The study focused on the early prediction of sepsis-related liver injury (SRLI) in ICU patients using machine learning models. They used data from the MIMIC-IV database (9,434 patients with sepsis) for training and internal validation, plus external validation on 120 patients from Nanjing Jinling Hospital. They developed seven ML models, of which Random Forest (RF) was the best with ROC-AUC 0.867 and PR-AUC 0.392 in internal validation and ROC-AUC 0.862 with PR-AUC 0.735 in external validation. The most important predictors in the RF model were total bilirubin, international normalized ratio, SOFA score, LODDS score, and prothrombin time in the first 24 hours after admission. Decision curve analysis confirmed the clinical utility of the RF model across a wide range of risk thresholds. Prothrombin time has been shown to be a significant marker for early risk stratification.