Machine learning to predict moderate-to-severe AKI in ICU acute pancreatitis patients

Acute kidney injury (AKI) is a serious complication in intensive care unit (ICU) patients with acute pancreatitis (AP), and moderate-to-severe AKI requires early clinical attention. This study developed and externally validated an interpretable machine learning model for early ICU risk stratification of moderate-to-severe AKI identified during the ICU stay, using clinical information available within 24 h after ICU admission. Data from 801 AP patients in MIMIC-IV were used for model development and internal validation, and 368 patients from eICU were used for external validation. Ten routinely available variables reflecting vital signs, comorbidities or early diagnostic conditions, laboratory findings, and early treatment requirements were included. Among the evaluated algorithms, performance differences were modest and no model was uniformly superior. XGBoost was retained as the primary model for detailed interpretation, achieving AUCs of 0.850 and 0.702 in the internal and external validation cohorts, respectively. SHAP-based explanations provided transparent individualized risk estimates. The model provides an interpretable summary of early-ICU AKI risk, but its temporal predictive value and clinical actionability require prospective validation.