Abstract: FR-PO569
Artificial Intelligence-Enabled Electrocardiography Helps Identify Severe Dyscalcemia and Provide Additional Prognostic Value
Session Information
- Fluid, Electrolyte, and Acid-Base Disorders: Clinical
November 04, 2022 | Location: Exhibit Hall, Orange County Convention Center‚ West Building
Abstract Time: 10:00 AM - 12:00 PM
Category: Fluid‚ Electrolyte‚ and Acid-Base Disorders
- 1002 Fluid‚ Electrolyte‚ and Acid-Base Disorders: Clinical
Authors
- Lin, Shih-Hua P., Tri-Service General Hospital Department of Internal Medicine, Taipei, Taiwan
- Sung, Chih-Chien, Tri-Service General Hospital Department of Internal Medicine, Taipei, Taiwan
Background
The detection of hypocalcemia and hypercalcemia (collectively dyscalcemia) relies on blood laboratory tests requiring turnaround time while abnormal serum calcium concentrations may affect the heart and alter the electrocardiogram (ECG). The study aimed to develop a bloodless artificial intelligence (AI)-assisted (ECG) to rapidly detect dyscalcemia and analyze its contribution to prognostic predictions.
Methods
This study collected 86,731 development, 15,611 tuning, 11,105 internal validation, and 8,401 external validation ECGs from electronic medical records with at least 1 ECG labeled by an albumin-adjusted calcium (aCa) value within 4 hours. The main outcomes were to assess the accuracy of AI-ECG to predict aCa and follow up these patients for all-cause mortality, new-onset acute myocardial infraction (AMI), and new-onset heart failure (HF) to validate the ability of AI-ECG-aCa for previvor identification.
Results
ECG-aCa had mean absolute errors (MAE) of 0.78/0.98 mg/dL and achieved an area under receiver operating characteristic curves (AUCs) 0.9219/0.8447 and 0.8948/0.7723 to detect severe hypercalcemia and hypocalcemia in the internal/external validation sets, respectively. Although <20% variance of ECG-aCa could be explained by traditional ECG features, the ECG-aCa was found to be associated with medical complexity. Patients with ECG-hypercalcemia but initially normal aCa were found to have a higher risk of subsequent all-cause mortality [hazard ratio (HR): 2.05, 95% conference interval (CI): 1.55-2.70], new-onset AMI (HR: 2.88, 95% CI: 1.72-4.83), and new-onset HF (HR: 2.02, 95% CI: 1.38-2.97) in the internal validation set, which were also seen in external validation.
Conclusion
The AI-ECG-aCa may help detect severe dyscalcemia for decision support and ECG-hypercalcemia also provides prognostic value for future cardiovascular outcomes.