., Bouzid, Z., Faramand, Z., Alrawashdeh, M.O., Gregg, R.E., Helman, S., Riek, N.T., Kraevsky-Phillips, K., Clermont, G., Akcakaya, M., Sereika, S.M., Van Dam, P., Smith, S.W., Birnbaum, Y., Saba, S., Sejdic, E., Callaway, C.W.: Machine learning for ECG diagnosis and risk stratification of occlusion myoc...

    work page doi:10.1038/s41591-023-02396-3 2023

  11. [11]

    Communications Medicine3, 73–73 (2023) https://doi.org/10.1038/ s43856-023-00278-w

    Holmstrom, L., Christensen, M., Yuan, N., Weston Hughes, J., Theurer, J., Jujjavarapu, M., Fatehi, P., Kwan, A., Sandhu, R.K., Ebinger, J., Cheng, S., Zou, J., Chugh, S.S., Ouyang, D.: Deep learning-based electrocardio- graphic screening for chronic kidney disease. Communications Medicine3, 73–73 (2023) https://doi.org/10.1038/ s43856-023-00278-w

    2023

  12. [12]

    Nature Medicine25(1), 70–74 (2019) https://doi.org/10.1038/s41591-018-0240-2

    Attia, Z.I., Kapa, S., Lopez-Jimenez, F., McKie, P.M., Ladewig, D.J., Satam, G., Pellikka, P.A., Enriquez-Sarano, M., Noseworthy, P.A., Munger, T.M., Asirvatham, S.J., Scott, C.G., Carter, R.E., Friedman, P.A.: Screening for cardiac contractile dysfunction using an artificial intelligence–enabled electrocardiogram. Nature Medicine25(1), 70–74 (2019) https...

    doi:10.1038/s41591-018-0240-2 2019

  13. [13]

    Journal of Cardiovascular Electrophysiology30(5), 668–674 (2019) https: //doi.org/10.1111/JCE.13889

    Attia, Z.I., Kapa, S., Yao, X., Lopez-Jimenez, F., Mohan, T.L., Pellikka, P.A., Carter, R.E., Shah, N.D., Friedman, P.A., Noseworthy, P.A.: Prospective validation of a deep learning electrocardiogram algorithm for the detection of left ventricular systolic dysfunction. Journal of Cardiovascular Electrophysiology30(5), 668–674 (2019) https: //doi.org/10.11...

    work page doi:10.1111/jce.13889 2019

  14. [14]

    Nature Reviews Cardiology14(10), 591–602 (2017) https://doi.org/10.1038/nrcardio.2017.65

    Dunlay, S.M., Roger, V.L., Redfield, M.M.: Epidemiology of heart failure with preserved ejection fraction. Nature Reviews Cardiology14(10), 591–602 (2017) https://doi.org/10.1038/nrcardio.2017.65

    work page doi:10.1038/nrcardio.2017.65 2017

  15. [15]

    Nature Reviews Cardiology21(10), 717–734 (2024) https://doi.org/10.1038/s41569-024-01046-6

    Khan, M.S., Shahid, I., Bennis, A., Rakisheva, A., Metra, M., Butler, J.: Global epidemiology of heart failure. Nature Reviews Cardiology21(10), 717–734 (2024) https://doi.org/10.1038/s41569-024-01046-6

    work page doi:10.1038/s41569-024-01046-6 2024

  16. [16]

    European Heart Journal43(20), 1917–1927 (2022) https://doi.org/10.1093/ eurheartj/ehac088

    Prinzen, F.W., Auricchio, A., Mullens, W., Linde, C., Huizar, J.F.: Electrical management of heart failure: from pathophysiology to treatment. European Heart Journal43(20), 1917–1927 (2022) https://doi.org/10.1093/ eurheartj/ehac088

    1917

  17. [17]

    Journal of the American College of Cardiology73(19), 2388–2397 (2019) https://doi.org/10

    Khan, S.S., Ning, H., Shah, S.J., Yancy, C.W., Carnethon, M., Berry, J.D., Mentz, R.J., O’Brien, E., Correa, A., Suthahar, N., Boer, R.A., Wilkins, J.T., Lloyd-Jones, D.M.: 10-Year risk equations for incident heart failure in the general population. Journal of the American College of Cardiology73(19), 2388–2397 (2019) https://doi.org/10. 1016/j.jacc.2019.02.057

    2019

  18. [18]

    In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp

    Abnar, S., Zuidema, W.: Quantifying attention flow in transformers. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp. 4190–4197. Association for Computational Linguistics, Online (2020). https://doi.org/10.18653/v1/2020.acl-main.385

    doi:10.18653/v1/2020.acl-main.385 2020

  19. [19]

    https://jacobgil.github.io/deeplearning/ vision-transformer-explainability

    Gildenblat, J.: Exploring Explainability for Vision Transformers (2020). https://jacobgil.github.io/deeplearning/ vision-transformer-explainability

    2020

  20. [20]

    Circulation145(2), 122–133 (2022) https://doi.org/10.1161/CIRCULATIONAHA.121.057480

    Khurshid, S., Friedman, S., Reeder, C., Di Achille, P., Diamant, N., Singh, P., Harrington, L.X., Wang, X., Al-Alusi, M.A., Sarma, G., Foulkes, A.S., Ellinor, P.T., Anderson, C.D., Ho, J.E., Philippakis, A.A., Batra, P., Lubitz, S.A.: ECG-based deep learning and clinical risk factors to predict atrial fibrillation. Circulation145(2), 122–133 (2022) https:...

    doi:10.1161/circulationaha.121.057480 2022

  21. [21]

    European Heart Journal42(38), 3948–3961 (2021) https://doi.org/10.1093/EURHEARTJ/EHAB588

    Prifti, E., Fall, A., Davogustto, G., Pulini, A., Denjoy, I., Funck-Brentano, C., Khan, Y., Durand-Salmon, A., Badilini, F., Wells, Q.S., Leenhardt, A., Zucker, J.D., Roden, D.M., Extramiana, F., Salem, J.E.: Deep learning analysis of electrocardiogram for risk prediction of drug-induced arrhythmias and diagnosis of long QT syndrome. European Heart Journa...

    work page doi:10.1093/eurheartj/ehab588 2021

  22. [22]

    npj Digital Medicine6, 169–169 (2023) https://doi.org/10.1038/s41746-023-00916-6

    Hughes, J.W., Tooley, J., Torres Soto, J., Ostropolets, A., Poterucha, T., Christensen, M.K., Yuan, N., Ehlert, B., Kaur, D., Kang, G., Rogers, A., Narayan, S., Elias, P., Ouyang, D., Ashley, E., Zou, J., Perez, M.V.: A deep learning- based electrocardiogram risk score for long term cardiovascular death and disease. npj Digital Medicine6, 169–169 (2023) h...

    work page doi:10.1038/s41746-023-00916-6 2023

  23. [23]

    The American Journal of Cardiology168, 105–109 (2022) https://doi.org/10.1016/j.amjcard.2021.12.017

    Khan, S.S., Barda, N., Greenland, P., Dagan, N., Lloyd-Jones, D.M., Balicer, R., Rasmussen-Torvik, L.J.: Validation of heart failure-specific risk equations in 1.3 million Israeli adults and usefulness of combining ambulatory and hospitalization data from a aarge integrated health care organization. The American Journal of Cardiology168, 105–109 (2022) ht...

    work page doi:10.1016/j.amjcard.2021.12.017 2022

  24. [24]

    Circulation: Heart 15 Failure14(2), 007761 (2021) https://doi.org/10.1161/CIRCHEARTF AILURE.120.007761

    Sinha, A., Gupta, D.K., Yancy, C.W., Shah, S.J., Rasmussen-Torvik, L.J., McNally, E.M., Greenland, P., Lloyd- Jones, D.M., Khan, S.S.: Risk-based approach for the prediction and prevention of heart failure. Circulation: Heart 15 Failure14(2), 007761 (2021) https://doi.org/10.1161/CIRCHEARTF AILURE.120.007761

    work page doi:10.1161/circheartf 2021

  25. [25]

    European Heart Journal46(11), 1044–1053 (2025) https://doi.org/10.1093/eurheartj/ehae914

    Dhingra, L.S., Aminorroaya, A., Sangha, V., Pedroso, A.F., Asselbergs, F.W., Brant, L.C.C., Barreto, S.M., Ribeiro, A.L.P., Krumholz, H.M., Oikonomou, E.K., Khera, R.: Heart failure risk stratification using artificial intelligence applied to electrocardiogram images: a multinational study. European Heart Journal46(11), 1044–1053 (2025) https://doi.org/10...

    work page doi:10.1093/eurheartj/ehae914 2025

  26. [26]

    Cardiovascular Digital Health Journal4(6), 183–190 (2023) https://doi.org/10.1016/j.cvdhj.2023.11.003

    Butler, L., Karabayir, I., Kitzman, D.W., Alonso, A., Tison, G.H., Chen, L.Y., Chang, P.P., Clifford, G., Soliman, E.Z., Akbilgic, O.: A generalizable electrocardiogram-based artificial intelligence model for 10-year heart failure risk prediction. Cardiovascular Digital Health Journal4(6), 183–190 (2023) https://doi.org/10.1016/j.cvdhj.2023.11.003 . Acces...

    work page doi:10.1016/j.cvdhj.2023.11.003 2023

  27. [27]

    European Heart Journal - Digital Health2(4), 626–634 (2021) https://doi.org/10.1093/ehjdh/ztab080

    Akbilgic, O., Butler, L., Karabayir, I., Chang, P.P., Kitzman, D.W., Alonso, A., Chen, L.Y., Soliman, E.Z.: ECG- AI: electrocardiographic artificial intelligence model for prediction of heart failure. European Heart Journal - Digital Health2(4), 626–634 (2021) https://doi.org/10.1093/ehjdh/ztab080 . Accessed 2025-07-12

    work page doi:10.1093/ehjdh/ztab080 2021

  28. [28]

    IEEE Transactions on Neural Systems and Rehabilitation Engineering27(3), 400–410 (2019) https://doi.org/10.1109/TNSRE.2019.2896659

    Phan, H., Andreotti, F., Cooray, N., Ch´ en, O.Y., De Vos, M.: SeqSleepNet: End-to-End Hierarchical Recurrent Neural Network for Sequence-to-Sequence Automatic Sleep Staging. IEEE Transactions on Neural Systems and Rehabilitation Engineering27(3), 400–410 (2019) https://doi.org/10.1109/TNSRE.2019.2896659 . Accessed 2025- 07-05

    arXiv 2019

  29. [29]

    Nature Medicine30(5), 1461–1470 (2024) https://doi.org/10.1038/s41591-024-02961-4

    Lin, C.-S., Liu, W.-T., Tsai, D.-J., Lou, Y.-S., Chang, C.-H., Lee, C.-C., Fang, W.-H., Wang, C.-C., Chen, Y.-Y., Lin, W.-S., Cheng, C.-C., Lee, C.-C., Wang, C.-H., Tsai, C.-S., Lin, S.-H., Lin, C.: AI-enabled electrocardiography alert intervention and all-cause mortality: a pragmatic randomized clinical trial. Nature Medicine30(5), 1461–1470 (2024) https...

    work page doi:10.1038/s41591-024-02961-4 2024

  30. [30]

    Jain, S., Wallace, B.C.: Attention is not Explanation. arXiv. arXiv:1902.10186 [cs] (2019). https://doi.org/10.48550/ arXiv.1902.10186 . http://arxiv.org/abs/1902.10186 Accessed 2025-07-04

    Pith/arXiv arXiv doi:10.48550/arxiv.1902.10186 1902

  31. [31]

    782–791 (2021)

    Chefer, H., Gur, S., Wolf, L.: Transformer Interpretability Beyond Attention Visualization, pp. 782–791 (2021)

    2021

  32. [32]

    Jo, S., Jang, G., Park, H.: GMAR: Gradient-Driven Multi-Head Attention Rollout for Vision Transformer Inter- pretability. arXiv. arXiv:2504.19414 [cs] (2025). https://doi.org/10.48550/arXiv.2504.19414 . http://arxiv.org/abs/ 2504.19414 Accessed 2025-07-04

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2504.19414 2025

  33. [33]

    Advanced Drug Delivery Reviews59(9-10), 940–951 (2007) https://doi.org/10.1016/J.ADDR.2006.10.011

    Portaluppi, F., Hermida, R.C.: Circadian rhythms in cardiac arrhythmias and opportunities for their chronotherapy. Advanced Drug Delivery Reviews59(9-10), 940–951 (2007) https://doi.org/10.1016/J.ADDR.2006.10.011

    work page doi:10.1016/j.addr.2006.10.011 2007

  34. [34]

    Journal of the American College of Cardiology66(2), 101–109 (2015) https://doi.org/10.1016/J.JACC.2015.04.062

    Dukes, J.W., Dewland, T.A., Vittinghoff, E., Mandyam, M.C., Heckbert, S.R., Siscovick, D.S., Stein, P.K., Psaty, B.M., Sotoodehnia, N., Gottdiener, J.S., Marcus, G.M.: Ventricular ectopy as a predictor of heart failure and death. Journal of the American College of Cardiology66(2), 101–109 (2015) https://doi.org/10.1016/J.JACC.2015.04.062

    work page doi:10.1016/j.jacc.2015.04.062 2015

  35. [35]

    The American Journal of Cardiology91(6), 2–8 (2003) https://doi.org/10.1016/S0002-9149(02)03373-8

    Maisel, W.H., Stevenson, L.W.: Atrial fibrillation in heart failure: epidemiology, pathophysiology, and rationale for therapy. The American Journal of Cardiology91(6), 2–8 (2003) https://doi.org/10.1016/S0002-9149(02)03373-8

    work page doi:10.1016/s0002-9149(02)03373-8 2003

  36. [36]

    Journal of Clinical Medicine11(9), 2510–2510 (2022) https://doi.org/10.3390/JCM11092510

    Bergau, L., Bengel, P., Sciacca, V., Fink, T., Sohns, C., Sommer, P.: Atrial fibrillation and heart failure. Journal of Clinical Medicine11(9), 2510–2510 (2022) https://doi.org/10.3390/JCM11092510

    work page doi:10.3390/jcm11092510 2022

  37. [37]

    JAMA310(1), 66–74 (2013) https://doi.org/10.1001/jama.2013.7588

    Ledwidge, M., Gallagher, J., Conlon, C., Tallon, E., O’Connell, E., Dawkins, I., Watson, C., O’Hanlon, R., Berming- ham, M., Patle, A., Badabhagni, M.R., Murtagh, G., Voon, V., Tilson, L., Barry, M., McDonald, L., Maurer, B., McDonald, K.: Natriuretic Peptide–Based Screening and Collaborative Care for Heart Failure: The STOP-HF Randomized Trial. JAMA310(1...

    arXiv 2013

  38. [38]

    JAMA310(1), 44–45 (2013) https://doi.org/10.1001/jama.2013.7589

    Hernandez, A.F.: Preventing Heart Failure. JAMA310(1), 44–45 (2013) https://doi.org/10.1001/jama.2013.7589 . Accessed 2024-12-12

    arXiv 2013

  39. [39]

    arXiv (2024)

    McKeen, K., Oliva, L., Masood, S., Toma, A., Rubin, B., Wang, B.: ECG-FM: An Open Electrocardiogram Foundation Model. arXiv (2024). https://doi.org/10.48550/ARXIV.2408.05178 . https://arxiv.org/abs/2408.05178

    Pith/arXiv arXiv doi:10.48550/arxiv.2408.05178 2024

  40. [40]

    Interpretable Pre-Trained Transformers for Heart Time-Series Data

    Davies, H.J., Monsen, J., Mandic, D.P.: Interpretable Pre-Trained Transformers for Heart Time-Series Data. arXiv (2024). https://doi.org/10.48550/ARXIV.2407.20775 . https://arxiv.org/abs/2407.20775

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2407.20775 2024

  41. [41]

    arXiv (2024)

    Li, J., Aguirre, A., Moura, J., Liu, C., Zhong, L., Sun, C., Clifford, G., Westover, B., Hong, S.: An electrocardiogram foundation model built on over 10 million recordings with external evaluation across multiple domains. arXiv (2024). 16 https://doi.org/10.48550/arXiv.2410.04133 . http://arxiv.org/abs/2410.04133

    Pith/arXiv arXiv doi:10.48550/arxiv.2410.04133 2024

  42. [42]

    SiamAF: Learning Shared Information from ECG and PPG Signals for Robust Atrial Fibrillation Detection

    Guo, Z., Ding, C., Do, D.H., Shah, A., Lee, R.J., Hu, X., Rudin, C.: SiamAF: Learning Shared Information from ECG and PPG Signals for Robust Atrial Fibrillation Detection. arXiv (2024). https://doi.org/10.48550/ARXIV. 2310.09203 . https://arxiv.org/abs/2310.09203

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2310.09203 2024

  43. [43]

    In: The Twelfth International Conference on Learning Representations - ICLR 2024 (2024).https://openreview.net/forum?id=pC3WJHf51j

    Abbaspourazad, S., Elachqar, O., Miller, A.C., Emrani, S., Nallasamy, U., Shapiro, I.: Large-scale Training of Foundation Models for Wearable Biosignals. In: The Twelfth International Conference on Learning Representations - ICLR 2024 (2024).https://openreview.net/forum?id=pC3WJHf51j

    2024

  44. [44]

    Transactions on Machine Learning Research (2023)

    D´ efossez, A., Copet, J., Synnaeve, G., Adi, Y.: High Fidelity Neural Audio Compression. Transactions on Machine Learning Research (2023)

    2023

  45. [45]

    In: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining

    Akiba, T., Sano, S., Yanase, T., Ohta, T., Koyama, M.: Optuna: A Next-generation Hyperparameter Optimization Framework. In: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. KDD ’19, pp. 2623–2631. Association for Computing Machinery, New York, NY, USA (2019). https://doi. org/10.1145/3292500.3330701

    arXiv 2019

  46. [46]

    Technical report, Jerusalem (2023)

    Israel Central Bureau of Statistics: The Population of Ethiopian Origin in Israel: Selected Data Published on the Occasion of the Sigd Festival 2023. Technical report, Jerusalem (2023). https://www.cbs.gov.il/en/mediarelease/ Pages/2023/The-Ethiopian-Population-in-Israel-2023.aspx

    2023

  47. [47]

    Behavior Research Methods53(4), 1689–1696 (2021) https://doi.org/10.3758/S13428-020-01516-Y/TABLES/3

    Makowski, D., Pham, T., Lau, Z.J., Brammer, J.C., Lespinasse, F., Pham, H., Scholzel, C., Chen, S.H.A.: NeuroKit2: A Python toolbox for neurophysiological signal processing. Behavior Research Methods53(4), 1689–1696 (2021) https://doi.org/10.3758/S13428-020-01516-Y/TABLES/3

    work page doi:10.3758/s13428-020-01516-y/tables/3 2021

  48. [48]

    In: 2021 Computing in Cardiology (CinC), vol

    Gendelman, S., Biton, S., Derman, R., Zvuloni, E., Levy, J., Lugassy, S., Alexandrovich, A., Behar, J.A.: Physio- Zoo ECG: Digital electrocardiography biomarkers to assess cardiac conduction. In: 2021 Computing in Cardiology (CinC), vol. 2021-September, pp. 1–4. IEEE, Brno, Czech Republic (2021). https://doi.org/10.23919/CinC53138. 2021.9662857

    arXiv 2021

  49. [49]

    Sobel, J., Alexandrovich, A., Charlton, P., Marton Aron Goda, D.: PhysioZoo: The Open Physiological Biomarkers Resource

    Behar, J.A., Levy, J., Zvuloni, E., Gendelman, S., Rosenberg, A., Biton, S., Derman, R., A. Sobel, J., Alexandrovich, A., Charlton, P., Marton Aron Goda, D.: PhysioZoo: The Open Physiological Biomarkers Resource. In: Computing in Cardiology. IEEE Computer Society, Atlanta, Georgia, USA (2023). https://doi.org/10.22489/CinC.2023.190

    work page doi:10.22489/cinc.2023.190 2023

  50. [50]

    Journal of Machine Learning Research12(85), 2825–2830 (2011)

    Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., Cournapeau, D., Brucher, M., Perrot, M., Duchesnay, E.: Scikit-learn: Machine Learning in Python. Journal of Machine Learning Research12(85), 2825–2830 (2011)

    2011

  51. [51]

    Journal of Open Source Software4(40), 1317 (2019) https://doi.org/10.21105/joss.01317

    Davidson-Pilon, C.: lifelines: survival analysis in Python. Journal of Open Source Software4(40), 1317 (2019) https://doi.org/10.21105/joss.01317

    doi:10.21105/joss.01317 2019

  52. [52]

    BMC Medical Research Methodology20(1), 53 (2020) https://doi.org/10.1186/s12874-019-0890-x

    Goyal, P., Mefford, M.T., Chen, L., Sterling, M.R., Durant, R.W., Safford, M.M., Levitan, E.B.: Assembling and validating a heart failure-free cohort from the Reasons for Geographic and Racial Differences in Stroke (REGARDS) study. BMC Medical Research Methodology20(1), 53 (2020) https://doi.org/10.1186/s12874-019-0890-x . Accessed 2024-11-01

    work page doi:10.1186/s12874-019-0890-x 2020

  53. [53]

    recording date before 2010 or after 2023

    McDonagh, T.A., Metra, M., Adamo, M., Gardner, R.S., Baumbach, A., B¨ ohm, M., Burri, H., Butler, J.,ˇCelutkien˙ e, J., Chioncel, O., Cleland, J.G.F., Coats, A.J.S., Crespo-Leiro, M.G., Farmakis, D., Gilard, M., Heymans, S., Hoes, A.W., Jaarsma, T., Jankowska, E.A., Lainscak, M., Lam, C.S.P., Lyon, A.R., McMurray, J.J.V., Mebazaa, A., Mindham, R., Muneret...

    2021