Abstract
Acute kidney injury (AKI) is a common and serious complication in patients with malignant obstructive jaundice (MOJ), yet no predictive model exists for this specific population. This retrospective study included 557 hospitalized MOJ patients, with 103 developing AKI. Propensity score matching was used to control confounding, and 103 matched pairs were analyzed. Least absolute shrinkage and selection operator (LASSO) regression and multivariate logistic regression were used to develop the prediction model. Model performance was evaluated using the area under the curve (AUC), concordance index, Brier score, calibration curve, decision curve analysis, and clinical impact curve. Seven variables were included in the final model: neutrophil-to-lymphocyte ratio (NLR), C-reactive protein (CRP), uric acid (UA), total bilirubin (TBil), potassium (K), carbon dioxide combining power (CO(2)CP), and prothrombin time (PT). The model showed excellent discrimination (AUC = 0.891) and clinical usefulness, with good calibration demonstrated by the Hosmer-Lemeshow test (p = 0.567) and internal validation using bootstrap resampling (B = 1000). A nomogram was constructed for individualized risk assessment. Risk stratification based on predicted probabilities showed a progressive increase in AKI incidence across tertiles (11.6%, 47.1%, and 91.3%). This model provides an accurate and practical tool for predicting AKI in MOJ patients using routine clinical parameters. Prospective multicenter studies are needed to confirm generalizability.