Renal function and electrolytes levels for estimation of in-hospital course in acute heart failure using artificial intelligence

Abstract

Acute heart failure is a major healthcare problem and is associated with severe clinical consequences. A significant number of patients with heart failure experience renal dysfunction, which can worsen during hospitalization due to acute heart failure. Studies have shown that in such patients, the risk of mortality and rehospitalization due to heart failure is increased. To reduce this burden on the healthcare system, models for estimating the risk of renal dysfunction in patients with acute heart failure are deemed necessary. Therefore, this study was conducted with the aim of determining renal function and electrolyte levels at the time of admission to estimate in-hospital outcomes in patients hospitalized with acute heart failure using artificial intelligence. Methods: This study was conducted from April 2023 to September 2023 at Shahid Madani Hospital in Tabriz, and patients admitted with chronic acute heart failure, new acute heart failure, or decompensated heart failure were included. Clinical information of the patients, including age, gender, cause of heart failure, underlying diseases, medications, laboratory and echocardiographic findings, as well as in-hospital outcomes such as length of hospital stay, in-hospital mortality, need for inotropes, mechanical ventilation, and dialysis were recorded. The relationship between levels of urea, creatinine, eGFR, sodium, and potassium with in-hospital outcomes was analyzed using artificial intelligence algorithms. The algorithms were trained using patient data, and their performance was evaluated based on criteria such as accuracy, sensitivity, and specificity. Ultimately, the algorithm with the best performance based on these criteria was selected.

Results: Our results showed that for predicting in-hospital mortality using serum electrolytes at the time of admission, the best artificial intelligence results were related to the SVM model with an accuracy of 97.9% and an F1 score of 0.97, and the Ensemble model with an accuracy of 97.4% and an F1 score of 0.97.