Leveraging the wearable 1-lead ECG signal. From cardiac rhythm to cardiac function assessment

Q: What is the clinical evidence from this study?

Study design: Cross-Sectional. Population: Acute coronary syndrome / Post-myocardial infarction (n=6703). Intervention: 1-lead smartwatch ECG with deep learning vs. 12-lead ECG. Primary outcome: Prediction of moderate/severe left ventricular function impairment (AUROC) (AUC 0.883, 95% CI 0.881-0.886).

Q: What does this research mean for the field?

Left ventricular function can be accurately predicted from single-lead wearable ECGs using deep learning, achieving diagnostic performance nearly equivalent to standard 12-lead ECGs. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Viktor van der Valk; Douwe Atsma; Roderick Scherptong; Marius Staring

doi:10.1038/s44325-026-00153-2

October 2, 2026npj Cardiovascular HealthOpen Access

Leveraging the wearable 1-lead ECG signal. From cardiac rhythm to cardiac function assessment

View Full Paper

Ask AI

Bookmark

Share

Key result

A deep learning model using 1-lead smartwatch ECGs accurately predicted left ventricular function with an AUC of 0.883, nearly matching the predictive performance of standard 12-lead ECGs (AUC 0.897).

Why the study?

Traditional 12-lead ECGs require clinical settings and expert interpretation, limiting accessibility, while the clinical utility of 1-lead ECGs beyond rhythm assessment is unclear.

Does a 1-lead smartwatch ECG combined with deep learning accurately predict left ventricular function compared to a 12-lead ECG in post-myocardial infarction patients?

Key Points

To determine whether wearable single-lead ECGs combined with explainable deep learning models can accurately assess left ventricular function compared to standard 12-lead ECGs in post-myocardial infarction patients.
Implemented an autocorrelation-based sieve using adjacent recordings to filter out low-quality wearable signals and conducted attribute manipulation analyses to assess model interpretability.
Pretrained single-lead ECG models predicted moderate-to-severe left ventricular dysfunction with an AUROC of 0.883 [95% CI, 0.881–0.886], closely matching the 12-lead ECG model AUROC of 0.897 [95% CI, 0.887–0.906].
At a set sensitivity of 0.90, specificities were 0.71 [95% CI, 0.70–0.72] for the pretrained single-lead model, 0.65 [95% CI, 0.64–0.66] for the standard single-lead model, and 0.64 [95% CI, 0.58–0.69] for the full 12-lead model.

Study Design

Type

Cross-Sectional (n=6,703)

Multicenter

No

Structured PICO

Does a 1-lead smartwatch ECG combined with deep learning accurately predict left ventricular function compared to a 12-lead ECG in post-myocardial infarction patients?

P

Population

6,703 patients recovering from acute coronary syndrome who had standard 12-lead ECGs and/or home-recorded 1-lead smartwatch ECGs matched with echocardiography data to assess left ventricular function.

E

Exposure

1-lead ECG recorded at home using a smartwatch, analyzed using an explainable deep learning model (partial β-variational autoencoder) to predict left ventricular function.

C

Comparator

Standard 10-second 12-lead ECG analyzed using the same explainable deep learning model architecture.

O

Outcome

Prediction of left ventricular function (LVF) categorized into four ordinal classes (normal ≥55%, mild impairment 45%-54%, moderate impairment 35%-44%, severe impairment <35%) based on echocardiogram reports within 7 days of the ECG recording.surrogate

Main Result

Effect estimate: AUC 0.883 (95% CI 0.881-0.886)

Absolute Event Rate: 0.883% vs 0.897%

Wearable 1-lead ECGs combined with explainable AI can accurately predict left ventricular function in post-myocardial infarction patients, offering a scalable tool for remote structural cardiac monitoring.

Limitations

Single-center dataset lacking external validation in independent cohorts or broader cardiac populations.
Real-world recording artifacts and data quality issues (e.g., device orientation errors, non-rest acquisition) in unsupervised home monitoring.
Left ventricular function was treated as a discrete classification task rather than a continuous regression task.
The models may not be directly generalizable to broader cardiac populations due to the confined heterogeneity of the post-myocardial infarction population.
Absence of external validation (single-center dataset)
1-lead data quality challenges including device orientation errors, device sharing, and non-rest acquisition
Models likely not directly generalizable to broader cardiac populations due to restriction to the post-myocardial infarction population

Ask about this paper

Answers are grounded in this paper and cite their sources.

Implication

Should not yet change post-MI monitoring; hypothesis-generating for wearable ECG-AI validation in prospective cohorts.

Abstract

Traditional 12-lead ECGs offer comprehensive insights into the electrical activity of the heart, but require clinical settings and expert interpretation, which limits their accessibility. Smartwatch 1-lead ECGs can be recorded at home, allowing more frequent and rapid monitoring, opening opportunities for early adverse event detection and enhanced patient autonomy. This study investigates whether 1-lead ECGs can provide clinically meaningful information beyond heart rhythm assessment. Using explainable deep learning models, we predict left ventricular function (LVF) from both 1-lead and 12-lead ECGs in a post-myocardial infarction population, and compare their respective performances. Our findings demonstrate that LVF can be accurately predicted from 1-lead ECGs alone (AUC = 0.883), nearly matching the predictive performance of 12-lead ECGs (AUC = 0.897). Explainability analyses further reveal that the models leverage physiologically plausible ECG features, supporting the validity of this approach. These results suggest that 1-lead ECGs, when combined with explainable AI, have the potential to support broader clinical applications and empower patients, particularly in resource-limited or remote settings where access to traditional cardiac diagnostics remains constrained.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Valk et al. (2026) conducted a cross-sectional in Acute coronary syndrome / Post-myocardial infarction (n=6,703). 1-lead smartwatch ECG with deep learning vs. 12-lead ECG was evaluated on Prediction of moderate/severe left ventricular function impairment (AUROC) (AUC 0.883, 95% CI 0.881-0.886). A deep learning model using 1-lead smartwatch ECGs accurately predicted left ventricular function with an AUC of 0.883, nearly matching the predictive performance of standard 12-lead ECGs (AUC 0.897).

synapsesocial.com/papers/6ac1ab93cc6172083c9289b0 https://doi.org/10.1038/s44325-026-00153-2

View Full Paper

Ask AI

Bookmark

Share

Leveraging the wearable 1-lead ECG signal. From cardiac rhythm to cardiac function assessment

Key Points

Study Design

Structured PICO

Main Result

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Leveraging the wearable 1-lead ECG signal. From cardiac rhythm to cardiac function assessment

Key Points

Study Design

Structured PICO

Main Result

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading