Development and external validation of a contrastive learning foundation model for ECG-based prediction of cardiovascular diseases and outcomes
Key result
The ECG-CLIP foundation model outperformed comparator models in low-label settings, achieving an AUC of 0.910 (95% CI 0.903-0.916) vs 0.884 for acute myocardial infarction detection.
Why the study?
Traditional machine learning approaches for ECG assessment are limited in diagnostic and predictive potential due to over-reliance on task-specific labels.
Does the ECG-CLIP foundation model improve the detection and prediction of cardiovascular diseases and outcomes from ECGs compared to supervised baseline and signal-only foundation models?
Key Points
- To develop and externally validate ECG-CLIP, a multimodal foundation model combining electrocardiogram signals and clinician reports for cardiovascular disease diagnosis and outcome prediction.
- Pretrained ECG-CLIP using masked signal reconstruction and text-signal contrastive learning on over 1.7 million ECGs paired with clinical reports from 542,288 patients.
- Externally validated performance on the MIMIC-IV dataset (>800,000 ECGs) across disease detection, disease prediction, and mortality forecasting tasks versus supervised and signal-only baseline models.
- ECG-CLIP matched the diagnostic performance of the best-performing comparator models while requiring an average of 90.8% less labeled training data.
- Under low-label training with only 10 positive labels, ECG-CLIP significantly outperformed runner-up models in acute myocardial infarction detection (AUC 0.910 [95% CI 0.903–0.916] vs 0.884 [0.876–0.892]), cardiac amyloidosis detection (0.790 [0.767–0.812] vs 0.777 [0.754–0.799]), hypertrophic cardiomyopathy detection (0.772 [0.751–0.793] vs 0.754 [0.733–0.774]), and atrial fibrillation prediction (0.777 [0.773–0.781] vs 0.765 [0.761–0.769]).
Study Design
Type
Observational (n=542,288)
Multicenter
Yes
Structured PICO
Does the ECG-CLIP foundation model improve the detection and prediction of cardiovascular diseases and outcomes from ECGs compared to supervised baseline and signal-only foundation models?
P
Population
542,288 patients with over 1.7 million ECGs and paired reports used for model training, with external validation on over 800,000 ECGs.
E
Exposure
ECG-CLIP, a contrastive learning foundation model combining ECG data with textual annotations
C
Comparator
Two supervised baseline models, one non-ECG foundation model, and three ECG signal-only foundation models (ECGFounder, DeepECG-SL, and DeepECG-SSL)
O
Outcome
Area under the receiver operating characteristic curve (AUC) for cardiovascular disease detection (acute myocardial infarction, cardiac amyloidosis, hypertrophic cardiomyopathy), cardiovascular disease prediction (atrial fibrillation from normal sinus rhythm), and adverse health outcome prediction (30-day emergency department and post-surgical mortality, and 3-year onset of chronic kidney disease and type 2 diabetes)surrogate
Main Result
Effect estimate: AUC 0.910 (95% CI 0.903-0.916)
Absolute Event Rate: 0.91% vs 0.884%
A novel contrastive learning foundation model (ECG-CLIP) combining ECG signals with textual annotations achieves high accuracy and label efficiency for detecting and predicting cardiovascular diseases.
Key expert perspectives
Captured external expert commentary on this paper, strongest first. Original sources are linked where available.
“For me personally, it's exciting that we can develop algorithms with the potential to really help clinicians and cardiologists.”
“We found that ECG-CLIP was better at detecting and predicting cardiovascular diseases, particularly in cases where there was much less data. This may be particularly useful in cases like rare diseases, where there are only a dozen or so positive examples of well-labeled ECGs that can be used for training the model.”
“The ECG has remarkably rich data that expert humans can't see. Our new paper @LancetDigitalH out today shows the ECG prediction of kidney disease, diabetes, atrial fibrillation, no less detection of hard…”
Implication
May aid label-efficient AMI detection on ECG; leaves open prospective validation before clinical adoption.
Abstract
Background Automated electrocardiogram (ECG) assessment tools to assist clinicians in diagnosis have improved substantially over the past decade; however, the full diagnostic and predictive potential of ECGs is limited when using traditional machine learning approaches because of an over-reliance on task-specific labels. We aimed to develop a large-scale ECG foundation model, ECG-CLIP, with improved generalisability and clinical applicability for ECG analysis. Methods This study used data from the Scripps Health GE MUSE system, a large-scale, retrospective dataset containing more than 1·7 million ECGs and paired clinician-overread ECG reports from 542 288 patients, collected in-clinic between Jan 15, 2008, and Jan 15, 2019. ECG-CLIP was pretrained in a two-stage pretraining framework consisting of masked ECG reconstruction and contrastive learning combining ECG data with textual annotations. MIMIC-IV, an independent, external dataset with more than 800 000 ECGs, was used to evaluate ECG-CLIP in cardiovascular disease detection (acute myocardial infarction, cardiac amyloidosis, and hypertrophic cardiomyopathy), cardiovascular disease prediction (atrial fibrillation from normal sinus rhythm), and adverse health outcome prediction (30-day emergency department and post-surgical mortality and 3-year onset of chronic kidney disease and type 2 diabetes) tasks. We compared the performance of ECG-CLIP against that of two supervised baseline models and one non-ECG foundation model, as well as three ECG signal-only foundation models: ECGFounder, DeepECG-SL, and DeepECG-SSL. Performance was evaluated using a five-fold cross-validation framework and reported as the area under the receiver operating characteristic curve (AUC). We assessed label efficiency by varying the number of positive labels in the training set of each downstream task. Findings ECG-CLIP consistently showed superior performance across all detection and prediction tasks compared with the supervised baseline and non-ECG foundation models, as well as a high label efficiency, reaching the same AUC in cardiovascular disease detection tasks as the best-performing comparator model with an average of 90·8% less training data. Compared with ECG signal-only foundation models, ECG-CLIP showed superior performance in low-label settings. When only ten positive labels were present in the dataset, ECG-CLIP significantly outperformed the runner-up model on tasks including acute myocardial infarction detection (AUC 0·910 [95% CI 0·903–0·916] vs 0·884 [0·876–0·892]), cardiac amyloidosis detection (0·790 [0·767–0·812] vs 0·777 [0·754–0·799]), hypertrophic cardiomyopathy detection (0·772 [0·751–0·793] vs 0·754 [0·733–0·774]), and atrial fibrillation prediction (0·777 [0·773–0·781] vs 0·765 [0·761–0·769]). The model has good performance in the detection of anterior and inferior acute myocardial infarction when using single ECG leads, outperforming supervised baseline models when using leads typically considered less informative. Interpretation ECG-CLIP is a generalisable foundation model that can learn clinically meaningful representations from ECGs and paired reports for diagnostic and prediction tasks. This approach enables accurate, data-efficient, and interpretable risk predictions across diverse clinical tasks, supporting scalable deployment in real-world and resource-limited settings.
Discussion
No takes yet. Share an insight, caveat, or question.
Cite This Study
Ko et al. (2026) conducted an observational in Cardiovascular diseases (n=542,288). ECG-CLIP foundation model vs. Supervised baseline models and ECG signal-only foundation models was evaluated on Acute myocardial infarction detection (low-label setting with 10 positive labels) (AUC 0.910, 95% CI 0.903-0.916). The ECG-CLIP foundation model outperformed comparator models in low-label settings, achieving an AUC of 0.910 (95% CI 0.903-0.916) vs 0.884 for acute myocardial infarction detection.