Abstract
BACKGROUND & AIMS: Validated cutoffs for liver failure with cirrhosis are required to define acute-on-chronic liver failure. We sought to determine the cutoff for serum bilirubin and international normalized ratio (INR), and influence of complications on inpatient mortality with the goal of defining liver failure and inpatient mortality prediction. METHODS: The derivation cohort for variables associated with mortality included 7239 patients with cirrhosis admitted nonelectively worldwide (CLEARED consortium) and followed until death/discharge. Admission variables, MELD 3.0, original MELD, and hepatic encephalopathy (HE) were used to create a decision rule to classify inpatient death using cutoffs to maximize negative predictive value and decision-tree models sorted each predictor's relative contribution. For external validation, data from 3 Mayo Clinic sites were used. RESULTS: A total of 7239 patients were included from 115 centers across 6 continents among whom 808 (11.2%) died in-hospital. Cutoffs maximizing negative predictive value for in-hospital transplant-free mortality were bilirubin ≤7.5 mg/dL, creatinine ≤1.5 mg/dL, INR ≤1.5, MELD ≤27, and MELD 3.0 ≤28. On decision tree analysis, the relative contributions to mortality were HE (42.2%), infection (22.2%), INR (18.6%), bilirubin (12.7), and creatinine (4.3%). Using only liver-specific variables, patients with bilirubin ≤7.5, INR ≤1.5, and without HE had only 3.3% mortality, whereas patients with bilirubin >7.5, INR >1.5, and HE had 33% mortality. External validation in 1634 patients from Mayo transplant evaluation cohort showed similar patterns for inpatient mortality prediction. CONCLUSIONS: Cutoffs associated with mortality are admission bilirubin >7.5 mg/dL, INR >1.5, creatinine >1.5 mg/dL, MELD >27, and MELD 3.0 >28. Liver failure may therefore be identified by the serum bilirubin >7.5 mg/dL and INR >1.5, with HE and admission infections as contributors toward mortality.