Development of deep learning algorithm for detecting dyskalemia based on electrocardiogram (Preprint)

Abstract

Dyskalemia is a common electrolyte abnormality that can cause fatal arrhythmias and cardiac arrest in severe cases. We detected dyskalemia quickly and easily using a deep learning-based model that has learned electrocardiograms (ECG), which are noninvasive and can be quickly measured. This retrospective cohort study was conducted at two hospitals from 2006 to 2020. The training set, validation set, internal testing cohort, and external validation cohort comprised 310,449, 15,828, 23,849, and 130,415 ECG-serum potassium laboratory samples, respectively. A DeepECG-Hyperkalemia model diagnosing whether the potassium was ≥ 5.5 mEq/L and a DeepECG-Hypokalemia model diagnosing whether the potassium was < 3.5 mEq/L were developed adopting convolutional neural network architecture. The areas under the rece iver operating characteristic curve (AUROC) of the DeepECG-Hyperkalemia model in the internal testing cohort were 0.929, 0.912, and 0.887 for 12-lead, limb-lead, and lead I, respectively. The AUROC of the DeepECG-Hypokalemia model in the internal testing cohort were 0.925, 0.896, and 0.885 for 12-lead, limb-lead, and lead I, respectively. The survival rate of the group that the model predicted to be positive was lower than that of the group that predicted to be negative (p < 0.001). Furthermore, when we masked the QRS complex and T wave features, a decrease in the AUROC performance for both models was observed. We demonstrated the high diagnostic performance of deep learning models for the noninvasive screening of dyskalemia based on electrocardiograms. By applying these models in clinical practice, it will be possible to diagnose dyskalemia simply and quickly, thereby contributing to the improvement of patient outcomes.