Continuous Prediction of Acute Kidney Injury from Patients with Sepsis in ICU Settings: A Sequential Transduction Model Based on Attention

Abstract

Septic patients admitted to the intensive care unit (ICU) are highly susceptible to acute kidney injury (AKI), which leads to reduced survival in these patients. It is thus necessary to develop a model that can predict the risk of AKI in septic patients in real time. Although continuous or near-continuous risk assessment is likely necessary, few risk models have been designed for this purpose. Therefore, we constructed a model to continuously predict sepsis-induced AKI in ICU. Our proposed model optimally achieved an area under the receiver operating characteristic curve (AUROC) of 79.5 and an area under the precision-recall curve (AUPRC) of 65.0, performed better than other methods, including logistic regression, XGBoost, and RNN, on a full set of performance evaluation processes. Discrimination as well as DCA were also shown the proposed algorithm performed superior to other methods.