Artificial intelligence-enabled electrocardiography identifies severe dyscalcemias and has prognostic value.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Abnormal serum calcium (dyscalcemia) impacts cardiac function and electrocardiogram (ECG) readings.
• Current detection methods rely on blood tests with significant turnaround times.

Purpose of the Study:

• To develop a bloodless artificial intelligence (AI)-enabled ECG method for rapid dyscalcemia detection.
• To assess the AI-ECG's utility in predicting patient outcomes.

Main Methods:

• Utilized a large dataset of 86,731 ECGs for development and over 24,000 for validation.
• Trained an AI model to predict albumin-adjusted calcium (aCa) levels from ECGs.
• Correlated AI-ECG predictions with patient outcomes like mortality, myocardial infarction, and heart failure.

Main Results:

• The AI-ECG model achieved high accuracy in detecting severe hypercalcemia and hypocalcemia (AUCs up to 0.92).
• AI-ECG predictions explained <20% of variance via traditional ECG features but showed strong prognostic value.
• Patients with AI-ECG-detected hypercalcemia, even with normal initial calcium, faced higher risks of mortality, AMI, and HF.

Conclusions:

• AI-ECG-aCa shows promise for early dyscalcemia detection, potentially reducing reliance on lab tests.
• ECG-hypercalcemia identified by AI has significant prognostic implications for adverse clinical events.