Serum Potassium Monitoring Using AI-Enabled Smartwatch Electrocardiograms
Authors' conclusion
We validate a deep learning model to predict serum potassium levels from both 12-lead ECGs and single-lead smartwatch data, demonstrating its utility for remote monitoring of hyperkalemia.
Study context
- Population
- Patients with end-stage renal disease (ESRD); serum potassium paired with electrocardiogram (ECGs) within 1 hour (Cedars Sinai, Stanford Healthcare, Chang Gung Memorial Hospital)training cohort
- Participants
- 152,508 patients
- Comparison
- AI-ECG (Kardio-Net) prediction of serum potassium from 12-lead, single-lead and smartwatch ECGs
- Outcome
- Detection of severe hyperkalemia (>6.5 mEq/L) (AUC) and serum potassium prediction (MAE)
Methodological detailsAI-ECG model development and validation, external validation, prospective international smartwatch cohort
- AI-ECG model development and validation
- External validation
- Prospective international smartwatch cohort
Performance
As reported in the abstract
Severe hyperkalemia (>6.5 mEq/L)
| Test or model | AUC (primary cohort) | Mean absolute error (MAE) (primary cohort) | AUC (external Stanford Healthcare (SHC)) | MAE (external SHC) | AUC | MAE |
|---|---|---|---|---|---|---|
| 12-lead ECG | 0.852 (95% CI: 0.745-0.956) | 0.527 mEq/L | 0.849 (95% CI: 0.823-0.875) | 0.599 mEq/L | ||
| Single-lead ECG | 0.876 (95% CI: 0.765-0.987) | 0.575 mEq/L | 0.807 (95% CI: 0.778-0.835) | 0.740 mEq/L | ||
| Smartwatch ECG | 0.831 (95% CI: 0.693-0.975) | 0.580 mEq/L |
AUC (primary cohort)
12-lead ECG0.852 (95% CI: 0.745-0.956)
Single-lead ECG0.876 (95% CI: 0.765-0.987)
Smartwatch ECGNot reported in the abstract
Mean absolute error (MAE) (primary cohort)
12-lead ECG0.527 mEq/L
Single-lead ECG0.575 mEq/L
Smartwatch ECGNot reported in the abstract
AUC (external Stanford Healthcare (SHC))
12-lead ECG0.849 (95% CI: 0.823-0.875)
Single-lead ECG0.807 (95% CI: 0.778-0.835)
Smartwatch ECGNot reported in the abstract
MAE (external SHC)
12-lead ECG0.599 mEq/L
Single-lead ECG0.740 mEq/L
Smartwatch ECGNot reported in the abstract
AUC
12-lead ECGNot reported in the abstract
Single-lead ECGNot reported in the abstract
Smartwatch ECG0.831 (95% CI: 0.693-0.975)
MAE
12-lead ECGNot reported in the abstract
Single-lead ECGNot reported in the abstract
Smartwatch ECG0.580 mEq/L
Original publication
Serum Potassium Monitoring Using AI-Enabled Smartwatch Electrocardiograms. JACC Clin Electrophysiol. Published 9 October 2024. doi:10.
Summarised from the published abstract; every number is checked against it