Clinical characteristics and construction of a predictive model for patients with sepsis related liver injury

Background: Sepsis-related liver injury (SRLI) is a common condition in critically ill patients, and it is associated with poor outcomes. Early identification of liver injury in sepsis can provide clinicians with the abundance of information for optimizing treatment strategies and improve quality of life. Therefore, the purpose of this study was to establish a predictive model to assess the early predictive value of liver injury in sepsis. Method: In this retrospective study, a total of 1116 patients with sepsis enrolled from the Biobank of First Affiliated Hospital of Xi'an Jiaotong University were included. According to the diagnosis of SRLI, all patients were divided into SRLI group and sepsis group. Multivariable analysis was performed using stepwise logistic regression to identify the independent risk factors of SRLI. Based on the results of multivariate regression analysis, we constructed a prediction model. The receiver operating characteristic curve (ROC) was used to determine the predictive value of the model on SRLI. Results: From December 2015 to December 2021, 1116 cases met the inclusion criteria and were included in this study. The median age was 58 years, of which 458 (41.04 %) were female. We discovered that procalcitonin (PCT), AST-to-platelet ratio index (APRI), alanine aminotransferase (ALT), lactate (Lac), blood urea nitrogen (BUN), Neutrophil and Cardiovascular disease were independent predictors for SRLI. We used to enter methods for constructing the predictive model and finally found that the indicators of model 2 (AUC = 0.832) were readily available and had good predictive value for SRLI. Furthermore, we also found that model 2 (AUC = 0.763), with a sensitivity of 81.4 %, demonstrated excellent predictive value for predicting 28-day mortality in patients with septic liver injury. Conclusions: This study explored the risk factors of SRLI, and established a prediction model that can accurately and effectively predict the occurrence of SRLI. Furthermore, model 2 is easy to obtain and has the highest sensitivity, which can contribute to early warning and appropriate clinical decision-making.