Discharge renal function and electrolyte levels for estimation of early outcome in hospitalized patients with acute heart failure using artificial intelligence

Abstract

Heart failure, re-hospitalization, and mortality are the main problems in the field of cardiology. On the other hand, it is simply not possible to identify these high-risk patients, and risk-determination methods have not been completely successful in these patients. Due to the close relationship between the heart and kidney, the use of renal indicators such as urea, creatinine, and serum electrolytes such as sodium and potassium can be useful indicators in this field. Considering the previous studies, a specific study on the combination of kidney and electrolyte disorders to predict readmission and early mortality has not been done. Because these tests are cheap and available, they can help identify high-risk patients during discharge so that more measures can be taken in this group. Methods: This cross-sectional study was conducted from September 2023 to March 2024 at Shahid Madani Hospital in Tabriz, and patients hospitalized with chronic, new onset, or decompensated heart failure were included in the study. The clinical information of the patients including age, sex, cause of heart failure, underlying diseases, medications, laboratory findings, and echocardiography, as well as one-month follow-up, were recorded. The relationship between renal function and the level of electrolytes at the time of discharge with the early prognosis of patients including readmission and mortality within one month after discharge was investigated using artificial intelligence (AI). Algorithms were trained using data from patients and their performance was evaluated based on criteria such as accuracy, sensitivity, and specificity. Finally, the algorithm that performed best based on these criteria was selected. Results: The results showed that for predicting readmission within one month after discharge using serum electrolytes, the best artificial intelligence results are related to the SVM model with 93.8% accuracy and F1 score equal to 0.94, and the Ensemble model with 94.6% accuracy and F1 -score was equal to 0.94.