AI-ECG for Detecting Left Ventricular Systolic Dysfunction in Chagas Disease: Diagnostic and Prognostic Value

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Left ventricular systolic dysfunction (LVSD) is a key mortality predictor in Chagas disease (ChD).
• Diagnosing LVSD typically requires cardiac imaging, often inaccessible in resource-limited areas.
• Affordable treatments for LVSD exist but depend on timely diagnosis.

Purpose of the Study:

• To evaluate an artificial intelligence-enabled electrocardiogram (AI-ECG) for detecting LVSD in Chagas disease patients.
• To assess the AI-ECG's ability to predict mortality and incident LVSD in this population.
• To compare AI-ECG performance against established biomarkers like NT-proBNP.

Main Methods:

• A fine-tuned AI-ECG model for LVSD was applied to the SaMi-Trop Brazilian ChD cohort.
• Diagnostic accuracy for LVSD was compared between AI-ECG and NT-proBNP, with echocardiography as the gold standard.
• Prognostic performance for mortality and incident LVSD was evaluated using Cox and log-binomial models, respectively.

Main Results:

• AI-ECG demonstrated high accuracy (AUC: 0.89) for LVSD detection, comparable to NT-proBNP (AUC: 0.90).
• AI-ECG accurately predicted 2- and 9-year all-cause mortality and incident LVSD over 7 years, similar to NT-proBNP.
• The AI-ECG model showed potential to substitute NT-proBNP in mortality risk scores with minimal accuracy loss.

Conclusions:

• A fine-tuned AI-ECG model effectively detects LVSD and predicts long-term mortality and incident LVSD in Chagas disease.
• While slightly less performant than NT-proBNP, AI-ECG offers a viable alternative in settings lacking NT-proBNP testing.
• The study highlights AI-ECG's potential to improve Chagas disease management, especially in underserved regions.