Development and validation of a nomogram based on high-sensitivity C-reactive protein-to-high-density lipoprotein cholesterol ratio (HCHR) for predicting secondary bacterial infections in patients with acute-on-chronic liver failure: a retrospective study

Secondary bacterial infections are common in acute-on-chronic liver failure (ACLF) and are associated with poor clinical outcomes and increased short-term mortality. Reliable tools for early risk stratification remain limited. Therefore, we evaluated the ability of the high-sensitivity C-reactive protein/high-density lipoprotein cholesterol ratio (HCHR) to predict in-hospital secondary bacterial infections in ACLF and constructed an individualized prediction model based on HCHR. In this retrospective study, 524 patients with ACLF were randomly divided into training ( n = 367) and validation ( n = 157) sets. Predictors were selected using univariate analysis and LASSO regression. Seven machine learning (ML) algorithms were compared using discrimination, calibration, and clinical utility, and a nomogram was developed from the optimal model. Secondary bacterial infections occurred in 51.0% of patients. HCHR was significantly higher in the infection group and associated with infection risk (OR = 1.028, 95% CI: 1.023–1.034, p < 0.001), showing a dose-response relationship. The final model incorporated HCHR, ascites, hepatic encephalopathy (HE), and hepatorenal syndrome (HRS). Among all models, the HCHR-based logistic regression (LR) model was selected as the final model because of its favorable discrimination (AUC = 0.868, 95% CI: 0.804–0.920), calibration (Brier score = 0.142), interpretability, and clinical utility. The model significantly outperformed the MELD score ( p < 0.001) and enabled effective risk stratification, with infection rates of 89.1% in the high-risk group versus 24.7% in the low-risk group. HCHR is a readily available biomarker associated with the risk of in-hospital secondary bacterial infection in patients with ACLF. The HCHR-based nomogram demonstrated promising performance in internal validation and may serve as a preliminary tool for individualized risk stratification. Not applicable.