rECHOmmend: An ECG-Based Machine Learning Approach for Identifying Patients at Increased Risk of Undiagnosed Structural Heart Disease Detectable by Echocardiography
Open accessPublished version of a preprint
Abstract
Background: Timely diagnosis of structural heart disease improves patient outcomes, yet many remain underdiagnosed. While population screening with echocardiography is impractical, ECG-based prediction models can help target high-risk patients. We developed a novel ECG-based machine learning approach to predict multiple structural heart conditions, hypothesizing that a composite model would yield higher prevalence and positive predictive values to facilitate meaningful recommendations for echocardiography. Methods: Using 2 232 130 ECGs linked to electronic health records and echocardiography reports from 484 765 adults between 1984 to 2021, we trained machine learning models to predict the presence or absence of any of 7 echocardiography-confirmed diseases within 1 year. This composite label included the following: moderate or severe valvular disease (aortic/mitral stenosis or regurgitation, tricuspid regurgitation), reduced ejection fraction <50%, or interventricular septal thickness >15 mm. We tested various combinations of input features (demographics, laboratory values, structured ECG data, ECG traces) and evaluated model performance using 5-fold cross-validation, multisite validation trained on 1 site and tested on 10 independent sites, and simulated retrospective deployment trained on pre-2010 data and deployed in 2010. Results: Our composite rECHOmmend model used age, sex, and ECG traces and had a 0.91 area under the receiver operating characteristic curve and a 42% positive predictive value at 90% sensitivity, with a composite label prevalence of 17.9%. Individual disease models had area under the receiver operating characteristic curves from 0.86 to 0.93 and lower positive predictive values from 1% to 31%. Area under the receiver operating characteristic curves for models using different input features ranged from 0.80 to 0.93, increasing with additional features. Multisite validation showed similar results to cross-validation, with an aggregate area under the receiver operating characteristic curve of 0.91 across our independent test set of 10 clinical sites after training on a separate site. Our simulated retrospective deployment showed that for ECGs acquired in patients without preexisting structural heart disease in the year 2010, 11% were classified as high risk and 41% (4.5% of total patients) developed true echocardiography-confirmed disease within 1 year. Conclusions: An ECG-based machine learning model using a composite end point can identify a high-risk population for having undiagnosed, clinically significant structural heart disease while outperforming single-disease models and improving practical utility with higher positive predictive values. This approach can facilitate targeted screening with echocardiography to improve underdiagnosis of structural heart disease.
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.
Integrity signals
Registry
Cites
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.
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?
Echocardiography is the principal confirmatory imaging modality for most forms of SHD, but broad referral strategies are difficult to scale because imaging capacity, expertise, and patient selection constrain efficiency.
“Echocardiography is the principal confirmatory test, but its broad use as a screening tool is constrained by imaging capacity, cost, and referral efficiency.”
From AI-ECG for Echocardiography Triage in Structural Heart Disease: Evidence, Implementation, and Future Directions (2026) · cites this paper as [9]
The paper calls echocardiography the primary diagnostic test for many structural conditions (Clinical Implications) and says population screening has been ineffective because disease prevalence is low (Discussion). The Discussion also identifies limits on the volume of echocardiograms and the need to select patients likely to benefit. It does not identify cost as a constraint on broad screening. (LLM, conf=0.85)
Experimental — accuracy not yet measuredMore general AI-ECG models can accurately detect a composite of low left ventricular systolic dysfunction, low left ventricular hypertrophy and moderate or greater VHD.
“More general AI-ECG models have also been developed that can accurately detect a composite of low left ventricular systolic dysfunction, low left ventricular hypertrophy and moderate or greater VHD4.”
From Detecting structural heart disease from electrocardiograms using AI (2025) · cites this paper as [4]
The Abstract (Methods) defines a composite of reduced ejection fraction (<50%), increased interventricular septal thickness (>15 mm), and moderate or severe valvular disease; the Abstract (Results) reports an AUROC of 0.91 for the composite model. The claim's “low left ventricular hypertrophy” does not match the paper's increased septal thickness criterion. (LLM, conf=0.88)
Experimental — accuracy not yet measuredThese ensemble models use raw ECG voltage data as inputs, which are inaccessible to clinicians at the point of care and often require modifications to the technical infrastructure to account for vendor-specific data formats.
“Nonetheless, these models use raw ECG voltage data as inputs, which are inaccessible to clinicians at the point of care and often require modifications to the technical infrastructure to account for vendor-specific data formats.19”
From Ensemble Deep Learning Algorithm for Structural Heart Disease Screening Using Electrocardiographic Images (2025) · cites this paper as [19]
Methods (Model Development; page not provided) says the CNN models use raw ECG trace data as input, supporting that part of the claim. The paper does not say those data are inaccessible to clinicians at the point of care or that vendor-specific formats often require infrastructure changes. Instead, the Discussion says the featured model uses data readily available from ECG systems and could be easily deployed across most health care systems. (LLM, conf=0.87)
Experimental — accuracy not yet measuredEnsemble models for detecting a composite of multiple structural heart diseases have been proposed to overcome low precision.
“To overcome this limitation, ensemble models for detecting a composite of multiple SHDs have been proposed.19”
From Ensemble Deep Learning Algorithm for Structural Heart Disease Screening Using Electrocardiographic Images (2025) · cites this paper as [19]
The Clinical Implications section says the authors combined seven structural heart disease outcomes into a composite prediction to increase diagnostic yield, and the Results report higher positive predictive value than the individual-disease models. But the Methods describe a model trained to predict the composite label, not an ensemble of models; that characterization is unsupported by our copy. No page numbers are available in the indexed passages. (LLM, conf=0.86)
Experimental — accuracy not yet measuredValvular and composite SHD AI-ECG models are promising for referral enrichment.
“Valvular and composite SHD models are promising for referral enrichment”
From AI-ECG for Echocardiography Triage in Structural Heart Disease: Evidence, Implementation, and Future Directions (2026) · cites this paper as [9]
The rECHOmmend composite model used age, sex, and ECG traces and achieved an AUROC of 0.91 with a positive predictive value of 42% at 90% sensitivity, outperforming corresponding single-disease models in practical referral yield.
“The composite rECHOmmend model used age, sex, and ECG traces and achieved an AUROC of 0.91 with a positive predictive value of 42% at 90% sensitivity, outperforming the corresponding single-disease models in practical referral yield.9”
From AI-ECG for Echocardiography Triage in Structural Heart Disease: Evidence, Implementation, and Future Directions (2026) · cites this paper as [9]
Many patients remain undiagnosed until disease is advanced, and ECG-based prediction models may help enrich referral toward higher-risk patients in a more targeted manner.
“Accordingly, many patients remain undiagnosed until disease is advanced, and electrocardiogram (ECG)-based prediction models may help enrich referral toward higher-risk patients in a more targeted manner.9”
From AI-ECG for Echocardiography Triage in Structural Heart Disease: Evidence, Implementation, and Future Directions (2026) · cites this paper as [9]
Other ECG-based SHD models, such as rECHOmmend, have demonstrated excellent performance with an AUROC of 91% in its retrospective dataset.
“Other ECG-based SHD models, such as rECHOmmend, have demonstrated excellent performance with an AUROC of 91% in its retrospective dataset4.”
From Detecting structural heart disease from electrocardiograms using AI (2025) · cites this paper as [4]
The performance of rECHOmmend was observed to drop (AUROC 0.88) when restricting to patients with echocardiography-confirmed disease, thereby increasing disease prevalence, as was done for the present study.
“the performance of rECHOmmend was observed to drop (AUROC 0.88) when restricting to patients with echocardiography-confirmed disease, thereby increasing disease prevalence, as was done for the present study.”
From Detecting structural heart disease from electrocardiograms using AI (2025) · cites this paper as [4]
The simultaneous detection of multiple SHDs increases the composite disease prevalence and improves model precision.
“The simultaneous detection of multiple SHDs increases the composite disease prevalence and improves model precision.19”
From Ensemble Deep Learning Algorithm for Structural Heart Disease Screening Using Electrocardiographic Images (2025) · cites this paper as [19]
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
Otto, C. M., Nishimura, R. A., Bonow, R. O., Carabello, B. A., Erwin, J. P., Gentile, F., Jneid, H., Krieger, E. V., Mack, M., McLeod, C., et al. (2021) · Circulation · doi:10.1161/cir.0000000000000923 · Cited by 3,686 · see citing works
Raghunath, S., Pfeifer, J. M., Ulloa-Cerna, A. E., Nemani, A., Carbonati, T., Jing, L., vanMaanen, D. P., Hartzel, D. N., Ruhl, J. A., Lagerman, B. F., et al. (2021) · Circulation · doi:10.1161/circulationaha.120.047829 · Cited by 312 · see citing works
Flagged first, then by citations
Research works printed but not linked: 31
- [1]Ross J Jr Braunwald E. Aortic stenosis. Circulation. 1968; 38(1 Suppl): 61– 67. doi: 10.1161/01.cir.38.1s5.v-61 4894151 · DOI
- [2]Cheitlin MD Gertz EW Brundage BH Carlson CJ Quash JA Bode RS Jr. Rate of progression of severity of valvular aortic stenosis in the adult. Am Heart J. 1979; 98: 689– 700. doi: 10.1016/0002-8703(79)90465-4 495418 · DOI
- [3]Davies SW Gershlick AH Balcon R. Progression of valvar aortic stenosis: a long-term retrospective study. Eur Heart J. 1991; 12: 10– 14. doi: 10.1093/oxfordjournals.eurheartj.a059815 2009886 · DOI
- [4]Curtis JP Sokol SI Wang Y Rathore SS Ko DT Jadbabaie F Portnay EL Marshalko SJ Radford MJ Krumholz HM. The association of left ventricular ejection fraction, mortality, and cause of death in stable outpatients with heart failure. J Am Coll Cardiol. 2003; 42: 736– 742. doi: 10.1016/s0735-1097(03)00789-7 12932612 · DOI
- [5]Martinez-Naharro A Baksi AJ Hawkins PN Fontana M. Diagnostic imaging of cardiac amyloidosis. Nat Rev Cardiol. 2020; 17: 413– 426. doi: 10.1038/s41569-020-0334-7 32042139 · DOI
- [7]Cheitlin MD Alpert JS Armstrong WF Aurigemma GP Beller GA Bierman FZ Davidson TW Davis JL Douglas PS Gillam LD. ACC/AHA guidelines for the clinical application of echocardiography. A report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines (Committee on Clinical Application of Echocardiography). Developed in collaboration with the American Society of Echocardiography. Circulation. 1997; 95: 1686– 1744. doi: 10.1161/01.cir.95.6.1686 9118558 · DOI
- [8]Lang RM Badano LP Mor-Avi V Afilalo J Armstrong A Ernande L Flachskampf FA Foster E Goldstein SA Kuznetsova T. Recommendations for cardiac chamber quantification by echocardiography in adults: an update from the American Society of Echocardiography and the European Association of Cardiovascular Imaging. Eur Heart J Cardiovasc Imaging. 2015; 16: 233– 270. doi: 10.1093/ehjci/jev014 25712077 · DOI
- [9]Nkomo VT Gardin JM Skelton TN Gottdiener JS Scott CG Enriquez-Sarano M. Burden of valvular heart diseases: a population-based study. Lancet. 2006; 368: 1005– 1011. doi: 10.1016/S0140-6736(06)69208-8 16980116 · DOI · Paper page
- [10]Alexander KM Orav J Singh A Jacob SA Menon A Padera RF Kijewski MF Liao R Di Carli MF Laubach JP. Geographic disparities in reported US amyloidosis mortality from 1979 to 2015: potential underdetection of cardiac amyloidosis. JAMA Cardiol. 2018; 3: 865– 870. doi: 10.1001/jamacardio.2018.2093 30046835 PMC6233639 · DOI
- [11]d’Arcy JL Coffey S Loudon MA Kennedy A Pearson-Stuttard J Birks J Frangou E Farmer AJ Mant D Wilson J. Large-scale community echocardiographic screening reveals a major burden of undiagnosed valvular heart disease in older people: the OxVALVE Population Cohort Study. Eur Heart J. 2016; 37: 3515– 3522. doi: 10.1093/eurheartj/ehw229 27354049 PMC5216199 · DOI · Paper page
- [12]Maron MS Hellawell JL Lucove JC Farzaneh-Far R Olivotto I. Occurrence of clinically diagnosed hypertrophic cardiomyopathy in the United States. Am J Cardiol. 2016; 117: 1651– 1654. doi: 10.1016/j.amjcard.2016.02.044 27006153 · DOI
- [13]Park SJ Enriquez-Sarano M Chang SA Choi JO Lee SC Park SW Kim DK Jeon ES Oh JK. Hemodynamic patterns for symptomatic presentations of severe aortic stenosis. JACC Cardiovasc Imaging. 2013; 6: 137– 146. doi: 10.1016/j.jcmg.2012.10.013 23489526 · DOI
- [14]Goliasch G Kammerlander AA Nitsche C Dona C Schachner L Öztürk B Binder C Duca F Aschauer S Laufer G. Syncope: the underestimated threat in severe aortic stenosis. JACC Cardiovasc Imaging. 2019; 12: 225– 232. doi: 10.1016/j.jcmg.2018.09.020 30553685 · DOI
- [15]Selzer A. Changing aspects of the natural history of valvular aortic stenosis. N Engl J Med. 1987; 317: 91– 98. doi: 10.1056/NEJM198707093170206 3295554 · DOI
- [17]Kwon JM Lee SY Jeon KH Lee Y Kim KH Park J Oh BH Lee MM. Deep learning-based algorithm for detecting aortic stenosis using electrocardiography. J Am Heart Assoc. 2020; 9: e014717. doi: 10.1161/JAHA.119.014717 32200712 PMC7428650 · DOI · Paper page
- [18]Attia ZI Kapa S Lopez-Jimenez F McKie PM Ladewig DJ Satam G Pellikka PA Enriquez-Sarano M Noseworthy PA Munger TM. Screening for cardiac contractile dysfunction using an artificial intelligence-enabled electrocardiogram. Nat Med. 2019; 25: 70– 74. doi: 10.1038/s41591-018-0240-2 30617318 · DOI · Paper page
- [19]Cohen-Shelly M Attia ZI Friedman PA Ito S Essayagh BA Ko WY Murphree DH Michelena HI Enriquez-Sarano M Carter RE. Electrocardiogram screening for aortic valve stenosis using artificial intelligence. Eur Heart J. 2021; 42: 2885– 2896. doi: 10.1093/eurheartj/ehab153 33748852 · DOI · Paper page
- [20]Samad MD Ulloa A Wehner GJ Jing L Hartzel D Good CW Williams BA Haggerty CM Fornwalt BK. Predicting survival from large echocardiography and electronic health record datasets: optimization with machine learning. JACC Cardiovas Imaging. 2019; 12: 681– 689. PMC6286869 29909114
- [21]Ommen SR Mital S Burke MA Day SM Deswal A Elliott P Evanovich LL Hung J Joglar JA Kantor P. 2020 AHA/ACC guideline for the diagnosis and treatment of patients with hypertrophic cardiomyopathy: a report of the American College of Cardiology/American Heart Association Joint Committee on Clinical Practice Guidelines. Circulation. 2020; 142: e558– e631. doi: 10.1161/CIR.0000000000000937 33215931 · DOI
- [22]Ioffe S Szegedy C. Batch normalization: accelerating deep network training by reducing internal covariate shift. Proceedings of the 32nd International Conference on Machine Learning. PMLR. 2015; 37: 448– 456.
- [23]Chen T Guestrin C. XGBoost: a scalable tree boosting system. In Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York: Association for Computing Machinery; 2016: 785– 794. doi: 10.1145/2939672.2939785 · DOI · Paper page
- [24]Platt J. Probabilistic outputs for support vector machines and comparisons to regularized likelihood methods. Advances in Large Margin Classifiers. 1999; 10: 61– 74.
- [25]Bodison SA Wesley YE Tucker E Green KJ. Results of screening a large group of intercollegiate competitive athletes for cardiovascular disease. J Am Coll Cardiol. 1987; 10: 1214– 1221. 2960727
- [26]Hada Y Sakamoto T Amano K Yamaguchi T Takenaka K Takahashi H Takikawa R Hasegawa I Takahashi T Suzuki J. Prevalence of hypertrophic cardiomyopathy in a population of adult Japanese workers as detected by echocardiographic screening. Am J Cardiol. 1987; 59: 183– 184. doi: 10.1016/s0002-9149(87)80107-8 3812242 · DOI
- [27]Baumgartner H Hung J Bermejo J Chambers JB Edvardsen T Goldstein S Lancellotti P LeFevre M Miller F Jr Otto CM. Recommendations on the echocardiographic assessment of aortic valve stenosis: a focused update from the European Association of Cardiovascular Imaging and the American Society of Echocardiography. J Am Soc Echocardiogr. 2017; 30: 372– 392. doi: 10.1016/j.echo.2017.02.009 28385280 · DOI
- [28]Davies M Hobbs F Davis R Kenkre J Roalfe AK Hare R Wosornu D Lancashire RJ. Prevalence of left-ventricular systolic dysfunction and heart failure in the Echocardiographic Heart of England Screening study: a population based study. Lancet. 2001; 358: 439– 444. doi: 10.1016/s0140-6736(01)05620-3 11513906 · DOI · Paper page
- [29]Banovic M Putnik S Penicka M Doros G Deja MA Kockova R Kotrc M Glaveckaite S Gasparovic H Pavlovic N; AVATAR Trial Investigators*. Aortic valve replacement versus conservative treatment in asymptomatic severe aortic stenosis: the AVATAR Trial. Circulation. 2022; 145: 648– 658. doi: 10.1161/CIRCULATIONAHA.121.057639 34779220 · DOI
- [30]Kagiyama N Piccirilli M Yanamala N Shrestha S Farjo PD Casaclang-Verzosa G Tarhuni WM Nezarat N Budoff MJ Narula J. Machine learning assessment of left ventricular diastolic function based on electrocardiographic features. J Am Coll Cardiol. 2020; 76: 930– 941. doi: 10.1016/j.jacc.2020.06.061 32819467 · DOI
- [31]Wolters FJ. An AI-ECG algorithm for atrial fibrillation risk: steps towards clinical implementation. Lancet. 2020; 396: 235– 236. doi: 10.1016/S0140-6736(20)31062-X 32711792 · DOI
- [32]Yao X Rushlow DR Inselman JW McCoy RG Thacher TD Behnken EM Bernard ME Rosas SL Akfaly A Misra A. Artificial intelligence-enabled electrocardiograms for identification of patients with low ejection fraction: a pragmatic, randomized clinical trial. Nat Med. 2021; 27: 815– 819. doi: 10.1038/s41591-021-01335-4 33958795 · DOI · Paper page
- [33]Noseworthy PA Attia ZI Brewer LC Hayes SN Yao X Kapa S Friedman PA Lopez-Jimenez F. Assessing and mitigating bias in medical artificial intelligence: the effects of race and ethnicity on a deep learning model for ecg analysis. Circ Arrhythm Electrophysiol. 2020; 13: e007988. doi: 10.1161/CIRCEP.119.007988 32064914 PMC7158877 · DOI · Paper page
Cited by
Sorted by citations
Detecting structural heart disease from electrocardiograms using AI
Poterucha, T. J., Jing, L., Ricart, R. P., Adjei-Mosi, M., Finer, J., Hartzel, D., Kelsey, C., Long, A., Rocha, D., Ruhl, J. A., et al. (2025) · Nature · doi:10.1038/s41586-025-09227-0 · Cited by 116 · see citing works
Self-citationPediatric ECG-Based Deep Learning to Predict Left Ventricular Dysfunction and Remodeling
Mayourian, J., La Cava, W. G., Vaid, A., Nadkarni, G. N., Ghelani, S. J., Mannix, R., Geva, T., Dionne, A., Alexander, M. E., Duong, S. Q., et al. (2024) · Circulation · doi:10.1161/circulationaha.123.067750 · Cited by 59 · see citing works
Dhingra, L. S., Aminorroaya, A., Sangha, V., Pedroso, A. F., Shankar, S. V., Coppi, A., Foppa, M., Brant, L. C. C., Barreto, S. M., Ribeiro, A. L. P., et al. (2025) · Journal of the American College of Cardiology · doi:10.1016/j.jacc.2025.01.030 · Cited by 53 · see citing works
Liang, Y., Sau, A., Zeidaabadi, B., Barker, J., Patlatzoglou, K., Pastika, L., Sieliwonczyk, E., Whinnett, Z., Peters, N. S., Yu, Z., et al. (2025) · European Heart Journal · doi:10.1093/eurheartj/ehaf448 · Cited by 26 · see citing works
Tang, Q., Deng, K., Cui, Y., Liu, H., & Qian, B. (2026) · International Journal of General Medicine · doi:10.2147/ijgm.s610298
“9 Even so, unlike low-LVEF AI-ECG—which already has pragmatic randomized trial evidence—broad SHD screening is still supported mainly by retrospective multicenter validation and nonrandomized prospective or population-based cohorts, and randomized implementation data, hard downstream outcome…”
Influential
- Authors
- Ulloa-Cerna, A. E., Jing, L., Pfeifer, J. M., Raghunath, S., Ruhl, J. A., Rocha, D. B., Leader, J. B., Zimmerman, N., Lee, G., Steinhubl, S. R., Good, C. W., Haggerty, C. M., Fornwalt, B. K., & Chen, R.
- Affiliations
- Department of Translational Data Science and Informatics (A.E.U-C., L.J., J.M.P., S.R., J.A.R., C.W.G., C.M.H., B.K.F., R.C.); Geisinger, Danville, PA.
- Tempus Labs (United States) (US)
- Phenomic Analytics and Clinical Data Core (D.B.R., J.B.L.); Geisinger, Danville, PA.
- Scripps Research Institute (US)
- UPMC Hamot (US)
- Geisinger Medical Center (US)
- Corresponding author
- Chen, Ruijun
- ORCID
- Venue
- Circulation
- Published
- 2022-07-05
- Volume
- 146
- Issue
- 1
- Pages
- 36-47
- ISSN
- 0009-7322 (ISSN-L)
- 1524-4539
- PMC
- PMC9241668
- OpenAlex
- W4229450466
- Cited by
- 101
- References
- 2
- Citations by year
- 2026: 28 · 2025: 28 · 2024: 26 · 2023: 14 · 2022: 5
- FWCI
- 13.68
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- Circulation (publishedVersion, cc-by-nc-nd)
- medRxiv (acceptedVersion)
- PubMed Central (publishedVersion, cc-by-nc-nd)
- PubMed (publishedVersion)
- Topics
- Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
- Health Sciences › Medicine › Cardiology and Cardiovascular Medicine › ECG Monitoring and Analysis
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
- Keywords
- structural heart disease, echocardiography, valvular heart disease, aortic stenosis, mitral stenosis, tricuspid regurgitation, reduced ejection fraction, interventricular septum thickness, risk stratification, high-risk patient identification, electronic health records
- MeSH terms
- Machine Learning, Adult, Echocardiography, Electrocardiography, Heart Diseases, Humans, Retrospective Studies
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
