Abstract: FR-PO0008
Artificial Intelligence-Enabled Electrocardiography for Real-Time Monitoring Serum Potassium Dynamics in Patients with Severe Hypokalemia
Session Information
- Artificial Intelligence and Digital Health at the Bedside
November 07, 2025 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Authors
- Ding, Jhao-Jhuang, Tri-Service General Hospital, Taipei City, Taiwan
- Chen, Chien-Chou, Tri-Service General Hospital, Taipei City, Taiwan
- Lin, Chin, Tri-Service General Hospital, Taipei City, Taiwan
- Sung, Chih-Chien, Tri-Service General Hospital, Taipei City, Taiwan
- Hsu, Yu-Juei, Tri-Service General Hospital, Taipei City, Taiwan
- Hsu, Shun-Neng, Tri-Service General Hospital, Taipei City, Taiwan
- Tseng, Min-hua, Linkou Chang Gung Memorial Hospital, Taoyuan, Taiwan
- Lin, Shih-Hua P., Tri-Service General Hospital, Taipei City, Taiwan
Background
Severe hypokalemia is a life-threatening emergency requiring prompt management and close surveillance. Although bloodless artificial intelligence-enabled electrocardiography (AI-ECG) has been developed to detect severe hypokalemia rapidly, its application to monitor serum potassium (K+) changes remains unexplored.
Methods
This multicenter retrospective study was performed at an emergency department (ED) over 5 years. Patients with severe hypokalemia defined as serum K+ level (Lab-K+< 2.5 mmol/L) with matched AI-ECG-K+ < 3.5 mmol/L were included. AI-ECG-K+ was quantified by ECG12Net analysis. The following paired ECG-K+ and Lab-K+ were measured at least once within 24 hours during K+ supplementation. Clinical characteristics and relevant laboratory data were analyzed.
Results
One hundred ninety-four patients (mean age 60.0±20.6 years; 113 males, 81 females) fulfilling the inclusion criteria exhibited initial Lab-K+ 2.2 ± 0.3 and ECG-K+ 2.5 ± 0.4 mmol/L. Their underlying hypokalemia included acute shifting K+ disorders (n=36) and chronic K+ deficits (n=158). There were 279 paired Lab-K+ and ECG-K+ measurements including 148 (76.3%) with one pair, 26 (13.4%) with two pairs and 20 (10.6%) with three or more pairs. The paired ECG-K+ and Lab-K+ were significantly increased in parallel and strongly correlated (p< 0.001) during K+ therapy. The sensitivity of detecting severe and moderate hypokalemia was 0.96 and 0.90, respectively. Of note, two acute hypokalemic patients developing with rebound hyperkalemia (K+ 8.8 and 6.5 mmol/L) was also rapidly detected by ECG-K+.
Conclusion
AI-ECG provides reliable real-time monitoring of serum K+ dynamics during treatment and potentially early detection of rebound hyperkalemia in patients with acute hyperkalemia.