Prediction of hyperkalemia in ESRD patients by identification of multiple leads and multiple features on ECG

• DOI: 10.1080/0886022x.2023.2212800
• Publication Date: 2023-05-18T10:17:44Z
• Authors: Daojun Xu; Bin Zhou; Jiaqi Zhang; Chenyu Li; Chen Guan; Yuxuan Liu; Lin Li; Haina Li; Li Cui; Lingyu Xu; Hang Liu; Li Zhen; Yan Xu
• Keywords: end-stage renal disease; logistic regression; Diseases of the genitourinary system. Urology; 3. Good health; Electrocardiogram; Machine Learning; Electrocardiography; machine learning; noninvasive hyperkalemia prediction; Clinical Study; Potassium; Humans; Hyperkalemia; Kidney Failure, Chronic; RC870-923

Patients with end-stage renal disease (ESRD) especially those undergoing dialysis have a high prevalence of hyperkalemia, which must be detected and treated immediately. But the initial symptoms hyperkalemia are insidious, traditional laboratory serum potassium concentration testing takes time. Therefore, rapid real-time measurement is urgently needed. In this study, different machine learning methods were used to make predictions degrees by analyzing ECG.A total 1024 datasets ECG concentrations analyzed from December 2020 2021. The data scaled into training test sets. Different models (LR, SVM, CNN, XGB, Adaboost) built for dichotomous prediction 48 features chest leads V2-V5. performance was also evaluated compared using sensitivity, specificity, accuracy, F1 score AUC.We constructed predict LR four other common machine-learning methods. AUCs ranged 0.740 (0.661, 0.810) 0.931 (0.912,0.953) when as diagnostic threshold respectively. As raised, accuracy precision model decreased various degrees. And AUC performed less well than predicting mild hyperkalemia.Noninvasive can achieved specific waveforms on overall, XGB had higher in but SVM better more severe hyperkalemia.