AI-guided ECG screening to target guideline heart-failure medications and prevent left ventricular dysfunction

AI-Enabled Electrocardiogram-Guided Guideline-Directed Medical Therapy on Incident Left Ventricular Dysfunction: A Target Trial Emulation Study

This project tests whether an AI-read ECG can find people with normal ejection fraction who are at high risk of future left ventricular dysfunction and whether starting guideline-directed heart-failure medicines can lower that risk.

Quick facts

• Phase: Not applicable
• Study type: Interventional
• Enrollment: 5000 (estimated)
• Ages: 18 Years and up
• Sex: All
• Sponsor: Tri-Service General Hospital (Academic / other)
• Locations: 1 site (Taipei, Taipei)
• Trial ID: NCT07355023 on ClinicalTrials.gov

What this trial studies

This is a retrospective multicenter analysis using electronic medical records from one academic center and seven regional hospitals in Taiwan from January 2016 to December 2024. Investigators trained an AI-ECG model on over 50,000 paired ECG–echocardiogram recordings (within 7 days) using an 82-layer convolutional neural network with attention, and then applied the model to a cohort with preserved LVEF (≥50%) to stratify high- versus low-risk patients. The team emulated a target trial to compare outcomes in patients who did or did not receive guideline-directed medical therapies (such as ACE inhibitors/ARBs) after AI-based risk classification. The primary outcome was incident left ventricular functional decline, defined as follow-up LVEF ≤40% on echocardiography.

Who should consider this trial

Good fit: Patients with preserved LVEF (≥50%), an ECG followed by an echocardiogram within 90 days, complete ECG lead data, and available follow-up within the participating Taiwan health system are the ideal candidates.

Not a fit: Patients with a prior LVEF <50%, missing essential ECG or clinical data, short or no follow-up, or contraindications to guideline heart-failure medications are unlikely to benefit from this approach.

Why it matters

Potential benefit: If successful, this approach could enable earlier identification of at-risk patients and targeted use of guideline therapies to reduce progression to severe left ventricular dysfunction.

How similar studies have performed: Previous AI-ECG research has reliably detected existing low ejection fraction and predicted future risk, but using AI-ECG to trigger preventive guideline-directed therapy in an emulated trial format is novel and has limited prior validation.

Eligibility criteria

Inclusion Criteria:

* With an ECG followed by an echocardiogram within a 90-day interval
* With preserved left ventricular ejection fraction (LVEF ≥ 50%)

Exclusion Criteria:

* missing essential variables or ECG lead data
* any prior LVEF \< 50%
* loss to follow-up or death during the 90-day assessment window

Where this trial is running

Taipei, Taipei

• Tri-Service General Hospital — Taipei, Taipei, Taiwan (Recruiting)

Study contacts

• Study coordinator: Wei-Ting Liu, M.D.

Conditions Left Ventricular Systolic Dysfunction; Artificial intelligence; Electrocardiogram; Deep learning; Left ventricular dysfunction