Development and external validation of a predictive model for in-hospital mortality in patients with liver cirrhosis and sepsis

Cirrhosis has an increasing prevalence globally, and sepsis is a common life-threatening comorbidity of cirrhosis. The cirrhotic population benefits less from the diagnostic decision-making in current guidelines for sepsis. To establish a predictive model and validate its efficiency for predicting the risk of all-cause in-hospital mortality (IHM) in cirrhosis with sepsis. We extracted data of cirrhosis patients with sepsis from the Medical Information Mart for Intensive Care IV (MIMIC-IV) and the eICU Collaborative Research Database (eICU-CRD). The MIMIC-IV dataset was assigned at 7:3 to a training set (n = 1701) and an internal validation set (n = 729), and the eICU-CRD dataset as an external validation set (n = 352). Statistically, variables were screened by LASSO regression. We assessed the model performance by ROC, calibration, and decision curve analysis (DCA) curves. Finally, we compared the nomogram with the SAPS-II score and conducted DeLong tests. The model achieved AUCs of 0.783 (95% CI 0.761–0.804), 0.763 (95% CI 0.729–0.796), and 0.745 (95% CI 0.692–0.797) in the training, internal validation, and external validation sets, respectively. Calibration curves showed good agreement. Decision curve analysis demonstrated favorable clinical utility. The nomogram is valuable in early identifying high-risk groups, implementing targeted interventions, reducing IHM, and ameliorating prognosis.