Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
January 8, 2026The Korean Journal of Internal MedicineOpen Access

The combination of ECG data and clinical metadata significantly improves the prediction accuracy of left ventricular ejection fraction in patients with heart failure.

View Full Paper
Ask AI
Bookmark
Share

Key result

The combination of ECG data and clinical metadata significantly improves the prediction accuracy of left ventricular ejection fraction in patients with heart failure.

Authors

HPHyun Woong ParkTKTaeseen KangYSYoung-Hoon Seo

Discussion

Loading...

Member takes

Overview

Retrospective study demonstrates improved LVEF prediction in heart failure, suggesting better screening methods.

Key Points

  • This study aims to improve the prediction of left ventricular ejection fraction (LVEF) using a combination of electrocardiography and clinical metadata.
  • Analyzed ECG and clinical metadata retrospectively from two tertiary teaching hospitals.
  • Utilized a deep neural network (EfficientNet B3) for predicting LVEF.
  • Incorporated clinical metadata, including age, NT-proBNP, and sodium levels.
  • Assessed model performance using area under the curve (AUC) and coefficient of determination (R2).
  • Achieved an AUC of 0.95 when combining ECG data with clinical metadata.
  • Outperformed ECG-only models, which had an AUC of 0.90.
  • Specificity of the model was high at 96.9%, with a sensitivity of 54.8%.
  • Indicates potential as a screening tool for heart failure with reduced ejection fraction.

Study Design

Type

Observational (n=33,500)

Multicenter

Yes

PICO

P
Population
Heart Failure with Reduced Ejection Fraction (HFrEF) (n=33,500)
I
Intervention / Comparator
Deep Neural Network (EfficientNet B3) with clinical metadata vs ECG alone
O
Primary Outcome
Prediction accuracy for HFrEF detection based on AUC — null (0.939-0.962), p=0.001

Main Result

Effect estimate: null (95% CI 0.939-0.962)

Absolute Event Rate: 0.95% vs 0.9%

p-value: p=0.001

Limitations

  • The study data is derived from only two institutions, limiting generalizability.
  • The study design is retrospective, lacking real-world clinical validation.
  • Variability in data quality and heterogeneity across institutions.

Cite This Study

Park et al. (2026) conducted an observational in Heart Failure with Reduced Ejection Fraction (HFrEF) (n=33,500). Deep Neural Network (EfficientNet B3) with clinical metadata vs. ECG alone was evaluated on Prediction accuracy for HFrEF detection based on AUC (null, 95% CI 0.939-0.962, p=0.001). The combination of ECG data and clinical metadata significantly improves the prediction accuracy of left ventricular ejection fraction in patients with heart failure.

synapsesocial.com/papers/69608779fa51ca23bb9809b4https://doi.org/10.3904/kjim.2025.104
View Full Paper
Ask AI
Bookmark
Share