An Interpretable Machine Learning Model for Continuous Prediction of Acute Kidney Injury in Critically Ill Patients With Atrial Fibrillation.

The occurrence of acute kidney injury (AKI) in hospitalized patients with atrial fibrillation (AF) significantly increases the mortality risk. Currently, effective tools for early identification and dynamic predictive models remain lacking for this high-risk population, and the roles and underlying mechanisms of cardiorenal crosstalk remain underexplored. This study aims to develop a clinically interpretable model that continuously predicts the risk of AKI in patients with AF following admission to the intensive care unit (ICU). By retrospectively analyzing 19,347 critically ill patients with AF from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database, we developed five machine learning models and two deep learning models. Model performance was assessed using Decision Curve Analysis (DCA) and calibration curves. Simultaneously, the SHapley Additive exPlanations (SHAP) framework was utilized to quantify feature contributions and generate clinically interpretable visualizations. The results demonstrated that the eXtreme Gradient Boosting (XGBoost) model achieved the highest predictive performance, with an area under the receiver operating characteristic curve (AUROC) of 0.866 (95% CI: 0.864-0.868) and an area under the precision-recall curve (AUPRC) of 0.723 (95% CI: 0.718-0.729). Combined with real-time electronic health record (EHR) data, this proposed model can provide reliable reference evidence for clinicians to formulate reasonable clinical decisions and facilitate rational patient management strategies.