PrognostiC Model F Complica For Advan Ated With I Ced Liver Infection Cirrhosis

Background & Aims: Cirrhosis patients have increased risk of bacterial infections. Bacterial infections are the common causes of cirrhotic complications, such as gastrointestinal bleeding, hyponatremia and hepato renal syndrome, and organs failure, such as liver failure, renal failure, shock or multiple organ failure. Bacterial infection is a leading cause of mortality in patients with advanced cirrhosis. We aimed to analyze the character of bacterial infections and organs failure and develop predictive model for 3-month mortality in Child-Pugh class C cirrhosis patients hospitalized with infection for optimizing treatment and improving outcomes.Methods: The study screened 1819 cases of cirrhsis patients, 519 cases from Tianjin Third Central Hospital, 1015 cases from Mayo Clinic, U.S. and 285 from Seoul St. Mary’s Hospital, Korea. According the inclusion and exclusion criteria 417 cases were enrolled in this study finally, 244 patients from Tianjin Third Central Hospital, China(between January 2010 and April 2013, test cohort), 91 patients from Mayo Clinic, Rochester, MN, U.S.(between January 2007 and December 2013, validation cohort 1) and 82 patients from Seoul St. Mary’s Hospital, Korea(between January 2011 and May 2014, validation cohort 2). Clinical informations were retrospectively abstracted from above patients. Factors associated with mortality were determined using univariate and multivariate logistic regression. The model for predicting 3-month mortality was then constructed using decision tree. The predictive performance of the model was compared to that of the MELD, MELD-Na, Child-Pugh, Chronic Liver Failure–Sequential Organ Failure Assessment(CLIF-SOFA) and the north American consortium for the study of end-stage liver disease(NACSELD) models. The model was further validated in validation cohort 1 and validation cohort 2.Results: In test cohort, validation cohort 1 and validation cohort 2 the 3-month mortality was 58%, 58% and 54%, respectively; the mean age was 52.1±11.6, 57.5±11.4 and 58.3±12.7, respectively; MELD score was 26±8, 30±9 and 28±12, respectively; the most common etiology was hepatitis B, alcoholic and hepatitis B, respectively; the most common bacterial infections were pneumonia and spontaneous bacterial peritonitis(SBP), SBP and urinary tract infection(UTI), SBP and pneumonia,respectively; the proportion of acute on chronic liver failure was 74%, 78% and 73%, respectively; the most common extra-hepatic organ failure was all renal failure. In test cohort, respiratory failure, renal failure, international normalized ratio, total bilirubin, and neutrophil percentage were determinants of 3-month mortality, with odds ratios of 16.6(1.9-143.1), 3.3(1.4-7.8), 2.0(1.3-3.1), 1.1(1.0-1.1) and 1.03(1.0-1.1), respectively(P<0.05) after univariate and multivariate logistic regression analysis. These parameters were incorporated in the decision tree model which best area under receiver operating characteristic(AUROC) of 0.804(0.748-0.859). The decision tree model could divide the patients into low, intermediate and high risk subgroups, which 28 days and 3-month mortality were 14% and 27%, 48% and 63%, 77% and 87%, respectively. Compared to MELD, MELD-Na, Child-Pugh, CLIF-SOFA and the NACSELD models the decision tree model had the best AUROC of 0.804(0.748-0.859), best sensitivity of 80%, best overall accuracy of 75% and best Youden index of 0.488. The model had excellent reproducibility in the U.S.(AUROC 0.808(0.706-0.910)) and Korea cohort(AUROC 0.809(0.711-0.906)) and was better than MELD score(U.S. AUROC 0.756(0.643-0.868), Korea cohort AUROC 0.771(0.664-0.878)).Conclusion: The mortality was higher in end-stage cirrhosis patients complicated with infection. The common infections were pneumonia, SBP and UTI. The common organs failure were acute on chronic liver failure and renal failure. The proposed model reliably predicts survival of advanced cirrhosis with infection in both Asian and U.S. population.