Development and validation of a prediction model for acute kidney injury in cirrhosis with normal serum creatinine levels

Acute kidney injury (AKI) is a serious complication in decompensated cirrhosis (DC). Conventional kidney function markers, including serum creatinine (sCr) and estimated glomerular filtration rate (eGFR), are often inaccurate in these patients. We aimed to develop a predictive model for AKI using routine admission data. We retrospectively analyzed patients with DC and normal sCr admitted to a local tertiary hospital (Dec 2016-Dec 2022). AKI was defined per International Club of Ascites criteria. Predictors were selected using the least absolute shrinkage and selection operator regression and incorporated into a logistic model, internally validated via enhanced bootstrap and externally validated in the MIMIC-III cohort. Model performance was assessed using AUROC, calibration, and decision curve analysis (DCA). A user-friendly nomogram was constructed. Of 1,863 patients screened, 1,091 met criteria; 122 (10.8%) developed AKI. AKI patients were older, had longer stays, worse liver function, more complications (hepatic encephalopathy, infections), and higher rates of hypertension (HT) and type 2 diabetes mellitus (DM2). sCr and eGFR had poor discrimination for AKI. The final model included age, neutrophil count, direct bilirubin, Child-Turcotte-Pugh score, DM2, and HT, achieving AUROC 0.755 (internal validation: AUROC 0.733, sensitivity 0.697, specificity 0.626). DCA showed greater clinical benefit than treat-all or treat-none strategies. External validation in 810 MIMIC-III patients yielded AUROC 0.729; bootstrap validation confirmed robustness (AUROC 0.752, sensitivity 0.748, specificity 0.677). We developed and validated a simple admission-based model for predicting AKI in DC, supporting its potential use for risk stratification and early intervention.