Development and validation of an early diagnostic model to distinguish bacterial infection in patients with acute-on-chronic liver disease (AoCLD) and systemic inflammatory response syndrome (SIRS): a post-hoc analysis of a prospective multicenter cohort

This is the authors' abstract. We don't add key points for this paper.

A practical diagnostic model was developed and validated using the multicenter CATCH-LIFE cohort to distinguish bacterial infection in patients with acute-on-chronic liver disease (AoCLD) and systemic inflammatory response syndrome (SIRS). Based on five routine clinical indicators—jaundice, ascites, neutrophil count, albumin, and CRP—the model demonstrated high accuracy and enables early risk stratification to guide timely antibiotic decisions. • Developed an early diagnostic model to distinguish infection in AoCLD with SIRS. • Based on a large, prospective multicenter cohort (CATCH-LIFE, n = 515) • Incorporated five routinely available clinical and laboratory variables. • Model enables infection risk stratification into low, intermediate, and high groups. • Supports early identification and targeted intervention to guide antibiotic use. Early differentiation between infectious and non-infectious systemic inflammatory response syndrome (SIRS) in patients with acute-on-chronic liver disease (AoCLD) remains challenging. This study aimed to develop and validate an early diagnostic model to accurately identify infection status in AoCLD patients with SIRS, thereby guiding targeted anti-infective therapy and improving clinical management. Based on two multicenter prospective cohorts from the Chinese Acute-on-Chronic Liver Failure (CATCH-LIFE) study, 515 AoCLD patients with SIRS were randomly divided into a training cohort (n = 361) and a validation cohort (n = 154). In the training cohort, predictive factors for infection were screened using logistic regression, least absolute shrinkage and selection operator (LASSO) regression, multicollinearity analysis, and stepwise regression. Restricted cubic splines (RCS) were employed to explore nonlinear relationships and interaction effects. A nomogram model was subsequently constructed. The model’s performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA), with internal validation via bootstrap method. The final model incorporated five clinical indicators: jaundice, ascites, neutrophil count (N), albumin (ALB), and C-reactive protein (CRP). CRP exhibited a significant nonlinear relationship with infection risk ( P < 0.001) and was included as a spline term, while an interaction term between ALB and jaundice was incorporated. The model demonstrated excellent discrimination, with area under the ROC curve (AUC) of 0.840 (95 % CI: 0.788–0.888) in the training cohort and 0.865 (95 % CI: 0.803–0.927) in the validation cohort, significantly outperforming single biomarkers. Calibration curves showed high consistency between predicted and observed probabilities. DCA indicated substantial clinical net benefit across threshold probabilities. A clinical decision pathway was developed based on the nomogram’s risk prediction and optimal DCA threshold. The early diagnostic model developed in this study effectively differentiates between infectious and non-infectious SIRS in AoCLD patients, potentially reducing unnecessary antimicrobial exposure and improving clinical outcomes.