Kidney Week

Abstract: TH-OR22

Artificial Intelligence-Enabled Electrocardiography for Predicting Dyskalemia Using Home-Based Electrocardiography Devices

Session Information

• Augmented Intelligence and Digital Health Advances
  October 24, 2024 | Location: Room 2, Convention Center
  Abstract Time: 05:00 PM - 05:10 PM

Category: Augmented Intelligence, Digital Health, and Data Science

• 300 Augmented Intelligence, Digital Health, and Data Science

Authors

• Lin, Shih-Hua P., Tri-Service General Hospital, Taipei, Taiwan

Shih-Hua P. Lin, MD

• Sung, Chih-Chien, Tri-Service General Hospital, Taipei, Taiwan

Chih-Chien Sung, MD, PhD

• Lu, Ang, Tri-Service General Hospital, Taipei, Taiwan

Ang Lu, MD

• Chen, Chien-Chou, Tri-Service General Hospital, Taipei, Taiwan

Chien-Chou Chen, MD

• Lin, Chin, Tri-Service General Hospital, Taipei, Taiwan

Chin Lin

Background

Dyskalemia is the common electrolytes abnormality associated with an increased morbidity and mortality. We have developed a bloodless artificial Intelligence-enabled electrocardiogram (AI-ECG) system capable of predicting hyperkalemia and hypokalemia using Philips ECG devices and providing quantitative serum potassium (K⁺) detection and monitoring. This study aims to validate the application of a newly-developed home-based ECG device for dyskalemia detection.

Methods

Home-based ECG device is a portable 12-lead ECG machine developed by QT Medical (PCA500). We conducted a preliminary study with 107 hospitalized patients for various causes including acute kidney injury and advanced CKD on hemodialysis at a single academic medical center, resulting in 1033 ECG recordings. Among them, eighty-eight ECGs had the corresponding serum K⁺ concentration measured within the previous three hours. The dataset included 12 patients with hyperkalemia (Lab-K⁺ ≥ 5.5 mmol/L) and 11 hypokalemia (Lab-K⁺ ≤ 3.5 mmol/L).

Results

The AI-ECG system demonstrated an AUC of >0.95 for predicting hyperkalemia and an AUC of 0.809 for hypokalemia. The correlation coefficient between ECG-derived K⁺ levels (ECG-K⁺) and laboratory-measured K⁺ levels (Lab-K⁺) was 0.796, with a mean absolute error (MAE) of 0.456

Conclusion

Using this home-based ECG device, these results show that AI-ECG predict hyperkalemia better than hypokalemia and may be applied to future remote detection or monitoring of hyperkalemia in high-risk patients. A further validation with larger sample sizes is still required to confirm these findings.