Interpretable machine learning prediction models for 28-day mortality in critically ill patients with atrial fibrillation and acute kidney injury.

Objective: The study aims to develop and externally validate interpretable machine learning (ML) models for predicting 28-day mortality in critically ill patients with coexisting atrial fibrillation (AF) and acute kidney injury (AKI). Methods: We conducted a retrospective analysis using two large public databases, Medical Information Mart for Intensive Care IV (MIMIC-IV) and eICU Collaborative Research Database (eICU-CRD). Critically ill adults with both AF and AKI were included. In MIMIC-IV, patients were randomly divided into a training set and an internal test set for model development and evaluation. Nine ML algorithms were compared, and the best-performing model was further validated in the external eICU-CRD cohort. Model performance was primarily assessed using the area under the receiver operating characteristic curve (AUC). Interpretability was examined with the SHapley Additive exPlanations (SHAP) method, and an online risk calculator was developed to support clinical application. Results: A total of 11,510 patients from MIMIC-IV and 2565 patients from eICU-CRD were included. The GBM model achieved the best predictive performance, with an AUC of 0.856 (95% CI: 0.839-0.873) in the internal test cohort and 0.761 (95% CI: 0.740-0.783) in external validation. SHAP analysis identified anion gap, heart rate, and age as the most influential predictors of 28-day mortality. The developed online application enables individualized risk stratification, supporting clinical decision-making. Conclusions: We developed and externally validated interpretable ML models for 28-day mortality prediction in ICU patients with AF and AKI. These models may enhance prognostic accuracy, facilitate earlier intervention, and support clinical management in this high-risk population.