Development and Validation of a Deep-Learning Model to Screen for Hyperkalemia From the Electrocardiogram

Area of Science:

• Nephrology
• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Hyperkalemia is a frequent and potentially fatal complication in patients with chronic kidney disease (CKD).
• Current monitoring of serum potassium levels is often underutilized, despite hyperkalemia being asymptomatic and linked to fatal arrhythmias.
• Noninvasive screening methods for hyperkalemia are needed to improve early detection in CKD patients.

Purpose of the Study:

• To evaluate the diagnostic performance of a deep-learning model for detecting hyperkalemia using electrocardiograms (ECGs) in patients with CKD.
• To assess the potential of artificial intelligence (AI) in noninvasively screening for hyperkalemia.

Main Methods:

• A deep convolutional neural network (DNN) was trained on over 1.5 million ECGs from a large patient cohort.
• The DNN model utilized 2 or 4 ECG leads to detect hyperkalemia (serum potassium ≤5.5 mEq/L).
• Model performance was validated retrospectively on over 60,000 CKD patients with contemporaneous ECG and serum potassium measurements.

Main Results:

• The deep-learning model demonstrated strong performance in detecting hyperkalemia across multiple validation datasets.
• Area under the receiver operating characteristic curve (AUC) ranged from 0.853 to 0.883 using only two ECG leads.
• At a high sensitivity operating point (90%), the model achieved sensitivities between 88.9% and 91.3% with specificities ranging from 54.7% to 63.2%.

Conclusions:

• A deep-learning model utilizing ECG data can effectively screen for hyperkalemia in patients with chronic kidney disease.
• AI-driven analysis of ECGs offers a promising, noninvasive approach for hyperkalemia detection.
• Further prospective studies are warranted to confirm these findings and explore clinical implementation.