Why the study?
Heart failure diagnosis is challenging, and previous AI-ECG models relied on echocardiographic labelling and often focused on specific ejection fraction phenotypes rather than identifying heart failure across the ejection fraction spectrum.
Does an AI-enhanced ECG model trained on ICD-10 codes and NT-proBNP levels accurately detect hospital-admitted heart failure regardless of ejection fraction?
Population
83,000 patients from a network of four tertiary hospitals
Comparison
AI-ECG model trained with ICD-10 and NT-proBNP labelling vs NT-proBNP levels and H2FPEF score
Design
Multicenter AI model development and prospective testing study
Key result
AI-ECG detected hospital-admitted heart failure with AUC 0.84 regardless of ejection fraction, outperforming NT-proBNP and identifying HFpEF even with low NT-proBNP.
Authors
Loading...
May support ECG-based HF screening across EF; hypothesis-generating pending prospective validation.
Does an AI-enhanced ECG model trained on ICD-10 codes and NT-proBNP levels accurately detect hospital-admitted heart failure regardless of ejection fraction?
An AI-enhanced ECG model trained without explicit echocardiographic labeling can accurately detect heart failure across the ejection fraction spectrum, potentially serving as an accessible screening tool in primary care.
Stenhede et al. (2025) studied this question. AI-ECG detected hospital-admitted heart failure with AUC 0.84 regardless of ejection fraction, outperforming NT-proBNP and identifying HFpEF even with low NT-proBNP.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: