Derivation and validation of prognostic models for predicting survival outcomes in Acute-on-chronic liver failure (ACLF) patients

Acute-on-chronic liver failure (ACLF) is a syndrome characterised by acute decompensation of chronic liver disease associated with high bacterial infection (BI) and short term mortality. However, many ACLF prognostic predictive modelsare complicated. The aim of this study is to develop prognostic models for ACLF patients to predict BI and mortality. We retrospective recruited 263 patients with ACLF from Shandong Provincial Hospital and Taizhou Enze Medical Center (Group) Enze Hospital. ACLF was defined according to the Asian Pacific Association for the Study of the Liver (APASL)criteria. Multivariable logistic regression was used to derive prediction models for occurring BI and 28-day mortality in ACLF patients. 97 of 263 patients (37%) occurred BI and 41 of 155 (26%) died within 28 days of admission. C-reactive protein (CRP), glucose, and albumin were the independent predictors for occurring BIduring the hospital stay. We also found that hepatic encephalopathy (HE), prothrombin time, activated partial thromboplastin time(APRI), and glucose were the independent predictors of 28-day mortality of ACLF patients. Using logistic regression model, we generated a new modified MELD model (M-MELD) by incorporating HE, APRI, and glucose. AUC of M-MELD model was 0.871, which were significantly higher than MELD score (AUC:0.734), MELD-Na score (AUC:0.742), and integrated MELD score (iMELD)(AUC:0.761). HE, MELD score, APRI, and blood glucose were independent risk factors for 28-day mortality of ACLF patients. The modified MELD model (M-MELD) by incorporating HE, APRI, and glucose has better discriminative performances compared with MELD in predicting 28-day mortality.