Detection of Hypokalemia, Hyponatremia, and Hyperkalemia in Heart Failure Patients Using Artificial Intelligence Techniques via Electrocardiography

Abstract

Objective: Detection and monitoring of electrolyte imbalances are essential for the appropriate treatment of many metabolic diseases. However, no reliable and noninvasive tool currently exists for such detection. Electrolyte disorders, particularly in heart failure patients, can lead to life-threatening situations, which may often develop as a result of medications used in routine treatment. Method: In this study, we developed a deep learning model (DLM) using electrocardiography (ECG) to detect electrolyte imbalances in heart failure patients and evaluated its performance in a multicenter setting. Seventeen different centers participated in this study. Heart failure patients (ejection fraction <= 45%) who had blood electrolyte measurements and ECG taken on the same day were included. Patients were divided into four groups: those with normal electrolyte values, those with hypokalemia, those with hyperkalemia, and those with hyponatremia. Patients who developed electrolyte disorders due to medications used for heart failure were classified in the relevant group. Confidence intervals (CI): We computed 95% CIs for area under the receiver operating characteristic curve (AUROC) via stratified bootstrap (2,000 resamples at the patient level) and 95% CIs for accuracy using the Wilson score interval for binomial proportions. Results: The accuracy rates of the DLM in detecting hyponatremia, hypokalemia, and hyperkalemia were 83.33%, 95.33%, and 95.77%, respectively. Conclusion: The proposed DLM demonstrated high performance in detecting electrolyte imbalances. These results suggest that a DLM can be used to detect and monitor electrolyte imbalances using ECG on a daily basis.