AIM To establish a clinical scoring model to predictrisk of acute-on-chronic liver failure (ACLF) in chronichepatitis B (CHB) patients.METHODS: This was a retrospective study of 1457patients hospitalized for CHB between October 2008and October 2013 at the Beijing Ditan Hospital, CapitalMedical University, China. The patients were dividedinto two groups: severe acute exacerbation (SAE)group (n = 382) and non-SAE group (n = 1075). TheSAE group was classified as the high-risk group basedon the higher incidence of ACLF in this group than inthe non-SAE group (13.6% vs 0.4%). Two-thirds ofSAE patients were randomly assigned to risk-modelderivation and the other one-third to model validation.Univariate risk factors associated with the outcomewere entered into a multivariate logistic regressionmodel for screening independent risk factors. Eachvariable was assigned an integer value based onthe regression coefficients, and the final score wasthe sum of these values in the derivation set. Modeldiscrimination and calibration were assessed using areaunder the receiver operating characteristic curve andthe Hosmer-Lemeshow test.RESULTS: The risk prediction scoring model included the following four factors: age ≥ 40 years, totalbilirubin ≥ 171 μmol/L, prothrombin activity 40%-60%,and hepatitis B virus DNA 〉 107 copies/mL. The sumrisk score ranged from 0 to 7; 0-3 identified patientswith lower risk of ACLF, whereas 4-7 identified patientswith higher risk. The Kaplan-Meier analysis showedthe cumulative risk for ACLF and ACLF-related deathin the two risk groups (0-3 and 4-7 scores) of theprimary cohort over 56 d, and log-rank test revealeda significant difference (2.0% vs 33.8% and 0.8%vs 9.4%, respectively; both P 〈 0.0001). In thederivation and validation data sets, the model hadgood discrimination (C index = 0.857, 95% confidenceinterval: 0.800-0.913 and C index = 0.889, 95%confidence interval: 0.820-0.957, respectively) andcalibration demonstrated by the Hosmer-Lemeshowte