What is the clinical evidence from this study?

Study design: Observational. Population: Heart failure (n=69027). Intervention: Artificial intelligence-enabled electrocardiogram model. Primary outcome: Detection of hospital-diagnosed heart failure (AUC 0.86, 95% CI 0.847-0.864).

What does this research mean for the field?

A deep learning model using 12-lead electrocardiograms can accurately detect hospital-diagnosed heart failure, achieving an AUC of 0.86 that improves to 0.96 when training labels are refined with strict NT-proBNP thresholds. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

This research aims to improve heart failure diagnosis by implementing a deep learning model that analyzes electrocardiograms.

May 21, 2026Open Access

Heart failure detection in electrocardiograms using Artificial Intelligence and pragmatic labelling

Key Result

An artificial intelligence model using 12-lead electrocardiograms detected hospital-diagnosed heart failure with an AUC of 0.86, which improved to 0.96 when validated with strict NT-proBNP thresholds.

Key Points

This research aims to improve heart failure diagnosis by implementing a deep learning model that analyzes electrocardiograms.
Developed a neural network trained on a cohort of 25,300 patients and validated on 43,727 patients.
Utilized NT-proBNP levels to refine labelling during training and improved model accuracy.
Conducted external validation using the MIMIC-IV cohort of 161,352 patients.
Achieved an AUC of 0.96 (95% CI 0.951–0.963) with age-adjusted NT-proBNP thresholds.
Model predictions validated against echocardiographic assessments demonstrated accuracy in capturing diastolic and systolic function.
Released the model as open source for wider accessibility.

Study Design

Type

Observational (n=69,027)

Multicenter

Yes

Structured PICO

Does a deep learning model using electrocardiograms accurately detect heart failure?

Population

230,379 patients across a development cohort (n=25,300), an independent test cohort (n=43,727), and an external validation cohort (MIMIC-IV, n=161,352)

Intervention

Deep learning neural network for heart failure detection from electrocardiograms, trained using diagnosis codes validated by NT-proBNP levels

Outcome

Area under the receiver operating characteristic curve (AUC) for detecting hospital-diagnosed heart failuresurrogate

A deep learning model trained on ECGs with NT-proBNP-validated labels can accurately detect heart failure, offering a scalable tool for addressing underdiagnosis.

Main Result

Effect estimate: AUC 0.86 (95% CI 0.847-0.864)

Limitations

Lack of a definitive gold standard for heart failure diagnosis in routinely collected data.
ICD-10 codes have limited validity and act as a noisy proxy for the underlying heart failure syndrome.
Imputation of missing values in the H2FPEF grading algorithm may have led to underestimation of scores.
Some analyses were specified post hoc.
The study population had limited sociodemographic diversity, with less than 20% being of non-Caucasian descent.

Abstract

Abstract The diagnosis of heart failure (HF) is resource-intensive, leading to severe underdiagnosis. This study proposes the use of a deep learning model to detect HF solely from electrocardiograms. HF has limited validity in diagnosis codes, lowering the viability of direct application of supervised learning. However, validating the codes with measured levels of circulating N-terminal proB-type natriuretic peptide (NT-proBNP) during training mitigates the impact of label noise. We developed and prospectively validated a neural network for HF detection, using a development cohort of 25,300 patients and an independent test cohort of 43,727 patients. The model achieved an AUC of 0.86 (95% CI 0.847–0.864) for hospital-diagnosed HF, which increased to 0.91 (95% CI 0.900–0.913) with age-adjusted NT-proBNP thresholds and 0.96 (95% CI 0.951–0.963) with stricter limits of 125 and 1000 ng/L. The model was externally validated using the MIMIC-IV cohort of 161,352 patients, yielding AUC values of 0.87 (95% CI 0.861–0.871), 0.90 (95% CI 0.895–0.908), and 0.96 (95% CI 0.950–0.964) with the same labelling strategies. Model predictions were further validated using echocardiographic assessments of diastolic function grade and the H2FPEF score, demonstrating that the model accurately captures both diastolic and systolic function. The model has been released as open source.

Demander à l'IA

Bookmark

View Full Paper