AI-ECG for Echocardiography Triage in Structural Heart Disease: Evidence, Implementation, and Future Directions

May combine two claims.

Hypertrophic cardiomyopathy, cardiac amyloidosis, and pulmonary hypertension AI-ECG applications remain earlier or pathway-incomplete.

“hypertrophic cardiomyopathy, cardiac amyloidosis, and pulmonary hypertension remain earlier or pathway-incomplete applications.”

• cites [26] Ko, 2020, J Am Coll Cardiol sourcePartly supported

  The Conclusions and Study Limitations sections describe the AI-ECG model for hypertrophic cardiomyopathy as promising but still requiring refinement and external validation before broader screening use. The paper does not address AI-ECG applications for cardiac amyloidosis or pulmonary hypertension, so it supports only the hypertrophic-cardiomyopathy part of the claim. Page numbers were not supplied in our copy. (LLM, conf=0.88)

  Experimental — accuracy not yet measured
• cites [27] Goto, 2022, Circulation sourcePartly supported

  The Discussion (page unavailable) says downstream workflows will inform clinical implementation of the HCM models, supporting the pathway-incomplete characterization for HCM. The paper mentions cardiac amyloidosis only as a condition that can resemble HCM; it does not assess an AI-ECG application for amyloidosis or pulmonary hypertension. (LLM, conf=0.87)

  Experimental — accuracy not yet measured
• cites [28] Siontis, 2024, Euro Heart J Digital Health sourcePartly supported

  The Discussion (page unavailable) says the utility of AI-ECG screening for HCM in asymptomatic people is unknown and that proposed screening groups require further study, supporting the pathway-incomplete characterization for HCM. The paper does not address AI-ECG applications for cardiac amyloidosis or pulmonary hypertension, so it cannot support that part of the claim. (LLM, conf=0.88)

  Experimental — accuracy not yet measured
• cites [29] Sangha, 2025, Nat Cardiovasc Res sourcePartly supported

  The Discussion (page unavailable) says the HCM AI-ECG model needs prospective validation before widespread screening deployment and that a role for universal screening remains to be established. This supports describing the HCM application as pathway-incomplete. The paper does not address the development or implementation status of AI-ECG applications for cardiac amyloidosis or pulmonary hypertension. (LLM, conf=0.88)

  Experimental — accuracy not yet measured
• cites [30] Goto, 2021, Nat Commun sourcePartly supported

  The Discussion says the cardiac amyloidosis ECG and echocardiography models would ideally be coupled with facilitated confirmatory testing, supporting the pathway-incomplete characterization for that application. It names hypertrophic cardiomyopathy and pulmonary arterial hypertension as other rare, difficult-to-detect diseases, but does not assess the maturity or clinical pathways of AI-ECG applications for either. (LLM, conf=0.84)

  Experimental — accuracy not yet measured
• cites [31] Kwon, 2020, J Heart Lung Transplant sourceNot checked — no full text of the source
• cites [32] DuBrock, 2024, Eur Respir J sourcePartly supported

  The Methods describe the pulmonary-hypertension AI-ECG algorithm as investigational, and the Discussion says future studies will assess outcomes for patients it identifies. That supports describing the pulmonary-hypertension application as early or pathway-incomplete. The paper does not establish the same status for hypertrophic cardiomyopathy or cardiac amyloidosis; its Introduction and References mention prior hypertrophic-cardiomyopathy work but do not make that assessment (no page numbers provided). (LLM, conf=0.86)

  Experimental — accuracy not yet measured
• cites [33] Aras, 2023, J Card Fail sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Cardiac amyloidosis is now recognized as an underdiagnosed and clinically important cause of heart failure with preserved ejection fraction, and substantial diagnostic delay remains common.

“Cardiac amyloidosis, once considered rare, is now recognized as an underdiagnosed and clinically important cause of heart failure with preserved ejection fraction, and substantial diagnostic delay remains common.2,6”

• cites [2] AbouEzzeddine, 2021, JAMA Cardiol sourcePartly supported

  The Discussion (page unavailable) reports that clinically recognized transthyretin amyloid cardiomyopathy (ATTR-CM) was 1.3% in the studied HFpEF cohort, versus 6.3% with systematic screening, supporting underrecognition. The Introduction (page unavailable) explains that ATTR-CM can cause an HFpEF syndrome and has specific treatment. The paper does not establish that substantial diagnostic delay remains common; its findings also concern ATTR-CM in older patients with increased ventricular wall thickness, rather than cardiac amyloidosis generally. (LLM, conf=0.86)

  Experimental — accuracy not yet measured
• cites [6] Witteles, 2019, JACC Heart Fail sourcePartly supported

  The Epidemiology of ATTRwt section says transthyretin amyloid cardiomyopathy (ATTR-CM) is overlooked and occurs at relatively high rates among people diagnosed with heart failure with preserved ejection fraction. The Delay to Diagnosis and Misdiagnosis section says its diagnosis is often delayed. These passages support the claim for ATTR-CM, but do not establish the same pattern for cardiac amyloidosis as a whole, which also includes other types. Page numbers were not provided in our copy. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

May combine two claims.

The available evidence for cardiac amyloidosis remains constrained by case-enriched designs, limited external validation, and the low-prevalence challenge intrinsic to real-world deployment.

“However, the available evidence remains constrained by case-enriched designs, limited external validation, and the low-prevalence challenge intrinsic to real-world deployment.30,43”

• cites [30] Goto, 2021, Nat Commun sourcePartly supported

  The Methods describe cases matched to controls, and the Introduction says low disease prevalence places high demands on model performance. However, the paper does not support the claim of limited external validation: the Methods report external ECG validation at two institutions and echocardiography validation at four. Our copy provides section labels but no page numbers. (LLM, conf=0.88)

  Experimental — accuracy not yet measured
• cites [43] Grogan, 2025, J Am Heart Assoc sourceSupported

May combine two claims.

Evidence for AI-ECG detection of pulmonary hypertension and right ventricular dysfunction remains comparatively early and predominantly retrospective, although reported discrimination has been encouraging with AUROCs of approximately 0.86-0.92 across internal and external validation cohorts.

“Evidence for AI-ECG detection of PH and right ventricular dysfunction remains comparatively early. Published studies are still predominantly retrospective, although reported discrimination has been encouraging, with AUROCs of approximately 0.86–0.92 across internal and external validation cohorts.31–33”

• cites [31] Kwon, 2020, J Heart Lung Transplant sourceNot checked — no full text of the source
• cites [32] DuBrock, 2024, Eur Respir J sourcePartly supported

  Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured
• cites [33] Aras, 2023, J Card Fail sourcePartly supported

  Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured

May combine two claims.

In Alexandrino et al., among 286 patients with available reference echocardiography, a two-tier workflow of AI-ECG followed by focused point-of-care cardiac ultrasound increased positive predictive value from 32% to 64%, maintained a negative predictive value of 93%, improved overall diagnostic accuracy from 67% to 88%, and yielded a number-needed-to-screen of 8 to identify 1 patient requiring正式?

“Among 286 patients with available reference echocardiography, the combined strategy increased positive predictive value from 32% to 64% while maintaining a negative predictive value of 93% and improving overall diagnostic accuracy from 67% to 88%, supporting referral enrichment rather than definitive rule-out. Applied to the full pragmatic cohort, this workflow yielded a number-needed-to-screen of”

• cites [35] Alexandrino, 2026, Am Heart J sourceSupported

May combine two claims.

Complementary multicentre work from five Mount Sinai hospitals using 617,338 ECG-echocardiogram pairs showed that ECG-based deep learning can detect both MR and AS, with external-validation AUROCs of 0.81 for MR and 0.86 for AS.

“Complementary multicentre work from five Mount Sinai hospitals using 617,338 ECG–echocardiogram pairs likewise showed that ECG-based deep learning can detect both MR and AS, with external-validation AUROCs of 0.81 for MR and 0.86 for AS.22”

• cites [22] Vaid, 2023, Commun Med sourceSupported

May combine two claims.

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”

• cites [9] Ulloa-Cerna, 2022, Circulation sourceSupported

May combine two claims.

EchoNext extended composite SHD screening to a multicenter, demographically diverse eight-hospital cohort and showed that broad referral-oriented detection can generalize across academic and community settings.

“EchoNext extended composite SHD screening to a multicenter, demographically diverse eight-hospital cohort and showed that broad referral-oriented detection can generalize across academic and community settings.3”

• cites [3] Poterucha, 2025, Nature sourceSupported

May combine two claims.

PRESENT-SHD found that ECG image-based ensemble deep learning achieved AUROC 0.853 in ELSA-Brasil, sensitivity 88%, specificity 62%, and a positive screen was associated with 2- to 4-fold higher incident SHD/HF risk.

“AUROC 0.853 in ELSA-Brasil; sensitivity 88%; specificity 62%; positive screen associated with 2- to 4-fold higher incident SHD/HF risk”

• cites [34] Dhingra, 2025, J Am Coll Cardiol sourceSupported

May combine two claims.

Goto et al. reported an ECG-echocardiography pipeline with C-statistics of 0.85-0.91 for ECG and 0.89-1.00 for echocardiography across multicenter cohorts, and in deployment simulation ECG prescreening improved the positive predictive value of downstream echocardiographic screening from 33% to 74-77% at 67% recall.

“Goto et al subsequently reported a complementary ECG–echocardiography pipeline with C-statistics of 0.85–0.91 for ECG and 0.89–1.00 for echocardiography across multicenter cohorts. In deployment simulation, ECG prescreening improved the positive predictive value of downstream echocardiographic screening from 33% to 74–77% at 67% recall.30”

• cites [30] Goto, 2021, Nat Commun sourcePartly supported

  Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured

May combine two claims.

Alexandrino et al 2026 found that AI-ECG followed by point-of-care cardiac ultrasound had PPV 32% and NPV 94%; adding POCUS increased PPV to 64%, maintained NPV at 93%, and improved accuracy from 67% to 88%.

“AI-ECG PPV 32% and NPV 94%; adding POCUS increased PPV to 64%, maintained NPV at 93%, and improved accuracy from 67% to 88%”

• cites [35] Alexandrino, 2026, Am Heart J sourceSupported

May combine two claims.

Unlike low-LVEF AI-ECG, 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 trials, and validated gatekeeper protocols for safely deferring echocardiography remain unavailable.

“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 trials, and validated gatekeeper protocols for safely deferring echocardiography remain unavailable.14,34”

• cites [14] Yao, 2021, Nat Med sourceNot checked — no full text of the source
• cites [34] Dhingra, 2025, J Am Coll Cardiol sourcePartly supported

  Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured

May combine two claims.

Ko et al. developed the first AI-ECG model for HCM detection, achieving an AUC of 0.96 in an independent testing cohort, with 87% sensitivity and 90% specificity at the prespecified threshold.

“Ko et al at the Mayo Clinic developed the first AI-ECG model for HCM detection, achieving an AUC of 0.96 in an independent testing cohort, with 87% sensitivity and 90% specificity at the prespecified threshold.”

• cites [26] Ko, 2020, J Am Coll Cardiol sourceSupported

Classified as probably not a substantial claim. May combine two claims.

Gatekeeper use requires more than discrimination: reliable calibration, very high sensitivity and negative predictive value in the intended population, prospective pathway-level validation, decision-curve evidence at the selected operating threshold, and explicit mechanisms for clinician override and follow-up.

“Gatekeeper use requires more than discrimination: reliable calibration, very high sensitivity and negative predictive value in the intended population, prospective pathway-level validation, decision-curve evidence at the selected operating threshold, and explicit mechanisms for clinician override and follow-up.36,39,40”

• cites [36] Lin, 2026, EMBO Mol Med sourcePartly supported

  Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured
• cites [39] Oikonomou, 2024, Eur Heart J sourcePartly supported

  Split into 2 atoms — 0 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 1 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured
• cites [40] Palermi, 2025, Eur J Intern Med sourceNot checked — no full text of the source

May combine two claims.

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.”

• Also stated in: § The Undiagnosed Burden of Structural Heart Disease · plain

  “Echocardiography remains the principal confirmatory imaging modality for most forms of structural heart disease, but broad referral strategies are difficult to scale because imaging capacity, expertise, and patient selection all constrain efficiency.3,9”

• cites [3] Poterucha, 2025, Nature sourceSupported
• cites [9] Ulloa-Cerna, 2022, Circulation sourcePartly supported

  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 measured

May combine two claims.

Valvular AI-ECG has moved beyond proof of concept and shows meaningful referral enrichment, but prospective pathway validation remains limited.

“From a deployment perspective, however, the practical conclusion is narrower: valvular AI-ECG has moved beyond proof of concept and shows meaningful referral enrichment, but prospective pathway validation remains limited.19,23”

• cites [19] Cohen-Shelly, 2021, Eur Heart J sourceNot checked — no full text of the source
• cites [23] Segar, 2026, Euro Heart J Digital Health sourcePartly supported

  The Discussion (page unavailable) reports community validation of the aortic-stenosis AI-ECG model and projects that screening higher-risk patients would improve positive predictive value, with positive screens prompting echocardiography. It does not demonstrate referral enrichment in an implemented pathway; that is a proposed application. The Discussion also calls for prospective validation and randomized trials, supporting the qualification that pathway validation remains limited. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

May combine two claims.

The Columbia-led ValveNet study in 77,163 patients achieved AUROC of 0.88 for AS, 0.77 for AR, and 0.83 for mitral regurgitation, with similar overall accuracy in external validation.

“In 77,163 patients, the model achieved area under receiver operating characteristic curve (AUROC) of 0.88 for AS, 0.77 for AR, and 0.83 for mitral regurgitation (MR), with similar overall accuracy in external validation.21”

• cites [21] Elias, 2022, J Am Coll Cardiol sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

EchoNext explicitly framed AI-ECG deployment along a spectrum ranging from a safety-net strategy that triggers additional echocardiography to a later-stage gatekeeper strategy intended to reduce unnecessary imaging.

“EchoNext explicitly framed AI-ECG deployment along a spectrum ranging from a “safety-net” strategy that triggers additional echocardiography to a later-stage “gatekeeper” strategy intended to reduce unnecessary imaging.3”

• cites [3] Poterucha, 2025, Nature sourceSupported

May combine two claims.

The Taiwanese ARISE economic analysis evaluated an inpatient AI-ECG mortality-alert strategy and reported an ICER of $59,500 per death averted; it should be regarded as adjacent evidence for AI-enabled cardiovascular alerting rather than direct corroboration of low-LVEF screening cost-effectiveness.

“By contrast, the Taiwanese ARISE economic analysis evaluated an inpatient AI-ECG mortality-alert strategy and reported an ICER of $59,500 per death averted; it should therefore be regarded as adjacent evidence for AI-enabled cardiovascular alerting rather than direct corroboration of low-LVEF screening cost-effectiveness.58”

• cites [58] Hsieh, 2025, NPJ Digit Med sourceSupported

May combine two claims.

In aortic stenosis, patients with positive AI-ECG results but no moderate-to-severe AS at baseline had a higher long-term risk of subsequently developing moderate-to-severe AS, and recent community validation showed a fourfold higher risk of future AS hospitalization among false-positive screens.

“Analogous findings have been reported in aortic stenosis: patients with positive AI-ECG results but no moderate-to-severe AS at baseline had a higher long-term risk of subsequently developing moderate-to-severe AS, and recent community validation likewise showed a fourfold higher risk of future AS hospitalization among false-positive screens.19,23”

• cites [19] Cohen-Shelly, 2021, Eur Heart J sourceNot checked — no full text of the source
• cites [23] Segar, 2026, Euro Heart J Digital Health sourceSupported

May combine two claims.

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”

• cites [9] Ulloa-Cerna, 2022, Circulation sourceSupported

May combine two claims.

Effective disease-modifying therapies are now available for selected SHD phenotypes-including obstructive HCM and transthyretin amyloid cardiomyopathy-so the clinical value of timely diagnosis has increased.

“Because effective disease-modifying therapies are now available for selected structural heart disease phenotypes—including obstructive HCM and transthyretin amyloid cardiomyopathy—the clinical value of timely diagnosis has increased.7,8”

• cites [7] Olivotto, 2020, Lancet sourcePartly supported

  The Implications section (p. 2) describes mavacamten as a targeted therapy for obstructive HCM and reports clinically meaningful improvements; the Discussion (p. 9) reports benefits over placebo. The trial does not establish that treatment modifies the disease’s natural history or that it was already available in clinical practice. The sentence’s conclusion about timely diagnosis is the citing authors’ inference, and transthyretin amyloid cardiomyopathy appears to be covered by the other citation. (LLM, conf=0.84)

  Experimental — accuracy not yet measured
• cites [8] Maurer, 2018, N Engl J Med sourceSupported

May combine two claims.

In a multinational Circulation study, ECG and echocardiography models trained across three academic centers in the United States and Japan showed substantially better cross-site generalizability than single-institution models and improved discrimination of HCM from other causes of hypertrophy.

“In a multinational Circulation study, ECG and echocardiography models trained across three academic centers in the United States and Japan showed substantially better cross-site generalizability than single-institution models and improved discrimination of HCM from other causes of hypertrophy.27”

• cites [27] Goto, 2022, Circulation sourceSupported

May combine two claims.

In PRESENT-SHD, an image-based composite AI-ECG model validated across four U.S. hospitals generalized to the prospective ELSA-Brasil cohort, achieving an AUROC of 0.853 with 88% sensitivity and 62% specificity; a positive screen was associated with a 2- to 4-fold higher risk of incident SHD or heart failure across external cohorts.

“In PRESENT-SHD, an image-based composite AI-ECG model validated across four U.S. hospitals also generalized to the prospective population-based ELSA-Brasil cohort, achieving an AUROC of 0.853 with 88% sensitivity and 62% specificity; a positive screen was further associated with a 2- to 4-fold higher risk of incident SHD or heart failure across external cohorts.34”

• cites [34] Dhingra, 2025, J Am Coll Cardiol sourceSupported

May combine two claims.

The main failure modes of safety-net AI-ECG deployment are excess downstream testing, incomplete follow-up, and alert fatigue.

“Its main failure modes are excess downstream testing, incomplete follow-up, and alert fatigue.”

• cites [16] Rushlow, 2022, Mayo Clin Proc sourcePartly supported

  The Effectiveness of Detecting Low EF section reports that low-adopting clinicians ordered echocardiograms after 43.0% of positive AI-ECG results, supporting a concern about incomplete follow-up. The Discussion says high adopters were not indiscriminately ordering more tests, and the paper does not identify excess downstream testing or alert fatigue as main failure modes. The study was conducted in a Midwest health system; it does not establish these as the main failure modes of safety-net deployment. Page numbers were not available in our copy. (LLM, conf=0.85)

  Experimental — accuracy not yet measured
• cites [38] Lopez-Jimenez, 2025, Am Heart J Plus sourceCould not verify

  The Methods (Data Collection and Endpoints; page not provided) describe tracking echocardiogram orders, dismissed notifications, and clinical outcomes. The Discussion (page not provided) identifies low software adoption as a possible limitation, but our copy does not identify excess downstream testing, incomplete follow-up, or alert fatigue as the main failure modes of safety-net deployment. Searched for those terms and related descriptions of unnecessary testing and missed follow-up. (LLM, conf=0.72)

  Experimental — accuracy not yet measured
• cites [52] Meunier, 2023, Ann Fam Med sourcePartly supported

  Table 3 lists alert fatigue as a barrier to clinical decision support systems in primary care. The cited review does not establish that excess downstream testing or incomplete follow-up are failure modes, nor does it identify these three as the main failure modes of safety-net AI-ECG deployment. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

May combine two claims.

Average subgroup performance is not sufficient: clinically relevant degradation may still emerge in specific age-race-sex strata.

“Even so, average subgroup performance is not sufficient: clinically relevant degradation may still emerge in specific age-race-sex strata, and fairness in medical AI is shaped by data acquisition and workflow as well as by model architecture.”

• cites [66] Kaur, 2024, Circ Heart Fail sourceSupported
• cites [67] Chen, 2023, Nat Biomed Eng sourcePartly supported

  The ‘Healthcare disparities arising from dataset shift’ section says demographic and image-acquisition differences can cause disparate performance at the subgroup level; the ‘Image acquisition and measurement variation’ section also links acquisition protocols to model performance. These passages support the concern about subgroup degradation and factors beyond model architecture, but they do not specifically establish that average subgroup results conceal clinically relevant degradation in combined age–race–sex strata (pages unavailable in our copy). (LLM, conf=0.83)

  Experimental — accuracy not yet measured

May combine two claims.

When AI-ECG is applied to a routine ECG obtained for another purpose, the output functions as an incidental risk signal and may trigger cascades of follow-up testing with uncertain net benefit.

“When AI-ECG is applied to a routine ECG obtained for another purpose, the output functions as an incidental risk signal and may trigger cascades of follow-up testing with uncertain net benefit.”

• cites [63] Ganguli, 2019, JAMA Netw Open sourcePartly supported

  The Introduction (page not provided) says incidental findings can trigger follow-up of uncertain value and gives an ECG anomaly leading to a stress test and catheterization as an example. The paper does not address AI-ECG outputs or establish that applying AI to an ECG obtained for another purpose produces an incidental risk signal. (LLM, conf=0.88)

  Experimental — accuracy not yet measured

May combine two claims.

Interpretability tools may make model behavior more inspectable, but they do not by themselves resolve questions of accountability, acceptable error, or downstream clinical action in opportunistic screening.

“Interpretability tools may make model behavior more inspectable, but they do not by themselves resolve questions of accountability, acceptable error, or downstream clinical action in opportunistic screening.”

• cites [36] Lin, 2026, EMBO Mol Med sourceSupported
• cites [62] Lewin, 2024, Can J Cardiol sourcePartly supported

  The Transparency and explainability section says interpretability tools can clarify aspects of model decisions, while Responsibility and accountability says responsibility for those decisions remains with clinicians and other parties. Promoting human health and the public interest also discusses balancing misdiagnosis and overdiagnosis risks against follow-up and actionability for incidental findings. The copy does not specifically frame these points as acceptable error in opportunistic screening. (LLM, conf=0.82)

  Experimental — accuracy not yet measured

May combine two claims.

In low- and middle-income settings, the appeal of AI-ECG lies not in an already mature external-validation literature, but in the possibility of adding a scalable, low-cost triage layer upstream of formal imaging.

“In such settings, the appeal of AI-ECG lies not in an already mature external-validation literature, but in the possibility of adding a scalable, low-cost triage layer upstream of formal imaging.”

• cites [64] Mbanze, 2025, Heart sourceNot checked — no full text of the source
• cites [65] Myhre, 2024, Lancet Digit Health sourcePartly supported

  The Introduction proposes AI-enabled ECG as an inexpensive screening tool in low- and middle-income countries, with patients identified for subsequent AI-guided diagnostic echocardiography. It does not assess whether the external-validation literature is mature, so that contrast is not supported by this source. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S2 (checkable).

A safety-net model may be useful when positive AI-ECG results enrich echocardiography yield and lead to timely confirmatory evaluation.

“A safety-net model may be useful when positive AI-ECG results enrich echocardiography yield and lead to timely confirmatory evaluation.”

• cites [16] Rushlow, 2022, Mayo Clin Proc sourceSupported
• cites [38] Lopez-Jimenez, 2025, Am Heart J Plus sourcePartly supported

  The Discussion (page not provided) reports that providing AI-ECG results in the EAGLE study increased new low-LVEF diagnoses despite similar echocardiogram-ordering rates, which is consistent with improved diagnostic yield. The Methods > Endpoints section describes echocardiographic evaluation within 90 days, but this design paper does not establish that positive results led to timely confirmatory evaluation or assess a safety-net model. (LLM, conf=0.83)

  Experimental — accuracy not yet measured
• cites [52] Meunier, 2023, Ann Fam Med sourceCould not verify

  The Quantitative Synthesis discusses clinical decision support systems used for test ordering and cardiovascular care, but does not address AI-ECG results or echocardiography yield. Searches for AI-ECG, echocardiography, safety-netting, and confirmatory follow-up found no passage supporting the proposed model. (LLM, conf=0.77)

  Experimental — accuracy not yet measured

May combine two claims.

The strongest prospective African data come from a pragmatic randomized trial in Nigeria, which showed that AI-assisted screening for peripartum cardiomyopathy can be operationalized in hospital-based care.

“The strongest prospective African data come from a pragmatic randomized trial in Nigeria, which showed that AI-assisted screening for peripartum cardiomyopathy can be operationalized in hospital-based care;”

• cites [15] Adedinsewo, 2024, Nat Med sourceSupported

May combine two claims.

Fairness is not only a model-performance issue; it is also shaped by ECG acquisition quality, care setting, access to confirmatory echocardiography, and linkage to disease-specific treatment.

“Fairness is not only a model-performance issue; it is also shaped by ECG acquisition quality, care setting, access to confirmatory echocardiography, and linkage to disease-specific treatment.”

• cites [60] Zhou, 2025, NPJ Digit Med sourcePartly supported

  The Discussion (page unavailable) warns that institutional differences in ECG acquisition protocols may limit generalizability. The Methods and Results (page unavailable) assess performance across care settings, with small differences between them. But the paper’s fairness analysis concerns model performance by age and sex; the cited passages do not connect fairness to access to confirmatory echocardiography or disease-specific treatment. (LLM, conf=0.83)

  Experimental — accuracy not yet measured
• cites [61] Davis, 2025, J Am Med Inform Assoc sourcePartly supported

  The Introduction and Discussion distinguish subgroup fairness from overall model performance and describe fairness drift in evolving clinical environments. But this study concerns surgical risk models; our copy does not support the attributed roles of ECG acquisition quality, care setting, access to confirmatory echocardiography, or linkage to disease-specific treatment. Page numbers were not provided in our copy. (LLM, conf=0.86)

  Experimental — accuracy not yet measured
• cites [62] Lewin, 2024, Can J Cardiol sourcePartly supported

  The “Minimisation of bias” section says fairness depends on more than model performance, including historically unequal access to care. “The case for AI: The promises and the peril” discusses access to echocardiographic screening in remote communities. Neither section links fairness specifically to ECG acquisition quality, care setting, confirmatory echocardiography, or access to disease-specific treatment; no page numbers are available in our copy. (LLM, conf=0.83)

  Experimental — accuracy not yet measured
• cites [63] Ganguli, 2019, JAMA Netw Open sourceCould not verify

  The Introduction (page unavailable) discusses incidental findings that trigger further testing and treatment, including an ECG anomaly followed by a stress test and catheterization. Neither that passage nor the searched passages address fairness, ECG acquisition quality, care setting, access to confirmatory echocardiography, or linkage to disease-specific treatment. (LLM, conf=0.76)

  Experimental — accuracy not yet measured

May combine two claims.

Emerging evidence supports longitudinal risk tracking in aortic stenosis and surveillance for treatment-related left ventricular dysfunction in cardio-oncology.

“with emerging evidence supporting longitudinal risk tracking in aortic stenosis23 and surveillance for treatment-related left ventricular dysfunction in cardio-oncology.18”

• cites [23] Segar, 2026, Euro Heart J Digital Health sourceSupported
• cites [18] Jacobs, 2024, Eur J Prev Cardiol sourceSupported

May combine two claims.

Equitable AI-ECG deployment depends not only on broadly similar performance across demographic subgroups, but also on transportability across institutions, acquisition workflows, and care environments.

“Equitable AI-ECG deployment depends not only on broadly similar performance across demographic subgroups, but also on transportability across institutions, acquisition workflows, and care environments.”

• cites [3] Poterucha, 2025, Nature sourcePartly supported

  Model development and validation reports broadly similar performance by race or ethnicity and sex, and stable performance across NYP hospitals and clinical contexts. External validation found a 5–7% AUROC drop at three other institutions, underscoring the importance of transportability. The copy does not establish performance across different ECG acquisition workflows, so that part of the attributed claim is unsupported. (LLM, conf=0.84)

  Experimental — accuracy not yet measured
• cites [27] Goto, 2022, Circulation sourcePartly supported

  The Discussion (page unavailable) reports that single-institution ECG models can perform poorly on external data and suggests that institutional differences may reflect vendor signal processing. The Results, under “Federated Learning Improves the Generalizability of the HCM-ECG Model” (page unavailable), report external validation and an outpatient-setting analysis. The paper does not establish broadly similar performance across demographic subgroups or test transportability across acquisition workflows and care environments generally. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

May combine two claims.

ECG foundation models may improve transportability and label efficiency, particularly for lower-prevalence phenotypes, but their translational value will depend on whether they improve calibration, robustness, and pathway performance rather than benchmark metrics alone.

“ECG foundation models may improve transportability and label efficiency, particularly for lower-prevalence phenotypes, but their translational value will depend on whether they improve calibration, robustness, and pathway performance rather than benchmark metrics alone.68”

• cites [68] Nolin-Lapalme, 2026, Eur Heart J sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Asymptomatic left ventricular systolic dysfunction is estimated to affect approximately 3-6% of the general population and is associated with a substantially increased risk of progression to overt heart failure.

“Asymptomatic left ventricular systolic dysfunction is estimated to affect approximately 3–6% of the general population and is associated with a substantially increased risk of progression to overt heart failure.4”

• cites [4] Wang, 2003, Ann Intern Med sourceNot checked — no full text of the source

May combine two claims.

Gatekeeper use of AI-ECG remains investigational and requires prospective pathway-level validation, calibration, and operational safeguards before routine imaging deferral can be justified.

“Gatekeeper use remains investigational and requires prospective pathway-level validation, calibration, and operational safeguards before routine imaging deferral can be justified.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Aortic stenosis is the most prevalent valvular heart disease in developed countries, often remains asymptomatic for years, but prognosis worsens sharply once symptoms develop.

“Aortic stenosis, the most prevalent valvular heart disease in developed countries, often remains asymptomatic for years, but prognosis worsens sharply once symptoms develop.5”

• cites [5] Carabello, 2009, Lancet sourceNot checked — no full text of the source

May combine two claims.

On current evidence, low LVEF most clearly satisfies the first two criteria (referral enrichment and prospective pathway validation) and has moved furthest toward the third (workflow readiness), making it the benchmark safety-net indication for near-term deployment.

“On current evidence, low LVEF most clearly satisfies the first two criteria and has also moved furthest toward the third, making it the benchmark safety-net indication for near-term deployment.”

May combine two claims.

In a Taiwanese prospective outpatient study of opportunistic AI-ECG screening for asymptomatic left ventricular dysfunction, the strategy yielded a negative incremental cost-effectiveness ratio of -$7,439 for 65-year-old patients and remained cost-saving across age groups in sensitivity analyses.

“In a Taiwanese prospective outpatient study of opportunistic AI-ECG screening for asymptomatic left ventricular dysfunction, the strategy yielded a negative incremental cost-effectiveness ratio (ICER) of -$7,439 for 65-year-old patients and remained cost-saving across age groups in sensitivity analyses.17”

• cites [17] Liu, 2024, Can J Cardiol sourceNot checked — no full text of the source

May combine two claims.

Current evidence still supports referral prioritization rather than imaging deferral, and the main trade-off remains reduced lesion-specific interpretability.

“However, current evidence still supports referral prioritization rather than imaging deferral, and the main trade-off remains reduced lesion-specific interpretability.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The prevalence of aortic stenosis rises steeply with age, affecting approximately 1-2% of adults older than 65 years and about 12% of those older than 75 years.

“The prevalence of AS rises steeply with age, affecting approximately 1–2% of adults older than 65 years and about 12% of those older than 75 years, while definitive diagnosis and severity grading still rely on echocardiography.41”

• cites [41] Otto, 2024, JAMA sourceNot checked — no full text of the source

May combine two claims.

AI-enabled detection improved case finding without materially increasing overall echocardiography utilization.

“Importantly, AI-enabled detection improved case finding without materially increasing overall echocardiography utilization, suggesting referral redirection rather than indiscriminate imaging expansion.”

• cites [14] Yao, 2021, Nat Med sourceNot checked — no full text of the source

Classified as probably not a substantial claim.

After deduplication and staged screening of titles, abstracts, and full texts, 26 core evidence publications were retained for qualitative evidence synthesis.

“After deduplication and staged screening of titles, abstracts, and full texts, 26 core evidence publications were retained for qualitative evidence synthesis (Figure 1).”

May combine two claims.

A landmark 2019 study demonstrated that AI-ECG could identify reduced LVEF from routine 12-lead recordings with strong discriminatory performance, establishing the principle that an inexpensive and widely available test could function as an upstream screening instrument for abnormalities ordinarily confirmed by echocardiography.

“A landmark 2019 study demonstrated that artificial intelligence (AI)-ECG could identify reduced left ventricular ejection fraction (LVEF) from routine 12-lead recordings with strong discriminatory performance, establishing the principle that an inexpensive and widely available test could function as an upstream screening instrument for abnormalities ordinarily confirmed by echocardiography.12”

• cites [12] Attia, 2019, Nat Med sourceNot checked — no full text of the source

May combine two claims.

Calibration is indispensable because a model may retain a high AUROC yet still provide unreliable absolute risk estimates and thereby misdirect referral decisions.

“Calibration is equally indispensable: a model may retain a high AUROC yet still provide unreliable absolute risk estimates and thereby misdirect referral decisions.”

Probably not a substantial claim: may not pass S3 (load-bearing).

HCM has an estimated prevalence of approximately 1 in 500 in the general population.

“HCM, with an estimated prevalence of approximately 1 in 500 in the general population, often remains clinically unrecognized and is associated with sudden cardiac death, including in younger individuals.42”

• cites [42] Maron, 2013, Lancet sourceNot checked — no full text of the source

May combine two claims.

Overall, the evidence base is clinically promising but uneven in translational maturity.

“Overall, the evidence base is clinically promising but uneven in translational maturity.”

May combine two claims.

Cardiac amyloidosis has low prevalence, which makes real-world positive predictive value intrinsically challenging.

“Cardiac amyloidosis represents a distinct screening scenario because its low prevalence makes real-world positive predictive value (PPV) intrinsically challenging.”

Probably not a substantial claim: may not pass S2 (checkable).

This AI-ECG-to-POCUS pathway should still be interpreted as early workflow-feasibility evidence rather than implementation-ready evidence.

“Even so, this pathway should still be interpreted as early workflow-feasibility evidence rather than implementation-ready evidence.”

May combine two claims.

Among current AI-ECG applications, the economic evidence is most developed for low-LVEF screening and aligns with an additive safety-net use case rather than a validated gatekeeper strategy.

“Among current AI-ECG applications, the economic evidence is most developed for low-LVEF screening, and importantly it aligns with an additive safety-net use case rather than a validated gatekeeper strategy.”

May combine two claims.

Safety-net screening is the most mature near-term role for AI-ECG, whereas gatekeeper triage remains aspirational pending stronger prospective evidence.

“By this standard, safety-net screening is the most mature near-term role, whereas gatekeeper triage remains aspirational pending stronger prospective evidence.”

May combine two claims.

Attia et al 2019 found that AI-ECG for reduced LVEF achieved AUC 0.93 and NPV 98.7%, and AI-positive/echo-negative patients had higher future ventricular dysfunction risk.

“AUC 0.93; NPV 98.7%; AI-positive/echo-negative patients had higher future ventricular dysfunction risk”

• cites [12] Attia, 2019, Nat Med sourceNot checked — no full text of the source

May combine two claims.

The EAGLE trial increased new diagnoses of low EF from 1.6% in the control arm to 2.1% in the intervention arm (OR 1.32, 95% CI 1.01-1.61; P = 0.007), equivalent to a 32% relative increase over usual care.

“The trial met this endpoint, increasing new diagnoses from 1.6% in the control arm to 2.1% in the intervention arm (OR 1.32, 95% CI 1.01–1.61; P = 0.007), equivalent to a 32% relative increase over usual care.14”

• cites [14] Yao, 2021, Nat Med sourceNot checked — no full text of the source

May combine two claims.

AI-ECG is most likely to affect care not by replacing echocardiography, but by improving patient selection, prioritizing limited imaging capacity, and identifying patients who might otherwise remain undiagnosed.

“The real-world relevance is that AI-ECG is most likely to affect care not by replacing echocardiography, but by improving patient selection, prioritizing limited imaging capacity, and identifying patients who might otherwise remain undiagnosed.”

Probably not a substantial claim: may not pass S2 (checkable).

The transition from retrospective algorithm validation to clinical deployment requires that AI-ECG outputs be incorporated into EHR-based clinical workflows rather than remain stand-alone analytic results.

“The transition from retrospective algorithm validation to clinical deployment requires that AI-ECG outputs be incorporated into EHR-based clinical workflows rather than remain stand-alone analytic results.”

May combine two claims.

In the Mayo Clinic study, a convolutional neural network detected moderate-to-severe AS with an AUC of 0.85, and patients with positive AI-ECG results but no moderate-to-severe AS at baseline had a 2.18-fold higher long-term risk of subsequently developing clinically significant AS.

“In the Mayo Clinic study, a convolutional neural network detected moderate-to-severe AS with an AUC of 0.85; importantly, patients with positive AI-ECG results but no moderate-to-severe AS at baseline had a 2.18-fold higher long-term risk of subsequently developing clinically significant AS.19”

• cites [19] Cohen-Shelly, 2021, Eur Heart J sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S2 (checkable).

Across indications, the decisive issue is clinical deployability rather than discrimination alone.

“Across indications, the decisive issue is clinical deploy ability rather than discrimination alone.”

May combine two claims.

At present, the economic literature supports the plausibility of downstream referral enrichment, but not the stronger claim that broader SHD gatekeeping can safely reduce imaging at the pathway level.

“At present, the economic literature supports the plausibility of downstream referral enrichment, but not the stronger claim that broader SHD gatekeeping can safely reduce imaging at the pathway level.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The evidence reviewed in this paper has primarily addressed AI-ECG as a tool for initial disease detection.

“The evidence reviewed in this paper has primarily addressed AI-ECG as a tool for initial disease detection.”

Probably not a substantial claim: may not pass S2 (checkable).

Unresolved transportability across input pathways is a reason to favor safety-net use over gatekeeper reliance for the present deployment question.

“For the present deployment question, unresolved transportability across input pathways is another reason to favor safety-net use over gatekeeper reliance.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

Multimodal integration is most likely to matter when it reduces referral uncertainty in clinically ambiguous cases or strengthens the safety of downstream triage decisions, not simply when it produces incremental retrospective gains in discrimination.

“Multimodal integration is most likely to matter when it reduces referral uncertainty in clinically ambiguous cases or strengthens the safety of downstream triage decisions, not simply when it produces incremental retrospective gains in discrimination.”

May combine two claims.

If implemented with appropriate safeguards, AI-ECG may improve clinical care by identifying otherwise-missed SHD, prioritizing limited echocardiography capacity, and accelerating referral to disease-specific treatment pathways.

“If implemented with appropriate safeguards, AI-ECG may improve clinical care by identifying otherwise-missed SHD, prioritizing limited echocardiography capacity, and accelerating referral to disease-specific treatment pathways.”