What does this research mean for the field?

The AI-enhanced ECG model achieved an AUC of 0.982 and sensitivity of 89.4% for detecting LVEF <50% in 7,225 participants from the Framingham Heart Study. Novelty: ClaimNovelty.CONFIRMATORY. Consensus alignment: ConsensusAlignment.SUPPORTS_CONSENSUS.

What question did this study set out to answer?

To validate an AI-enhanced ECG model for detecting left ventricular systolic dysfunction.

February 8, 2026

AI-enhanced ECG screening for LV systolic dysfunction in the Framingham Heart Study

Q: What are the key findings of this study?

AI-enhanced ECG detected LVEF <50% with AUC 0.982, sensitivity 89.4%, specificity 97.2%, and NPV 99.8% in 7,225 Framingham Heart Study participants.

Q: What question did this study set out to answer?

To validate an AI-enhanced ECG model for detecting left ventricular systolic dysfunction.

Key Result

AI-enhanced ECG detected LVEF <50% with AUC 0.982, sensitivity 89.4%, specificity 97.2%, and NPV 99.8% in 7,225 Framingham Heart Study participants.

Key Points

To validate an AI-enhanced ECG model for detecting left ventricular systolic dysfunction.
Retrospective analysis of participants from the Framingham Heart Study with ECG and echocardiographic data
Assessment of left ventricular ejection fraction (LVEF) using two AI models
Evaluation of model performance through AUC, sensitivity, specificity, PPV, and NPV using echocardiography as standard.
AI-ECG model achieved an AUC of 0.982 for detecting LVEF <50% and 0.977 for LVEF ≤40%
Sensitivity was 89.4% for LVEF <50% and 90.7% for LVEF ≤40%
Specificity was 97.2% for LVEF <50% and 97.1% for LVEF ≤40%
NPV was very high, at 99.8% for LVEF <50% and 99.9% for LVEF ≤40%.

Structured PICO

Does an AI-enhanced 12-lead ECG model accurately detect left ventricular systolic dysfunction in a community-based cohort?

Population

7,225 participants (9,101 visits) from the Framingham Heart Study cohort with paired ECG and echocardiographic data, mean age 62.2 ± 12.6 years, 46% male.

Intervention

AI-enhanced 12-lead ECG model for detecting LVEF <50% and LVEF ≤40%

Comparator

Echocardiographically measured ejection fraction (reference standard)

Outcome

Model performance for detecting LVEF <50% and LVEF ≤40%, evaluated through area under the receiver operating characteristic curve (AUC), sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV)surrogate

An AI-enhanced 12-lead ECG model demonstrated high accuracy (AUC >0.97) for detecting left ventricular systolic dysfunction in a large community-based cohort, highlighting its potential as a scalable screening tool.

Main Result

Absolute Event Rate: 0% vs 0%

Abstract

Abstract Background Left ventricular (LV) systolic dysfunction is a major contributor to cardiovascular morbidity and mortality, yet it often remains undiagnosed until advanced stages. Artificial intelligence (AI) applied to 12-lead electrocardiography (ECG) offers a promising, noninvasive method for rapid screening for LV systolic dysfunction on a population level, however its performance in community-based settings has not been validated. Purpose To validate the performance of an AI-enhanced ECG model for detecting LV systolic dysfunction in the large, community-based cohort of the Framingham Heart Study. Methods Participants from the Framingham Heart Study who underwent routine ECG and echocardiographic examinations during scheduled study visits were included in this retrospective analysis. LV systolic dysfunction was assessed using two AI models, detecting LVEF 50% and LVEF ≤40% on 12-lead ECG. Model performance was evaluated through area under the receiver operating characteristic curve (AUC), sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) using echocardiographically measured EF as the reference standard. Results A total of 7,225 participants (9101 visits, 46% male, mean age 62.2 ± 12.6 years) from the Framingham Heart Study cohort with paired ECG and echocardiographic data were analyzed. The prevalence of LVEF 50% and LVEF ≤40% assessed echocardiographically was 1.5% and 0.6%, respectively. For detecting LVEF 50%, the AI-enhanced ECG model achieved an AUC of 0.982 (95% CI, 0.968-0.99), sensitivity 89.4% (95% CI, 83.7-94.3%), specificity 97.2% (95% CI, 96.9-97.6%), PPV 32.2% (95% CI, 27.4-37.1%), and NPV 99.8% (95% CI, 99.7-99.9%). For detecting LVEF ≤40%, the model yielded an AUC of 0.977 (95% CI, 0.946-0.994), sensitivity 90.7% (95% CI, 82.1-98%), specificity 97.1% (95% CI, 96.8-97.5%), PPV 15.9% (95% CI, 12-20.1%), and NPV 99.9% (95% CI, 99.9-100%). Model performance remained consistent across subgroups stratified by age and sex. Conclusions This study provides validation of an AI-ECG model for detecting reduced LVEF in a large, well-characterized, community-based cohort. The high accuracy and ease of application suggest that AI-augmented ECG analysis may serve as an effective, scalable screening tool for early detection of LV dysfunction in the general population.ROC curve on Framingham Heart Study data

Bookmark