To investigate the predictive value of integrated model for end-stage liver disease (iMELD) scoring system for short-term and long-term prognosis of (ACLF) patients with chronic and acute hepatic failure and its guiding significance for the selection of artificial liver support system (ALSS).

From January 2003 to December 2007, the clinical data of 232 patients with chronic hepatitis B and acute hepatic failure (HBV-ACLF) in the Third People’s Hospital of Nantong were analyzed, retrospectively, who were divided into survival group (83 cases) and death group (149 cases). According to the baseline of liver and kidney function, the electrolytes, prothrombin time and complications, the scores of the model for end-stage liver disease (MELD), MELD-Na, integrated MELD (iMELD), Child-Turcotte-Pugh (CTP) and modified CTP (mCTP) were calculated, respectively. The survival rates of 90 d and 5 years were compared between the two groups during the follow-up. The accuracy of prognostic prediction of the five scoring systems was evaluated from three aspects: discriminant ability, calibration degree and overall performance. The accuracy and reliability of selecting artificial liver treatment of single index plasma prothrombin activity (PTA) and the five models under the guidance of critical value reported in literature were investigated. The net benefit of each model and "total therapy" strategy was calculated by the decision curve analysis (DCA) method, and the clinical significance of the model in predicting curative effect were evaluated.

MELD, MELD-Na, iMELD, CTP and mCTP scoring systems could make good use of data information and have a significant correlation with the prognosis of ACLF patients. The area under the receiver operating characteristic curve (AUC) for these models were 0.63, 0.64, 0.68, 0.62 and 0.64 for 90-day survival and 0.65, 0.71, 0.80, 0.78 and 0.78 for 5-year survival, respectively. IMELD was superior to other scoring systems (t= 8.318,P< 0.001). The best critical value of iMELD score has the best judgment of death risk. Nagelkerke’s R²and Brier scores indicated that iMELD had the best overall judgment ability. With 55 as the threshold, the sensitivity and specificity of short-term prognostic judgement of iMELD were 86.8% and 49.5%, respectively; AUC = 0.68. The predictive ability of iMELD was superior to that of single index of PTA (t= 5.866,P< 0.001). When the threshold probability (Pt) was 23%-65%, the net benefit of artificial liver therapy based on the iMELD score was higher than that of the "total treatment" strategy, equivalenting of reduce unnecessary artificial liver therapy to up to 20 cases per 100 patients. It is suggested that the use of iMELD model could save 20% of artificial liver resources, and not increase the probability of missing out the patients who need artificial liver treatment.

Integrated MELD model may be the best model to predict short-term and long-term prognosis in patients with ACLF, which may have clinical implications for diagnosis and treatment.