Development and validation of a diagnostic prediction model for acute liver injury in sepsis based on serum ferritin level.

BACKGROUND AND

objectiveThe early diagnosis and treatment of acute liver injury in sepsis are crucial determinants of the prognosis for patients with sepsis. The study aimed to investigate the early predictive value of serum ferritin level for acute liver injury in sepsis, to construct and validate a predictive model for acute liver injury in patients with sepsis.

methodThe training group data were selected from the Medical Information Mart for Intensive Care (MIMIC)-IV database. Least absolute shrinkage and selection operator (LASSO) regression, Boruta algorithm, univariate and multivariate logistic regression analyses were used to identify relevant independent risk factors, and these factors were incorporated into the predictive model. A sensitivity analysis was performed to validate the result. We constructed a nomogram prediction model based on ferritin level and evaluated its performance using the Receiver operating characteristic (ROC) curve, the area under the curve (AUC), calibration, and decision curve analysis (DCA). Collected patients in the intensive care unit (ICU) of the Xinjiang Uygur Autonomous Region People's Hospital who met the sepsis-3 diagnostic criteria as clinical validation group data, and clinical data were applied to the prediction model to validate its predictive performance.

resultsThis study included 1109 sepsis patients from the MIMIC database and 122 sepsis patients from the Xinjiang Uygur Autonomous Region People's Hospital. Based on the outcome of total bilirubin (TBIL) and/or alanine aminotransferase (ALT) after ICU admission, patients were divided into two groups: the sepsis-associated liver injury (SALI) group and the non-sepsis-associated acute liver injury (non-SALI) group. Analysis of the two groups' data revealed the following: the Lasso regression and Boruta algorithm results for the SALI group and the non-SALI group intersected to identify 12 differential factors; after propensity score matching (PSM), ferritin level remained statistically different between the two groups. Logistic regression analysis showed that machine ventilation, continuous renal replacement therapy (CRRT), vasoactive agent, alkaline phosphatase (ALP), international normalized ratio (INR), and ferritin level were independent risk factors for secondary acute liver injury (P < 0.05); ROC curve analysis showed that the AUC of the prediction models in the training and clinical validation groups were 0.765 and 0.773 respectively, with no statistically significant difference; Calibration curves were tightly aligned to the ideal line, indicating good agreement between predicted and actual outcomes; decision curve analysis provided evidence that the prediction model has high clinical utility with significant net benefit.

conclusionHigher serum ferritin level is an independent risk factor for acute liver injury in patients with sepsis. Based on the variable, we constructed a model including six clinical features to predict the risk of acute liver injury in patients with sepsis. After validation with an external clinical dataset, the model demonstrated excellent predictive ability and high value for assisting clinical decision-making.