Development and Validation of a Clinical Predictive Model for Bacterial Infection in Hepatitis B Virus-Related Acute-on-Chronic Liver Failure.

Introduction: Bacterial infection is one of the most frequent complications in hepatitis B virus-related acute-on-chronic liver failure (HBV-ACLF), which leads to high mortality. However, a predictive model for bacterial infection in HBV-ACLF has not been well established. This study aimed to establish and validate a predictive model for bacterial infection in two independent patient cohorts.

Methods: Admission data from a prospective cohort of patients with HBV-ACLF without bacterial infection on admission was used for derivation. Bacterial infection development from day 3 to 7 of admission was captured. Independent predictors of bacterial infection development on multivariate logistic regression were used to develop the predictive model. External validation was performed on a separate retrospective cohort.

Results: A total of 377 patients were enrolled into the derivation cohort, including 88 patients (23.3%) who developed bacterial infection from day 3 to 7 of admission. On multivariate regression analysis, admission serum globulin (OR 0.862, 95% CI 0.822-0.904; P < 0.001), interleukin-6 (OR 1.023, 95% CI 1.006-1.040; P = 0.009), and C-reactive protein (OR 1.123, 95% CI 1.081-1.166; P < 0.001) levels were independent predictors for the bacterial infection development, which were adopted as parameters of the predictive model (GIC). In the derivation cohort, the area under the curve (AUC) of GIC was 0.861 (95% CI 0.821-0.902). A total of 230 patients were enrolled into the validation cohort, including 57 patients (24.8%) who developed bacterial infection from day 3 to 7 of admission, and the AUC of GIC was 0.836 (95% CI 0.782-0.881). The Hosmer-Lemeshow test showed a good calibration performance of the predictive model in the two cohorts (P = 0.199, P = 0.746). Decision curve analysis confirmed the clinical utility of the predictive model.

Conclusion: GIC was established and validated for the prediction of bacterial infection development in HBV-ACLF, which may provide a potential auxiliary solution for the primary complication of HBV-ACLF.