AI-based identification of patients who benefit from revascularization: a multicenter study

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Informatics

Background:

• Revascularization decisions in stable coronary artery disease (CAD) traditionally rely on ischemia severity.
• Current guidelines and randomized controlled trials (RCTs) may not fully address individual patient needs for revascularization.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI)-driven approach to estimate individualized treatment effects for revascularization in stable CAD.
• To introduce the AI-REVASC score for personalized revascularization decision-making.

Main Methods:

• Developed an AI model using a large, international registry (13 centers) to simulate outcomes under different therapeutic strategies.
• Emulated RCTs using 1:1 propensity score matching to create balanced patient pairs differing only in treatment strategy (early revascularization vs. medical therapy).
• Derived and validated the AI-REVASC score in a held-out cohort using Cox regression.

Main Results:

• The AI model analyzed 45,252 patients with a median follow-up of 3.6 years; 9.6% experienced myocardial infarction (MI) or death.
• A specific subgroup (n=1,335, 5.9%) identified by the AI model showed significant benefit from early revascularization (propensity-adjusted hazard ratio: 0.50).
• Patients benefiting from early revascularization had higher rates of hypertension, diabetes, dyslipidemia, and lower left ventricular ejection fraction (LVEF).

Conclusions:

• Pioneered a scalable, data-driven method to emulate RCTs using retrospective data for personalized medicine.
• The AI-REVASC score facilitates precision revascularization decisions, particularly in cases where existing guidelines and RCT evidence are insufficient.