Artificial Intelligence-Enhanced Electrocardiography Identifies Patients With Normal Ejection Fraction at...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• An artificial intelligence (AI)-based electrocardiogram (ECG) model aids in identifying patients at higher risk for low ejection fraction (EF).
• Patients identified as false positives (FP) by AI-ECG, exhibiting normal EF but abnormal AI-ECG scores, demonstrated a greater likelihood of developing future low EF.

Purpose of the Study:

• To investigate the echocardiographic characteristics of FP patients.
• To assess the association between FP status and all-cause mortality risk.

Main Methods:

• Retrospective classification of patients into true negatives (TN), FP, false negatives (FN), and true positives (TP) based on AI-ECG scores and EF measurements.
• Detailed analysis of echocardiographic parameters, including left ventricular function, valve disease, pulmonary pressures, and right heart function.
• Application of Cox regression models to determine factors linked to all-cause mortality.

Main Results:

• Out of 100,586 patients, 7% were classified as FP. FP patients exhibited more echocardiographic abnormalities than TN patients (97% of FPs had abnormalities).
• FP patients faced increased all-cause mortality risk (HR: 1.64) compared to TN patients over a median follow-up of 2.7 years.
• Mortality risk was significantly higher in FP patients with abnormal echocardiograms compared to those with normal echocardiograms and TN patients.

Conclusions:

• AI-ECG models can identify patients with normal EF who have underlying echocardiographic abnormalities.
• FP patients, particularly those with concurrent echocardiographic abnormalities, exhibit elevated mortality risk.
• Combining AI-ECG and echocardiography enhances risk stratification for patients with normal left ventricular ejection fraction (LVEF).