Review of Acute Kidney Injury Classification Using Machine Learning

The incidence of acute kidney injury (AKI) across hospitalized patients, especially in the intensive care unit (ICU) is worrying due to its prevalence and association with mortality. The sudden decrease in kidney function can be identified by an increase in serum creatinine or decreasing urine output. The severity of AKI stages can be defined according to Kidney Disease: Improving Global Outcomes (KDIGO) classifications. Several studies have reported AKI associated risk factors such as sepsis and rates of mortality. Due to this concern, machine learning has been implemented to predict AKI incidences utilizing several techniques such as Decision Tree, Random Forest, Support Vector Machine, k-Nearest Neighbour, and Gradient Boosting Method. The performances of these models were measured by area under the receiver operating characteristic curve (AUROC). This review examines ICU-based AKI incidences and the use of machine learning techniques to predict AKI incidences. It highlights the complementary data used to perform the prediction and its performance based on AUROC. The models studied in this review demonstrated AUROCs between 0.57 to 0.95. Diabetes and hyperglycemia have been demonstrated as significant risk factors for AKI in the ICU. Hence, insulin sensitivity representing a patient's metabolic variation is suggested as another variable to predict AKI incidence.