Research Article Cirrhosis and Liver Failure The PREDICT study uncovers three clinical courses of acutely decompensated cirrhosis that have distinct pathophysiology Graphical abstract Authors Jonel Trebicka, Javier Fernandez, Maria Papp, ., Richard Moreau, Paolo Angeli, Different clinical courses of acutely decompensated cirrhosis Vicente Arroyo Infections ACLF Correspondence Death jonel.trebicka@kgu.de (J. Trebicka). Pre-ACLF Lay summary Herein, we describe, for the first time, 3 dif- ferent clinical courses of acute decom- pensation (AD) of cirrhosis after hospital admission. The first clinical course includes Unstable decompensated cirrhosis patients who develop acute-on-chronic liver failure (ACLF) and have a high short-term risk of death – termed pre-ACLF. The second clinical course (unstable decompensated cir- Stable decompensated cirrhosis rhosis) includes patients requiring frequent hospitalizations unrelated to ACLF and is associated with a lower mortality risk than pre-ACLF. Finally, the third clinical course 0 90 180 270 360 Days (stable decompensated cirrhosis), includes two-thirds of all patients admitted to hospi- tal with AD – patients in this group rarely require hospital admission and have a much Highlights lower 1-year mortality risk.  Patients with acutely decompensated cirrhosis without ACLF develop 3 different clinical courses.  Patients with pre-ACLF develop ACLF within 90 days and have high systemic inflammation and mortality.  Patients with unstable decompensated cirrhosis suffer from complications of severe portal hypertension.  Patients with stable decompensated cirrhosis have less frequent complications and lower 1-year mortality risk. https://doi.org/10.1016/j.jhep.2020.06.013 e57 © 2020 European Association for the Study of the Liver. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). J. Hepatol. 2020, 73, 842–854 Research Article Cirrhosis and Liver Failure The PREDICT study uncovers three clinical courses of acutely decompensated cirrhosis that have distinct pathophysiologyq Jonel Trebicka1,2,*, Javier Fernandez1,4, Maria Papp5, Paolo Caraceni6, Wim Laleman13, Carmine Gambino7, Ilaria Giovo8, Frank Erhard Uschner2, Cesar Jimenez9, Rajeshwar Mookerjee10, Thierry Gustot11, Agustin Albillos12, Rafael Bañares14, Martin Janicko15, Christian Steib16, Thomas Reiberger17, Juan Acevedo18, Pietro Gatti19, William Bernal20, Stefan Zeuzem2, Alexander Zipprich21, Salvatore Piano7, Thomas Berg22, Tony Bruns23,34, Flemming Bendtsen24, Minneke Coenraad25, Manuela Merli26, Rudolf Stauber27, Heinz Zoller28, José Presa Ramos29, Cristina Solè4, Germán Soriano30, Andrea de Gottardi31, Henning Gronbaek32, Faouzi Saliba33, Christian Trautwein34, Osman Cavit Özdogan35, Sven Francque36, Stephen Ryder37, Pierre Nahon38, Manuel Romero-Gomez39, Hans Van Vlierberghe40, Claire Francoz41,42, Michael Manns43, Elisabet Garcia1, Manuel Tufoni6, Alex Amoros1, Marco Pavesi1, Cristina Sanchez1, Anna Curto1, Carla Pitarch1, Antonella Putignano11, Esau Moreno1, Debbie Shawcross20, Ferran Aguilar1, Joan Clària1,4, Paola Ponzo8, Christian Jansen3, Zsuzsanna Vitalis5, Giacomo Zaccherini6, Boglarka Balogh5, Victor Vargas9, Sara Montagnese7, Carlo Alessandria8, Mauro Bernardi6, Pere Ginès4, Rajiv Jalan1,10, Richard Moreau1,41,42, Paolo Angeli1,7,†, Vicente Arroyo1,†, for the PREDICT STUDY group of the EASL-CLIF Consortium‡ 1 European Foundation for Study of Chronic Liver Failure, EF-Clif, Barcelona, Spain; 2JW Goethe University Hospital, Frankfurt, Germany; 3 University Hospital Bonn, Bonn, Germany; 4Hospital Clinic, IDIBAPS and CIBEehd, Barcelona, Spain; 5University of Debrecen, Faculty of Medicine, Institute of Medicine, Department of Gastroenterology, Debrecen, Hungary; 6University of Bologna, Bologna, Italy; 7Department of Medicine, University of Padova, Padova, Italy; 8Division of Gastroenterology and Hepatology, Città della Salute e della Scienza Hospital, University of Torino, Italy; 9Liver Unit, Hospital Vall d’Hebron, Universitat Autònoma de Barcelona, CIBEREHD, Barcelona, Spain; 10UCL Medical School, Royal Free Hospital, London, United Kingdom; 11Universite Libre de Bruxelles, Bruxelles, Belgium; 12Department of Gastroenterology, Hospital Universitario Ramón y Cajal, IRYCIS, University of Alcalá, CIBEREHD, Madrid, Spain; 13University of Leuven, Leuven, Belgium; 14Gastroenterology and Hepatology Department, Hospital General Universitario Gregorio Marañón, Facultad de Medicina, Universidad Complutense, CIBERehd, Madrid, Spain; 15Pavol Jozef Safarik University in Kosice, Kosice, Slovakia; 16Department of Medicine II, University Hospital Munich, Munich, Germany; 17Medical University of Vienna, Vienna, Austria; 18University Hospitals Plymouth NHS Trust, Plymouth, UK; 19Internal Medicine PO Ostuni, ASL Brindisi, Italy; 20King’s College Hospital, London, United Kingdom; 21University Hospital Halle-Wittenberg, Halle (Saale), Germany; 22Division of Hepatology, Department of Medicine II, Leipzig University, Medical Center, Leipzig, Germany; 23Jena University Hospital, Jena, Germany; 24Hvidovre University Hospital, Hvidovre, Denmark; 25Leiden University Medical Center, Leiden, Netherlands; 26Department of Translational and Precision Medicine, Universitá Sapienza Roma, Roma, Italy; 27Medical University of Graz, Graz, Austria; 28Medical University of Innsbruck, Innsbruck, Austria; 29CHTMAD Vila Real, Vila Real, Portugal; 30Hospital de la Santa Creu i Sant Pau and CIBERehd, Barcelona, Spain; 31University Clinic of Visceral Surgery and Medicine-Inselspital, Bern and Ente Ospedaliero Cantonale, Universita della Svizzera Italiana, Lugano, Switzerland; 32Aarhus University Hospital, Aarhus, Denmark; 33AP-HP Hôpital Paul Brousse, Centre Hépato-Biliaire, Universite Paris Saclay, INSERM Unit 1193, Villejuif, France; 34Aachen University Hospital, Aachen, Germany; 35 Marmara University, Istanbul, Turkey; 36University Hospital Antwerp, Antwerpen, Belgium; 37NIHR Biomedical Research Centre at Nottingham University Hospitals NHS Trust and the University of Nottingham, Nottingham, United Kingdom; 38AP-HP, Hôpital Jean Verdier, Service d’Hépatologie, Bondy; Université Paris 13, Sorbonne Paris Cité, “Equipe labellisée Ligue Contre le Cancer”, Saint-Denis; Inserm, UMR- 1162, “Génomique fonctionnelle des tumeurs solides”, Paris, France; 39Virgen del Rocío University Hospital, Sevilla, Spain; 40Ghent University Keywords: Chronic liver disease; Non-elective admission; Acute complications; Outcome; Risk factors. Received 16 March 2020; received in revised form 5 June 2020; accepted 6 June 2020; available online 13 July 2020 q Guest Editor: Dominique Valla. * Corresponding author. Address: European Foundation for Study of Chronic Liver Failure, EF-Clif, Travesera de Gracia 11, 7th Floor, 08021 Barcelona, Spain. Tel.: +34 93 227 14 11; fax +34 93 227 14 19. E-mail address: jonel.trebicka@kgu.de (J. Trebicka). † Denotes shared last authorship. ‡ EASL-CLIF consortium collaborators listed at the end of the manuscript. https://doi.org/10.1016/j.jhep.2020.06.013 Journal of Hepatology 2020 vol. 73 j 842–854 41 Hospital, Ghent, Belgium; APHP, Hôpital Beaujon, Service d’Hépatologie, Clichy, France; 42Inserm, Université de Paris, Centre de Recherche sur LInflammation, Paris, France; 43Hannover Medical School, Hannover, Germany See Editorial, pages 755–756 Background & Aims: Acute decompensation (AD) of cirrhosis is AD is an extremely relevant feature during the clinical course of defined as the acute development of ascites, gastrointestinal cirrhosis. The first episode of AD signals the transition from hemorrhage, hepatic encephalopathy, infection or any combi- compensated to decompensated cirrhosis.4 Decompensated nation thereof, requiring hospitalization. The presence of organ cirrhosis is characterized by recurrent episodes of AD. Finally, failure(s) in patients with AD defines acute-on-chronic liver recent data from the CANONIC study have shown that AD has 2 failure (ACLF). The PREDICT study is a European, prospective, distinct clinical presentations, depending on the presence or observational study, designed to characterize the clinical course absence of organ failures and the grade of systemic inflamma- of AD and to identify predictors of ACLF. tion.5–8 The presence of both organ failures and high-grade sys- Methods: A total of 1,071 patients with AD were enrolled. We temic inflammation is the hallmark of acute-on-chronic liver collected detailed pre-specified information on the 3-month failure (ACLF), a syndrome associated with a very high 28-day period prior to enrollment, and clinical and laboratory data at mortality rate, while AD is associated with moderate systemic enrollment. Patients were then closely followed up for 3 months. inflammation and a low 28-day mortality rate. Systemic inflam- Outcomes (liver transplantation and death) at 1 year were also mation in AD and ACLF frequently develops in association with recorded. exogenous precipitating events (mainly bacterial infections or Results: Three groups of patients were identified. Pre-ACLF pa- acute alcoholic liver injury). However, it might also be secondary tients (n = 218) developed ACLF and had 3-month and 1-year to translocation of intestinal bacterial immunogenic material to mortality rates of 53.7% and 67.4%, respectively. Unstable the systemic circulation.9,10 Systemic inflammation may induce decompensated cirrhosis (UDC) patients (n = 233) required > −1 organ dysfunction/failure via a direct immunopathological effect readmission but did not develop ACLF and had mortality rates of on peripheral organs or via mitochondrial dysfunction, both of 21.0% and 35.6%, respectively. Stable decompensated cirrhosis which have been identified in decompensated cirrhosis.8 (SDC) patients (n = 620) were not readmitted, did not develop The CANONIC study was specifically designed to characterize ACLF and had a 1-year mortality rate of only 9.5%. The 3 groups ACLF but did not provide detailed information on the clinical differed significantly regarding the grade and course of systemic context prior to and after ACLF and AD development. Yet, the inflammation (high-grade at enrollment with aggravation during CANONIC study showed that patients with AD had very low follow-up in pre-ACLF; low-grade at enrollment with subsequent mortality rate (2%) at 28 days but a substantial mortality rate steady-course in UDC; and low-grade at enrollment with sub- (10%) at 90 days, suggesting a heterogeneity of clinical course in sequent improvement in SDC) and the prevalence of surrogates patients with AD. Detailed information on this period is an of severe portal hypertension throughout the study (high in UDC unmet medical need for the rational management of patients vs. low in pre-ACLF and SDC). with AD and the prevention of ACLF development. Conclusions: Acute decompensation without ACLF is a hetero- To answer these questions, we designed the PREDICT study geneous condition with 3 different clinical courses and 2 major (PREDICTing Acute-on-Chronic Liver Failure), the second pathophysiological mechanisms: systemic inflammation and prospective large-scale observational investigation performed portal hypertension. Predicting the development of ACLF by the European Association for the Study of the Liver remains a major future challenge. (EASL)-Chronic Liver Failure (CLIF) Consortium. It included 1,071 ClinicalTrials.gov number: NCT03056612. patients with cirrhosis hospitalized for the treatment of an Lay summary: Herein, we describe, for the first time, 3 different episode of AD without ACLF. The current article reports the clinical courses of acute decompensation (AD) of cirrhosis after results of the first study derived from this investigation, the aim hospital admission. The first clinical course includes patients of which was to characterize the clinical course and patho- who develop acute-on-chronic liver failure (ACLF) and have a physiology of AD, and to predict the development of ACLF. high short-term risk of death – termed pre-ACLF. The second clinical course (unstable decompensated cirrhosis) includes Patients and methods patients requiring frequent hospitalizations unrelated to ACLF Study oversight and is associated with a lower mortality risk than pre-ACLF. The PREDICT study is a European, multicenter, prospective, Finally, the third clinical course (stable decompensated observational study performed in 48 hospitals. Each hospital had cirrhosis), includes two-thirds of all patients admitted to hospital a liver unit, specific ward(s) for liver patients and intensive care with AD – patients in this group rarely require hospital admis- facilities, and all of them had access to a liver transplantation sion and have a much lower 1-year mortality risk. program. The study protocol (available with the full text of this © 2020 European Association for the Study of the Liver. Published by article) was approved by the institutional review board (IRB) at Elsevier B.V. This is an open access article under the CC BY-NC-ND each participating center. Patients were screened and enrolled license (http://creativecommons.org/licenses/by-nc-nd/4.0/). from March 2017 to July 2018. Written informed consent was obtained from patients or their legal surrogates before enroll- Introduction ment. An investigator was responsible for enrolling patients in Acute decompensation (AD) of cirrhosis is defined as the acute the study at each center, ensuring adherence to the protocol, and development of ascites, hepatic encephalopathy, gastrointestinal completing the electronic case-report form. Data were continu- hemorrhage or bacterial infections or any combination thereof.1–3 ously monitored on-line by the Data Management Center of the Journal of Hepatology 2020 vol. 73 j 842–854 843 Research Article Cirrhosis and Liver Failure Past medical history: a) the time point of the onset of decompensation 3-month period prior to enrollment V: planned visits E: visits in case of ACLF Medical history during the last 3 months: A: additional planned visits b) complications of AD AD: readmission visits c) treatment of complications d) any hospitalization OBSERVATIONAL PERIOD Enrollment No ACLF (Visit V1) CLIF-C AD <50 CLIF-C AD ≥50 3-month follow-up period Visit d7-10 (V2) Visit d7-10 (V2) Week 4/8 (A1/A2) ACLF (E1) Re-admission (AD) Visit d7-10 (V2) Week 12 (V3) Week 12 (V3) Outcome in Outcome in Outcome in 6 and 12 months 6 and 12 months 6 and 12 months Fig. 1. Scheme of visits and collection of data and samples during the 6-month observational period. The 6-month period which included the 3-month period prior to enrollment, the enrollment visit, and the 3-month follow-up period after enrollment. At enrollment patients were initially stratified into 2 12 groups based on the risk of ACLF development: high-risk group (CLIF-C AD-score > −50) and low-risk group (CLIF-C AD-score <50). In the high-risk group, the scheduled visits were performed at enrollment (visit 1) and 1, 4, 8 and 12 weeks after enrollment (visits 2–5). In the low-risk group, scheduled visits were performed only at enrollment (visit 1) and week 1 (visit 2) and 12 (visit 5) after enrollment. Any patient of both groups developing ACLF within the follow-up period received unplanned ACLF visits at the time of diagnosis of ACLF and 7 days later. Additional study visits were performed whenever a patient had to be readmitted for any reason except ACLF (readmission visits) in the high-risk group only, but the number of readmissions was recorded in both groups. Finally, data on liver transplantation or death and causes of death were prospectively collected 3, 6 and 12 months after enrollment in all patients. ACLF, acute-on-chronic liver failure; CLIF-C AD, Chronic Liver Failure Consortium acute decompensation. EF-Clif. All authors had access to the study data and reviewed sequentially during the follow-up visits (Fig. 1). The electronic and approved the final manuscript. case-report form was designed to collect granularity in the clinical data and the detailed queries answered remaining issues Patients in case of inconsistencies. Herein, only clinical and standard laboratory data are analyzed. A total of 1,421 patients non-electively admitted for the treatment of an episode of AD were eligible, of whom 148 patients met exclusion criteria (Table S1), 202 patients presented with ACLF Data obtained at enrollment and 1,071 patients were analyzed. Among these, 218 developed Two categories of pre-specified information were obtained at AD for the first time, and the remaining 853 had a prior history of enrollment. The first category included general characteristic and AD. The diagnosis of cirrhosis was based on previous liver biopsy demographic data, specific data related to the AD episode at findings or a composite of clinical signs and findings provided by enrollment, results of physical examination and standard labo- laboratory test results, endoscopy and ultrasonography. Diag- ratory analysis, including differential white-cell blood count nostic criteria for AD upon hospitalization were based on the (WBC) and C-reactive protein (CRP) levels, as markers of development of ascites, hepatic encephalopathy, gastrointestinal systemic inflammation. Cultures were routinely performed in hemorrhage, infection, or any combination of these. Importantly, patients with suspected bacterial infections. AD was not due to an isolated bacterial infection in any of the The second category of pre-specified data was related to the enrolled patients. Diagnosis of ACLF during follow-up was per- past medical history and included: a) the timepoint of the onset formed according to the CANONIC study criteria.7 Organ failure of decompensated cirrhosis (as defined by the first episode of and organ dysfunction were defined according to the CLIF AD); b) the complications of AD occurring within the last 3 consortium (CLIF-C) organ failure (OF) score.11 months prior to enrollment; c) treatment of complications (including prior transjugular portosystemic shunt stent [TIPS] Study design and its indication); and d) any hospitalization during the last Pre-specified clinical data, standard laboratory data and biolog- 3 months prior to enrollment. Data regarding onset of decom- ical samples for biobanking were obtained at enrollment and pensated cirrhosis could be obtained in 612 patients. Data 844 Journal of Hepatology 2020 vol. 73 j 842–854 regarding the occurrence of ascites, gastrointestinal hemorrhage, square test was used for categorical variables, whereas the hepatic encephalopathy, bacterial infections and hospitalizations Student’s t test or analysis of variance were used for normally within the last 3-month period prior to enrollment were distributed continuous variables and the Wilcoxon signed-rank obtained in 860, 796, 793, 791 and 831 patients, respectively. test or the Kruskal-Wallis test were used for non-normally distributed continuous variables. In all statistical analyses, Data obtained during follow-up significance was set at p <0.05. After enrollment, patients were prospectively followed-up for a period of 3 months. The scheme of visits and collection of data and Tools to predict ACLF development samples at enrollment and during the 3-month follow-up period For the prediction of ACLF development during the 90-day after enrollment is indicated in Fig. 1. Finally, data on liver trans- follow-up period, the CLIF-C ACLF development score (CLIF-C plantation or death and causes of death were prospectively ACLF-D score) was fitted according to the TRIPOD recommen- collected 3, 6 and 12 months after enrollment in all patients. dations (see TRIPOD checklist). There were no missing data in most potential predictors of ACLF development at enrollment, Defining the 6-month observational period except for serum albumin and plasma CRP levels, whose values Of note, according to the pattern of data collection described were not available, respectively, in 5%, 9% and 8% of patients from earlier, we defined a 6-month observational period, which the pre-ACLF, UDC, and SDC groups and in 20%, 13% and 11% of included the 3-month period prior to enrollment, the enrollment patients from the 3 groups (Table S2). Therefore, for multivariate visit and the 3-month follow-up period after enrollment (Fig. 1). analysis, we assumed that these missing values could be considered at random and carried out a multiple imputation Amendment to the initial study protocol based on a mixed model including all potential predictors During the first 8 months of the study, 720 patients were significantly associated with ACLF in the univariate analysis.13 consecutively enrolled, and used for prevalence calculations. We used the proportional-hazards model for competing risks Subsequently, since the number of patients developing ACLF was proposed by Fine and Gray to identify the best subset of inde- low, we amended the study protocol to enroll only high-risk pendent predictors associated with the onset of ACLF and to patients. After IRB approval of this amendment, the last 351 develop a new predictive score (the CLIF-C ACLF-D score).14 Liver patients were enrolled in the study. transplantation and death could be considered as ‘competing’ events in the competing risks model. The initial model included Statistical analysis the most relevant characteristics at enrollment found to be Patient stratification significantly associated (both clinically and statistically) with Patient stratification was performed based on the clinical course ACLF development at 3 months in the univariate analysis during the 3-month follow-up period for several reasons: i) the main (Table S3). In the final CLIF-C ACLF-D score model, the best subset objective of the study was the characterization of the clinical course of independent predictors was selected based on a stepwise after enrollment; ii) a preliminary analysis of an incomplete set of forward procedure with p-in <0.05 and p-out <0.10 for the consecutive patients included in the PREDICT study showed that AD change in model log-likelihood (Table S4). The coefficients consisted of a single complication (either ascites, encephalopathy or estimated for each predictor were used as relative weights to gastrointestinal hemorrhage) in only 50% of patients. The remaining compute the score. patients had 2 or 3 simultaneous complications, making stratifica- Because the PREDICT study is the only thorough investigation tion based on complications at enrollment extremely complex. iii) By on the factors leading to ACLF, no other cohort could be used for contrast, stratification of patients based on ACLF development (yes external validation. As a result, we had to carry out a random or no) and clinical course profile (unstable vs. stable, among ACLF- split-sample derivation and validation processes for the new free patients) during the 3-month follow-up was simpler and score. The subset of patients used to derive the score included more appropriate for addressing the main objective of the study. two-thirds of patients (n = 707) randomly selected from each Therefore, our patients were stratified into 3 groups for data patient group. The internal score validation was performed on analysis: i) pre-ACLF group: patients who developed ACLF within the remaining third of patients (n = 364) and compared the 90 days of enrollment; ii) unstable decompensated cirrhosis predictive ability of the CLIF-C ACLF-D score with those of the (UDC) group: patients who experienced at least 1 hospital CLIF-C AD score, MELD, MELD-sodium and Child-Pugh scores by readmission, but without ACLF development within the 90-day estimating the corresponding Harrel’ C-indexes and 95% CIs both follow-up period; and iii) stable decompensated cirrhosis (SDC) in the derivation and validation sets. group: patients without ACLF development or readmissions As a complementary tool to predict ACLF development, a de- within the 90-day follow-up period. cision tree model was fitted using the 980 patients with Because bacterial infections are major precipitants of AD and information about the development of ACLF. Patients, who died or ACLF, and systemic inflammation is the hallmark of AD and ACLF, were transplanted without presenting ACLF before 3 months were infections and systemic inflammation were considered in detail excluded. The clinical variables selected for the model were the when characterizing these groups. independent predictors of ACLF development obtained in the multivariate analysis for the CLIF-C ACLF-D score. The decision Data analysis tree algorithm selected the most relevant of these clinical Discrete variables are summarized as counts (percentages) and variables, their position within the decision tree and their optimal continuous variables as mean ± SD. Non-normally distributed cut-off values. The model was fitted using R software (version variables are summarized as median (IQR) and were log- 3.6.3) rpart package with settings minsplit = 10 and maxdepth = 5. transformed for some statistical analyses and for graphical Also, the complexity parameter was set by default to 0.01. Model comparisons. In univariate statistical comparisons, the chi- parameters were estimated using the function tune.rpart from the Journal of Hepatology 2020 vol. 73 j 842–854 845 Research Article Cirrhosis and Liver Failure Table 1. Patient characteristics prior to, at, and after enrollment. Pre-ACLF (n = 218) UDC (n = 233) SDC (n = 620) p value Characteristic Age, years, mean ± SD 61.1 ± 10.0 60.9 ± 10.6 57.9 ± 11.0a <0.001 Female sex, n (%) 70 (32.1) 74 (31.8) 200 (32.3) 0.990 Etiology of cirrhosis, n (%) Alcohol 107 (49.1) 143 (61.4)b 346 (55.9) 0.032 HCV 14 (6.4) 12 (5.2) 41 (6.6) 0.727 Alcohol and HCV 10 (4.6) 8 (3.4) 33 (5.3) 0.506 Non-alcoholic steatohepatitis 16 (7.3) 17 (7.3) 48 (7.8) 0.965 Other etiologies 70 (32.1) 51 (21.9)b 150 (24.2)b 0.028 Events prior to enrollment, n (%) Ascites 130 (66.7) 122 (65.9) 229 (47.7)a <0.001 Hepatic encephalopathy 46 (25.4) 54 (31.4) 75 (17.1)a <0.001 Gastrointestinal hemorrhage 17 (9.6) 29 (17.1)b 62 (13.9) 0.125 Any hospitalization 106 (56.7) 119 (65.0) 210 (45.6)a <0.001 Data at enrollment Clinical data, organ failures and organ dysfunctions, n (%) Ascites 173 (79.4) 170 (73.0) 415 (66.9)b 0.002 Hepatic encephalopathy 65 (29.8) 73 (31.3) 168 (27.1) 0.428 Gastrointestinal hemorrhage 16 (7.3) 39 (16.7)b 97 (15.6)b 0.005 No organ failure or dysfunction 50 (22.9) 80 (36.5)b 291 (46.9)a <0.001 Liver failure 29 (13.3) 11 (4.7)b 30 (4.8)b <0.001 Liver dysfunction 51 (23.4) 36 (15.5)b 84 (13.5)b 0.003 Circulatory dysfunction 20 (9.2) 43 (18.5)b 50 (8.1)c <0.001 Renal dysfunction 51 (23.4) 17 (7.3)b 40 (6.5)b <0.001 Coagulation failure 8 (3.7) 4 (1.7) 7 (1.1)b 0.050 Coagulation dysfunction 29 (13.3) 19 (8.2) 46 (7.4)b 0.029 Brain failure 4 (1.8) 4 (1.7) 16 (2.6) 0.676 Brain dysfunction 59 (27.1) 67 (28.8) 144 (23.2) 0.197 Respiratory dysfunction 10 (4.6) 8 (3.4) 29 (4.7) 0.722 Main reason for hospitalization Ascites 105 (48.4) 106 (45.5) 267 (43.1) 0.382 Hepatic encephalopathy 29 (13.4) 34 (14.6) 82 (13.2) 0.870 Gastrointestinal hemorrhage 13 (6.0) 37 (15.9)b 110 (17.7)b <0.001 Bacterial infection 32 (14.7) 27 (11.6) 84 (13.5) 0.603 Other 38 (17.5) 29 (12.4) 77 (12.4) 0.147 Biomarkers of systemic inflammation, median (IQR) White-cell count, ×109/L 7.2 (4.9–9.8) 6.1 (4.3–8.5)b 6.0 (4.2–8.7)b 0.002 Serum C-reactive protein, mg/L 23 (11–41) 16 (8–35)b 15 (6–36)b <0.001 Measurements estimating organ function Serum bilirubin, mg/dl, median (IQR) 3.9 (1.9–9.0) 2.6 (1.3–5.4)b 2.3 (1.4–4.5)b <0.001 Serum albumin, g/dl, mean ± SD 2.7 ± 0.7 2.8 ± 0.6 3.0 ± 0.6a <0.001 INR, median (IQR) 1.6 (1.4–1.9) 1.4 (1.3–1.7)b 1.4 (1.2–1.7)b <0.001 Serum creatinine, mg/dl, median (IQR) 1.1 (0.8–1.5) 0.9 (0.7–1.2)b 0.8 (0.7–1.1)a <0.001 Plasma sodium, mEq/L, mean ± SD 134 ± 6 135 ± 5 136 ± 5a <0.001 Severity scores, mean ± SD Child-Pugh 9.8 ± 1.8 9.2 ± 1.7b 8.7 ± 1.8a <0.001 MELD 19 ± 5 16 ± 5b 15 ± 5a <0.001 MELD-sodium 23 ± 5 19 ± 5b 18 ± 5a <0.001 CLIF-C AD 57 ± 8 53 ± 8b 50 ± 8a <0.001 Data after enrollment Mortality rates, n (%) 90-day mortality rate 117 (53.7) 49 (21.0) – 1-year mortality rate 147 (67.4) 83 (35.6) 59 (9.5) Main causes of death, n (%) ACLF 130 (88.4) 25 (30.1)b 29 (49.2)a <0.001 Hypovolemic shock 4 (2.7) 14 (16.9)b 3 (5.1)c <0.001 Other causes of death 6 (4.1) 15 (18.1)b 15 (25.4)b <0.001 Unknown 7 (4.8) 29 (34.9)b 12 (20.3)b <0.001 Liver transplantation within 12 months after enrollment 33 (15.1) 39 (16.7) 73 (11.8) 0.125 (continued on next page) 846 Journal of Hepatology 2020 vol. 73 j 842–854 Table 1. (continued) Pre-ACLF (n = 218) UDC (n = 233) SDC (n = 620) p value Characteristic Data after enrollment Indicators of severe portal hypertension, n (%) TIPSd 18 (8.3) 33 (14.2)b 63 (10.2) 0.107 TIPS for gastrointestinal hemorrhage 4 (1.8) 12 (5.4) 26 (4.2) 0.145 Any episode of gastrointestinal hemorrhaged 48 (22.0) 76 (32.6)b 155 (25.0)c 0.016 p values were obtained using chi-square test. ACLF, acute-on-chronic liver failure; CLIF-C AD, Chronic Liver Failure Consortium acute decompensation; INR, international normalized ratio; MELD, model for end-stage liver disease; SDC, stable decompensated cirrhosis; TIPS, transjugular intrahepatic portosystemic shunt; UDC, unstable decompensated cirrhosis. a Significantly different from the pre-ACLF group and UDC groups. b Significantly different from the pre-ACLF group. c Significantly different from the UDC group. d At any time of the 6-month observational period, this being defined by the 3 months prior to, and the 3 months as of enrollment. R package e1071, to select the best decision tree model, according at least 1 hospital readmission within the 3-month follow-up to accuracy, sensitivity and specificity. A decision tree plot was period, had 3-month and 1-year mortality rates of 21.0% and generated based on the model fitted. A 10-Fold cross validation 35.6%, respectively; 177 of these patients required 1 readmission, was used to reduce over-fitting and to assess the discriminative 32 patients 2 readmissions, and 17 patients > −3 readmissions. ability of the model, by estimating the corresponding sensitivity Fourteen patients with UDC were transplanted after readmission and specificity of the model and computing the area under the for an AD episode within the 3-month follow-up period. Finally, receiver-operating-characteristic curve (AUC). the 620 patients included in the SDC group, who did not develop ACLF, require hospital readmission, nor die during the 3-month follow-up period after enrollment, showed very low mortality Results (9.5%) within the 1-year follow-up period after enrollment. Heterogeneity of the clinical course of AD Among the 720 patients consecutively enrolled during the first Clinical course of patients with AD 8 months after the onset of the study, 425 (59%) were in SDC As expected, the pre-ACLF group, which included 218 patients group. Twenty-eight patients with SDC were transplanted from who developed ACLF during the 3-month follow-up period after the waitlist without ACLF or a new episode of AD within the enrollment, had the highest 3-month and 1-year mortality rates 3-month follow-up period. (53.7% and 67.4%, respectively) (Table 1). Twenty-two patients The clinical course of patients with pre-ACLF was character- with pre-ACLF were transplanted after ACLF developed within ized by a huge density of bacterial infections, episodes of ACLF the 3-month follow-up period. The 233 patients included in the and death, which are summarized as events (Fig. 2). A total of 120 UDC group, who did not develop ACLF, but who died or required patients (55% of this group) developed ACLF during the first hospitalization and 98 developed the syndrome from first Pre-ACLF group (n = 218) discharge to the end of the 3-month follow-up period. The bac- terial infection density curve chronologically preceded the ACLF density curve, and both curves preceded the mortality density curve, supporting a cause to effect relationship between the 3 events. The extreme proximity between the bacterial infection and ACLF density curves reflects that ACLF is a hyperacute process UDC group (n = 233) with a very short time period between precipitating events and the onset of the syndrome. Fig. 3A shows the cumulative rate of weekly occurrence of ACLF during the first 90 days after enroll- ment of patients with pre-ACLF. Fig. 3B shows that using the 90th day after enrollment as a landmark, the cumulative incidence of death 1 year after enrollment was also higher among patients assigned to the pre-ACLF group than among those assigned to 1 of SDC group (n = 620) the other 2 groups. The density of events in the UDC group was remarkably lower than the density of events in the pre-ACLF Infections group. Although this feature was mainly due to the lack of ACLF ACLF Death episodes in the UDC group, the density of bacterial infections and deaths were also lower. Finally, although the density of bacterial infections at first presentation in the SDC group was as high as in the UDC group, it was remarkably lower during the rest of the 0 30 60 90 Days 3-month follow-up period. There were no significant differences between the 3 groups of Fig. 2. Density curves of events during the 3-month follow-up period after patients regarding the etiology of cirrhosis (Table 1), prevalence enrollment in patients with pre-ACLF, UDC and SDC. The zero timepoint corresponds to enrollment into the study. Bacterial infections are represented of active alcoholism (26.6%, 23.2% and 27.6%, respectively) or in red, ACLF in green, and deaths in blue. ACLF, acute-on-chronic liver failure; presence of hepatocellular carcinoma (within Milan criteria) at SDC, stable decompensated cirrhosis; UDC, unstable decompensated cirrhosis. enrollment (5.4%, 6.5% and 3%, respectively). Moreover, there was Journal of Hepatology 2020 vol. 73 j 842–854 847 Research Article Cirrhosis and Liver Failure 100 Prevalence and severity of bacterial infections A Cumulative percentage of ACLF 90 Table 2 provides information about infections during the 3 80 months before enrollment, at enrollment and during the 3 70 months after enrollment. Overall, 178 (22.4%) out of the 796 60 50 patients with data developed at least 1 bacterial infection during 40 the 3-month period prior to enrollment. Of the 1,071 patients 30 included in the analysis, 29.3% (n = 314) and 24% (n = 257) had 20 infections at enrollment and during the 3-month follow-up 10 period, respectively. These 571 patients with infections at 0 enrollment or during follow-up (53.3%) presented a total of 674 1 2 3 4 5 6 7 8 9 10 11 12 13 Weeks between inclusion and ACLF development infections. Considering bacterial infections, Table 2 shows that the B 1.0 Cumulative incidence function of death Gray’s test p <0.0001 distinctive features of patients with pre-ACLF relative to patients 0.8 of the 2 other groups included a higher proportion of patients with at least 1 infectious episode during the 6-month observational period (see also Fig. 4C); higher proportion of pa- 0.6 tients with sepsis at enrollment and during follow-up; higher proportion of patients with pneumonia during follow-up; and a Pre-ACLF group 0.4 higher proportion of patients receiving therapeutic antibiotics; all these differences being significant. During follow-up, the pro- 0.2 UDC group portion of patients with community-acquired infection was SDC group significantly lower among patients with pre-ACLF than among the 2 other groups (Table 2). These findings are consistent with 0.0 higher prevalence and severity of bacterial infections in the 90 120 150 180 210 240 270 300 330 360 Days pre-ACLF group. At any time, the proportion of patients with in- fections caused by multi-drug-resistant bacteria was significantly Fig. 3. Cumulative rates of ACLF and death. (A) Cumulative rate of ACLF per higher between the pre-ACLF group and the UDC group (Table 2). week during the 90-day follow-up period in patients with pre-ACLF. (B) Cu- mulative incidence of death between the 90-day (landmark) and 1 year after enrollment in patients with pre-ACLF, UDC and SDC. ACLF, acute-on-chronic Clinical features prior to enrollment liver failure; SDC, stable decompensated cirrhosis; UDC, unstable decom- By definition, patients with pre-ACLF and UDC exhibited greater pensated cirrhosis. clinical instability during the first 3 months after enrollment than patients with SDC. However, their clinical courses were also no between-group difference in the number of patients with more unstable within the 3-month period prior to enrollment, as alcohol cessation (52 [23.9%] patients for pre-ACLF, 46 [19.7%] for indicated by the significantly higher frequency of bacterial in- UDC, and 146 [23.6%] for SDC; p = 0.456) and the number of fections, ascites or hepatic encephalopathy and, consequently, those receiving HCV therapy (4 [1.9%] patients for pre-ACLF, 3 hospital admissions in these groups of patients (Tables 1 and 2). [1.3%] for UDC, and 14 [2.3%] for SDC; p = 0.650). Clinical features and laboratory data at enrollment and Duration of the decompensated phase of cirrhosis during follow-up The time-course density curves of liver transplantation or death Markers of systemic inflammation across groups in the 234 patients developing these events are shown in Fig. 4A. The WBC count and the CRP levels were significantly higher at Time zero in this figure represents the onset of decompensated enrollment in patients with pre-ACLF than in patients from the cirrhosis. Therefore, this analysis estimates the between-group other 2 groups (Table 1). In contrast, there were no significant dif- differences in the length of the entire phase of decompensated ferences in these biomarkers between patients with UDC and SDC. cirrhosis. The pre-ACLF density curve preceded the UDC density We compared the CRP levels and WBC measured at enroll- curve, and both curves preceded the SDC density curve. These ment in patients with SDC, UDC and pre-ACLF, with those findings clearly indicate that ACLF development in patients with measured at the time of follow-up diagnosis of ACLF in 176 pre-ACLF significantly reduced the duration of the decom- patients from the pre-ACLF group (including 103 patients with pensated phase of the disease. Confirming these observations, ACLF-1, 52 with ACLF-2, and 21 with ACLF-3), and those the median time from the onset of decompensated cirrhosis to measured in a control group of 34 patients with compensated death or liver transplantation was 12 months (IQR 5.2–25.8) in cirrhosis (no prior history of AD) (Fig. 4D) previously patients with pre-ACLF, 14 months (9.6–24.3) in patients with described.5,6 Of note, the last 2 groups were included to facilitate UDC (p = 0.01 vs. patients with pre-ACLF), and 20 months the comparison of systemic inflammation throughout the whole (11.4–41.3) in patients with SDC (p = 0.04 vs. patients with UDC). spectrum of cirrhosis. There was a progressive increase in the These findings are confirmed by comparing individual values of grade of systemic inflammation across the different groups. the time period between the onset of decompensated cirrhosis We also performed within-group comparisons of the levels of and liver transplantation, death or end of follow-up between the inflammatory markers measured at enrollment vs. those 3 groups (Fig. 4B). Considering the between-group differences in measured during follow-up (Table 3). The follow-up timepoint mortality, the distinct duration of the decompensation phase was the time of diagnosis of ACLF for the pre-ACLF group, while would have been even more marked if follow-up had been for the other 2 groups of patients, it was the last measurement longer than 1 year. prior to liver transplantation, or death, or the end of the 3-month 848 Journal of Hepatology 2020 vol. 73 j 842–854 p <0.001 A 12 (6; 26) Pre-ACLF B p <0.001 C 80 72.5 70 p <0.001 p = 0.01 Pre-ACLF 60 57.5 14 (10; 24) UDC Percentage 50 p = 0.04 UDC 40.5 40 20 (11; 41) SDC 30 SDC 20 10 0 10 20 30 40 50 60 Months 0 0 30 60 90 LF - A C Pre C C Months UD SD D 45 p <0.001 E 100 F 50 p <0.001 p <0.001 44.6 40 p = 0.002 45 80 40 35 35 30 31.3 CRP (mg/L) CRP (mg/L) Percentage 60 30 27.1 25 25 20 40 20 15 15 10 20 10 5 5 0 0 0 LF - Ascites HE GIB A C Pre C C LF C C LF CC UD SD UD SD AC AC e- Pr Fig. 4. Liver transplantation or death, as well as surrogates of systemic inflammation and portal hypertension in patients with pre-ACLF, UDC and SDC. (A) Density curves of liver transplantation or death during the 1-year follow-up period after enrollment in patients with pre-ACLF (in red), UDC (in blue) and SDC (in green) taking the zero-point as the onset of acute decompensation. The median time (IQR) from the onset of clinically decompensated cirrhosis (as defined by the date of first the episode of acute decompensation) to death or liver transplantation (duration of the decompensated phase of cirrhosis) was significantly shorter in patients with pre-ACLF than in those with UDC, and in patients with UDC than in those with SDC. p values were obtained using Mann-Whitney U test. (B) Individual time period between the onset of decompensated cirrhosis and liver transplantation, death or the end of the 1-year follow-up period after enrollment in the 3 groups of patients. For clarity, the figure does not include patients with values over the 75% IQR. Differences between groups were highly significant (p <0.001). p values were obtained using Kruskal-Wallis test. (C) The percentage of patients developing at least 1 bacterial infection during the 6-month obser- vational period in patients with pre-ACLF, UDC and SDC. p values were obtained using chi-square test. (D) Plasma levels of CRP (median and 75% CI) in a control group of 34 patients with compensated cirrhosis (CC, no prior history of AD), SDC, UDC, pre-ACLF and ACLF. Patients with CC were studied previously.5,6 The ACLF group includes 176 patients from the pre-ACLF group who develop ACLF during the 3-month follow-up period. Samples for PCR measurements in these patients were obtained at the time of ACLF development, p values were obtained using Kruskal-Wallis test. (E) Serum concentration of C-reactive protein in 134 patients without prior history of AD who were enrolled only for ascites, encephalopathy or gastrointestinal hemorrhage. p values were obtained using Kruskal-Wallis test. (F) Percentage of patients presenting at least one surrogate of severe portal hypertension during the 6-month observational period in the Pre-ACLF, UDC and SDC groups. P values were obtained using chi-square test. ACLF, acute-on-chronic liver failure; AD, acute decompensation; CC, compensated cirrhosis; CRP, C-reactive protein; SDC, stable decompensated cirrhosis; UDC, unstable decompensated cirrhosis. follow-up period. Within each group, there was a close remarkably higher (p <0.002) in patients with ascites (23.4 relationship between changes in inflammatory markers and the [12.5–38.0]) than in those with encephalopathy (11.0 [4.4–21.6]) clinical course (Table 3). Progression of AD to ACLF in the pre- and gastrointestinal hemorrhage (5.0 [3.0–22.4]) (Fig. 4E). ACLF group occurred in the setting of a significant increase in WBC count and serum concentration of CRP. In patients with UDC there were no significant changes in WBC count and a small, but significant decrease in CRP, suggesting minor improvement Organ function and scores of systemic inflammation. Finally, patients with SDC had a The prevalence of liver failure, liver dysfunction and renal significant reduction in WBC and PCR. dysfunction (as defined by the CLIF-C OF score11) at enrollment was significantly higher among patients with pre-ACLF group than among those with UDC and SDC (Table 1). Moreover, lab- Association between systemic inflammation and complications that oratory measurements estimating liver and renal function at define AD enrollment were significantly more impaired among patients In order to assess the association between systemic inflammation with pre-ACLF than among those with UDC and SDC, suggesting and the 3 major complications that define AD, we explored 134 that a significant deterioration of organ function existed prior to patients who had no prior history of AD and were enrolled only enrollment in patients with pre-ACLF. for ascites (n = 99), encephalopathy (n = 14) or gastrointestinal CLIF-C AD and MELD-sodium scores significantly worsened hemorrhage (n = 21). The median (IQR) levels of plasma CRP was during the progression of pre-ACLF to ACLF and improved in Journal of Hepatology 2020 vol. 73 j 842–854 849 Research Article Cirrhosis and Liver Failure Table 2. Characteristics of infections at enrollment and during the 90-day follow-up period. Characteristic Pre-ACLF UDC SDC p value (n = 218) (n = 233) (n = 620) Number of patients with infections n (%)* 3 months prior to enrollment 58 (31.0) 45 (26.5) 75 (17.1)a <0.001 At enrollment 74 (33.9) 61 (26.2) 178 (28.7) 0.176 3 months after enrollment 106 (48.6) 83 (35.6)b 68 (11.0)a <0.001 Throughout the 6-month observational period 158 (72.5) 133 (57.1)b 251 (40.5)a <0.001 Infections at enrollment Number of infections 83 67 189 Site of infection, n/N (%)* Urinary tract 19/83 (22.9) 15/67 (22.4) 44/189 (23.2) 0.985 Spontaneous bacterial peritonitis 18/83 (21.7) 13/67 (19.4) 26/189 (13.8) 0.232 Pneumonia 10/83 (12.0) 14/67 (20.9) 24/189 (12.8) 0.213 Spontaneous bacteremia 9/83 (10.8) 5/67 (7.5) 9/189 (4.8) 0.184 Cellulitis 4/83 (4.8) 6/67 (9.0) 18/189 (9.6) 0.414 Suspected infections 6/83 (7.2) 8/67 (11.9) 35/189 (18.6)b 0.040 Otherc 17/83 (20.5) 6/67 (9.0) 32/189 (17.0) 0.150 Severity of infection, n/N (%)* Community-acquired 52/83 (62.6) 35/67 (52.2) 149/189 (78.8)a <0.001 Health-care- or hospital-acquired 31/83 (37.4) 32/67 (47.8) 40/189 (21.2)a <0.001 Sepsis 26/83 (31.3) 11/67 (16.4)b 28/189 (15.1)b 0.005 Infection caused by MDR bacteria 6/83 (7.2) 3/67 (4.9) 18/189 (10.3) 0.379 Infections during the 3-month follow-up period Number of infections 140 117 76 Site of infection, n/N (%)* Urinary tract 35/140 (25.0) 31/117 (26.5) 22/76 (28.9) 0.821 Spontaneous bacterial peritonitis 21/140 (15.0) 24/117 (20.5) 5/76 (6.6)c 0.030 Pneumonia 27/140 (19.3) 10/117 (8.5)b 10/76 (13.2) 0.047 Spontaneous bacteremia 10/140 (7.1) 9/117 (7.7) 2/76 (2.6) 0.319 Cellulitis 6/140 (4.3) 8/117 (6.8) 4/76 (5.3) 0.665 Suspected infections 16/140 (11.4) 13/117 (11.1) 16/76 (21.1) 0.091 Otherc 25/140 (17.9) 22/117 (18.8) 17/76 (22.4) 0.717 Severity of infection, n/N (%)* Community-acquired 14/140 (10.0) 15/117 (12.8)b 15/76 (19.7)b 0.129 Health-care- or hospital-acquired 126/140 (90.0) 102/117 (87.2) 61/76 (80.3)b 0.129 Sepsis 70/140 (50.0) 21/117 (18.1)b 4/76 (5.3)a <0.001 Infection caused by MDR bacteria 44/140 (33.8) 29/117 (28.2) 11/76 (16.2)b 0.031 p values were obtained using chi-square, * is calculated over the available data, no imputation was included in the table. ACLF, acute-on-chronic liver failure; MDR, multidrug resistant; SDC, stable decompensated cirrhosis; UDC, unstable decompensated cirrhosis. a Significantly different from the pre-ACLF group and UDC groups. b Significantly different from the pre-ACLF group. c Other: catheter-related infection, cholecystitis, cholangitis, secondary peritonitis, pseudomembranous colitis, other gastrointestinal infection. patients with SDC (Table 3). Scores also improved in patients in patients with UDC (44.6%) than in patients with pre-ACLF with UDC, although to a lesser extent than in patients with SDC. (27.1%) and SDC (31.3%). Increased prevalence of features suggesting severe portal Tools to predict development of ACLF hypertension in patients with UDC The CLIF-C ACLF-D score was developed to predict, at the time Whereas severe systemic inflammation and organ failure or of hospital admission, the probability of a patient with AD dysfunction were the most prominent features in patients from developing ACLF during the following 3 months. The initial the pre-ACLF group, surrogates of severe portal hypertension model was fitted including all the main characteristics at were the hallmark of patients with UDC. First, the prevalence of enrollment found to be associated with the development of circulatory dysfunction at enrollment (Table 1) and of gastroin- ACLF in the univariate analysis (Table S3). Patients age (years), testinal hemorrhage within the 6-month observational period presence of ascites, WBC count (×109/L), serum albumin (g/dl), (32% vs. 22% [p = 0.01] and 25% [p = 0.03], respectively) were serum bilirubin (mg/dl), and serum creatinine (mg/dl) at study significantly higher among patients with UDC than among those enrollment were subsequently identified as the best subset of with pre-ACLF and those with SDC. Second, the percentage of independent predictors in the final model (Table S4) and their patients who received TIPS during this period was also higher coefficients were used as relative weight to compute the cor- in the UDC group than in the other 2 groups (14.2% vs. 8.3% responding score. The equation for CLIF-C ACLF-D score is as [p = 0.04] and 10.2% [p = 0.1], respectively). Finally, the preva- follows: lence of hypovolemic shock as the main cause of death was CLIF-C ACLF-D score = ((0.03 × Age) + (0.45 × Ascites) + (0.26 × 6- and 3-times higher in patients with UDC group (16.9%) than in ln(WBC)) − (0.37 × Albumin) + (0.57 × ln(Bilirubin)) + (1.72 × those with pre-ACLF (2.7%; p <0.001) and SDC (5.1%; p <0.001). ln(Creatinine)) + 3 × 10. Fig. 4F shows that the percentage of patients with at least 1 The prognostic accuracy of CLIF-C ACLF-D score (Fig. 5A) was surrogate of severe portal hypertension was significantly higher higher than those of CLIF-C AD, MELD, MELD-sodium and Child- 850 Journal of Hepatology 2020 vol. 73 j 842–854 Table 3. Inflammatory markers and severity scores at enrollment and during the 90-day follow-up period. Enrollment Follow-up p value Pre-ACLF (n = 218) Blood biomarkers of systemic inflammation, median (IQR) White-cell count, ×109/L 7.2 (4.9–9.8) 8.3 (5.7–12.9) <0.001 Serum C-reactive protein, mg/L 23 (11–41) 29 (14–52) 0.033 Severity scores, mean ± SD MELD-sodium 23 ± 5 28 ± 6 <0.001 CLIF-C AD 57 ± 7 64 ± 9 <0.001 Unstable decompensated cirrhosis (n = 233) Blood biomarkers of systemic inflammation, median (IQR) White-cell count, ×109/L 6.1 (4.3–8.5) 5.9 (4.0–8.0) 0.343 Serum C-reactive protein, mg/L 16 (8–35) 12 (5–26) 0.004 Severity scores, mean ± SD MELD-sodium 19 ± 5 18 ± 6 0.006 CLIF-C AD 53 ± 7 51 ± 8 0.031 Stable decompensated cirrhosis (n = 620) Blood biomarkers of systemic inflammation, median (IQR) White-cell count, ×109/L 6.0 (4.2–8.7) 5.4 (3.9–7.3) <0.001 Serum C-reactive protein, mg/L 15 (6–36) 8 (4–17) <0.001 Severity scores, mean ± SD MELD-sodium 18 ± 5 16 ± 5 <0.001 CLIF-C AD 50 ± 8 48 ± 7 <0.001 p values were obtained using the Wilcoxon signed-rank test or the Student’s t test where appropriate. ACLF, acute-on-chronic liver failure; CLIF-C AD, Chronic Liver Failure Consortium acute decompensation; MELD, model for end-stage liver disease. A Severity scores Derivation set (n = 707) Validation set (n = 364) Harrel’ C-index (95% confidence interval) CLIF-C ACLF-D 0.76 (0.72-0.80) 0.77 (0.72-0.82) CLIF-C AD 0.70 (0.66-0.74) 0.75 (0.70-0.80) MELD-sodium 0.70 (0.66-0.74) 0.74 (0.69-0.80) MELD 0.70 (0.66-0.74) 0.73 (0.67-0.79) Child-Pugh 0.64 (0.59-0.68) 0.67 (0.60-0.73) Probability of ACLF B 0.22 development within 90 day 100% Yes No 0 0.5 1.0 Creatinine <1.3 creat < 1.2 0.16 0.43 79.5% 20.5% Bilirubin <3 creat < 1.2 Bilirubin creat < <5.5 1.2 0.27 0.38 37.8% 17.0% Bilirubin creat <<16.6 1.2 Creatinin <1.5 creat < 1.2 0.24 0.63 0.26 0.51 35.0% 2.8% 9.1% 8.0% Albumin ≥1.5 creat < 1.2 Age <41.5 creat < 1.2 Bilirubin <2.7 creat < 1.2 WBC creat <<3.0 1.2 0.48 0.57 2.3% 7.0% Creatinin ≥1.4 creat < 1.2 Creatinin ≥1.9 creat < 1.2 0.07 0.23 1.00 0.00 0.71 0.18 0.13 0.67 0.11 0.14 0.61 0.71 41.7% 34.6% 0.4% 0.3% 2.4% 6.7% 0.8% 1.5% 0.9% 0.7% 6.3% 3.5% Fig. 5. Predictive ability of the CLIF-C ACLF-D score. (A) Comparison between the predictive ability of the CLIF-C ACLF-D score with those of the CLIF-C AD, MELD, MELD-sodium and Child-Pugh scores using Harrel’ C-indexes and 95% CIs both in the derivation and validation sets. (B) Decision tree plot for the prediction of ACLF development during the 90-day follow-up period after enrollment. Each node shows the percentage of patients classified and their probability of ACLF development within the 90-day follow-up period after enrollment (also represented by the colors and color intensity). The blue color represents a probability of ACLF development >0.5. The green color represents a probability of ACLF development ACLF <0.5. The intensity of the color represents the estimated probability value. The upper node (root node) represents the entire population of patients (980 patients, 100%) included in the analysis and its corresponding probability of ACLF development before entering the model (0.22). Each node includes the estimated probability of subsequent subsets of patients. ACLF, acute-on-chronic liver failure; CLIF-C ACLF-D, Chronic Liver Failure Consortium acute-on-chronic liver failure development; CLIF-C AD, Chronic Liver Failure Consortium acute decompensation; MELD, model for end-stage liver disease. Journal of Hepatology 2020 vol. 73 j 842–854 851 Research Article Cirrhosis and Liver Failure Pugh scores in the derivation set. In the validation set, the CLIF-C The median time between the onset of decompensated ACLF-D showed a similar accuracy but smaller differences with cirrhosis to liver transplantation or death, which covers regards to the other scores. Therefore, we were unable to design the complete phase of clinically decompensated cirrhosis, a new score to predict ACLF development more accurately than was remarkably shorter in patients in the pre-ACLF group the traditional clinical scores. (12 months) than in those with SDC (20 months), indicating an The most relevant clinical variable selected by the decision accelerated clinical course of the decompensated phase of the tree model was creatinine, with a threshold of 1.3 mg/dl (Fig. 5B). disease towards death in patients with pre-ACLF. Bilirubin, albumin, age and WBC were also selected to Finally, the clinical course during the first 3-month period subsequently discriminate patients. The terminal nodes with a prior to admission, as estimated by the prevalence of ascites, probability of ACLF higher than 0.5, classifying patients likely to encephalopathy and bacterial infections, was significantly more develop ACLF, included 14.1% of the patients. The model achieved unstable in the pre-ACLF group than in the SDC group. This a discriminating ability (AUC) of 0.76 (0.72–0.79), with high finding suggests that patients with pre-ACLF were already more specificity (95%) but low sensitivity (38%), indicating an severely ill than patients with SDC months before reaching the important misclassification among those patients who actually pre-ACLF status. We presume that the intensity of systemic developed ACLF. inflammation during this period was probably sufficient to induce this frequent development of complications requiring hospital admission, but not enough to reach the critical threshold Discussion beyond which ACLF develops.15 Therefore, pre-ACLF should be The most noteworthy finding of the current study was the suspected in patients hospitalized for AD with prior unstable identification of 3 different clinical courses with distinct patho- clinical course, very high levels of inflammatory markers and physiology and prognosis in patients hospitalized for the treat- liver failure or liver or kidney dysfunction. Unfortunately, we ment of an episode of AD. These 3 clinical courses were unrelated were unable to design new specific tools that improve the ac- to the etiology of cirrhosis, or to active alcoholism in patients curacy of the CLIF-C AD and MELD-sodium scores for predicting with alcohol-related cirrhosis, indicating that they were largely ACLF development. dependent on other mechanisms. Patients with UDC shared many characteristics with patients The 3 distinct types of clinical courses coincided with specific with pre-ACLF and SDC. Like patients with pre-ACLF, they pre- changes in the grade of systemic inflammation. Patients with sented clinical course instability within the 3-month period prior pre-ACLF showed significantly higher grade of systemic to and after enrollment. However, they did not present severe inflammation at enrollment than patients with UDC and SDC. By systemic inflammation at enrollment or a clear increase of contrast, there was no significant difference in systemic systemic inflammation level during follow-up. This probably inflammation between patients with UDC and SDC. Moreover, explains the lack of development of ACLF in this group of whereas the levels of inflammatory markers increased signifi- patients. A second important finding in patients with UDC was cantly during follow-up, accompanying the progression of AD to their significantly higher prevalence of features suggestive of ACLF in patients with pre-ACLF, they decreased intensely in severe portal hypertension. This finding supports that the second patients with SDC, while they did not show clear changes in major pathophysiological mechanism of AD is likely related to patients with UDC. Therefore, a distinct progression of systemic changes in portal hypertension. inflammation is likely a major pathogenetic mechanism un- Therefore, the most severe course of AD corresponds to pa- derlying the 3 clinical courses of patients with AD. This finding tients with pre-ACLF who develop rapid progression of systemic is a key feature in the new comprehensive hypothesis for AD inflammation leading to ACLF development and death. The sec- presented in the current article. ond course corresponds to patients with UDC, who have an Thus, patients with SDC developed the index episode of AD in increased incidence of complications related to severe portal the context of moderate systemic inflammation. In addition, hypertension, such as circulatory dysfunction at enrollment, systemic inflammation decreased rapidly and remained at low increased incidence of gastrointestinal hemorrhage and TIPS intensity during the 3-month follow-up. Probably due to this, all placement during the 6-month observational period and higher patients recovered from the index episode of AD, most presented mortality due to hypovolemic shock. However, since the grade of a long-term relatively benign clinical course and only 9.5% died systemic inflammation did not progress to the critical threshold within the 1-year follow-up. Around half of the few patients who level to induce extrahepatic organ failure, only a minority of died within the 1-year follow-up period reproduced the clinical patients with UDC developed ACLF. Consequently, they lived course of the pre-ACLF group and developed multiorgan failure. longer than patients with pre-ACLF. Finally, the third course of In contrast, hypovolemic shock was reported as the main cause AD, which is by far the most frequent, corresponds to patients of death in only 5% of cases. with SDC and is likely the consequence of a slow progression of In contrast, patients with pre-ACLF developed AD in the these 2 pathophysiological mechanisms, leading to a relatively context of more intense systemic inflammation, which further benign course and much longer survival. increased with ACLF development during follow-up. These This hypothesis is further supported by our findings showing patients differed significantly from patients with SDC in many that ascites, which is the complication associated with the most other features reported at enrollment, clearly supporting that extensive organ dysfunction (liver, kidney, heart and systemic they were in a pre-ACLF stage. They exhibited a significantly circulation),16,17 was associated with the most intense systemic higher prevalence of liver failure, liver dysfunction, renal inflammation in comparison with hepatic encephalopathy and dysfunction, ascites, encephalopathy and bacterial infections gastrointestinal hemorrhage. and significantly worse prognostic scores than patients with SDC Bacterial infections were frequently associated with AD. and UDC. Roughly, 1 in every 3-4 patients included in each group were 852 Journal of Hepatology 2020 vol. 73 j 842–854 infected at the time of AD development. Two mechanisms have data, critical revision of the manuscript regarding important been proposed for this association. The first is that bacterial in- intellectual content. fections, by increasing the intensity of systemic inflammation, precipitate the development of AD.2,5,6 The second is that bac- terial infections would be the consequence of a compensatory Acknowledgements The authors are very grateful to the patients, their families and immunomodulatory reaction to systemic inflammation, which the personal of the hospitals for making this possible. In addition impairs the antibacterial activity of immune cells (immunopar- a special thank you is dedicated to Mrs. Yolanda Godoy, Dr. Anna alysis).18–20 Our findings suggest that these 2 mechanisms are Bosch, Dr. Josep-Maria Torner, Mrs. Cecilia Ducco, Montserrat not mutually exclusive. Carreras, Marites Abans and Paul Sauerbruch for excellent In summary, the PREDICT study suggests that AD in cirrhosis assistance in the accomplishment of the study. is a clinical condition with 3 different courses and 2 major pathophysiological mechanisms. Pre-ACLF is predominantly † related to rapid progression of systemic inflammation, ACLF EASL-CLIF consortium collaborators development and an extremely high short-term mortality rate. Miriam Maschmeier1, David Semela2, Laure Elkrief3, Ahmed UDC occurs in the context of rapid progression of portal hyper- Elsharkawy4, Tamas Tornai5, Istvan Tornai5, Istvan Altorjay5, tension and is associated with a less severe clinical course and Agnese Antognoli6, Maurizio Baldassarre6, Martina Gagliardi6, lower short-term mortality. Finally, both mechanisms progress Eleonora Bertoli7, Sara Mareso7, Alessandra Brocca7, Daniela slowly in SDC, and patients follow a relatively benign course with Campion8, Giorgio Maria Saracco8, Martina Rizzo8, Jennifer longer survival. Predicting the outcome of patients who present Lehmann9, Alessandra Pohlmann9, Michael Praktiknjo9, Robert with AD is a major future research challenge. Schierwagen9,18, Elsa Solà10, Nesrine Amari11, Miguel Rodri- guez12, Frederik Nevens13, Ana Clemente14, Peter Jarcuska15, Abbreviations Alexander Gerbes16, Mattias Mandorfer17, Christoph Welsch18, ACLF, acute-on-chronic liver failure; AD, acute decompensation; Emanuela Ciraci19, Vish Patel20, Cristina Ripoll21, Adam Herber22, AUC, area under the receiver-operating characteristic curve; Paul Horn23, Karen Vagner Danielsen24, Lise Lotte Gluud24, Jelte CLIF-C ACLF-D, Chronic Liver Failure Consortium acute-on- Schaapman25, Oliviero Riggio26, Florian Rainer27, Jörg Tobiasch chronic liver failure development; CLIF-C AD, Chronic Liver Moritz28, Mónica Mesquita29, Edilmar Alvarado-Tapias30, Osagie Failure Consortium acute decompensation; CLIF-C OF, Chronic Akpata31, Peter Lykke Eriksen32, Didier Samuel33, Sylvie Liver Failure Consortium organ failure; CRP, C-reactive protein; Tresson33, Pavel Strnad34, Roland Amathieu35, Macarena Simón- HR, hazard ratio; MELD, model for end-stage liver disease; SDC, Talero36, Francois Smits11, Natalie van den Ende13, Javier Marti- stable decompensated cirrhosis; TIPS, transjugular intrahepatic nez12, Rita Garcia14, Daniel Markwardt16, Harald Rupprechter17, portosystemic shunt; UDC, unstable decompensated cirrhosis; Cornelius Engelmann22 WBC, white blood cell count. Collaborator affiliations 1 Financial support Munster University Hospital, Münster, Germany; 2University of The study was supported by the European Foundation for the Basel-St Gall Cantonal Hospital, Switzerland; 3Hôpitaux Uni- Study of Chronic Liver Failure (EF-Clif). The EF-Clif is a non- versitaires de Genève, Genève, Switzerland; 4University of Bir- profit private organization. The EF-Clif receives unrestricted mingham, Birmingham, UK; 5University of Debrecen, Faculty of donations from Cellex Foundation and Grifols. EF-Clif is partner, Medicine, Institute of Medicine, Department of Gastroenterology, contributor and coordinator in several EU Horizon 2020 pro- Debrecen, Hungary; 6University of Bologna, Bologna, Italy; 7 gram projects. JT was appointed as visiting Professor in EF-Clif University of Padova, Padova, Italy; 8A.O.U. Città della Salute e for the execution of the study by a grant from Cellex Founda- della Scienza Torino, Torino, Italy; 9University Hospital Bonn, tion. The funders had no influence on study design, data Bonn, Germany; 10Hospital Clinic of Barcelona, Barcelona, Spain; 11 collection and analysis, decision to publish or preparation of C.U.B. Erasme, Bruxelles, Belgium; 12Department of Gastroen- the manuscript. terology, Hospital Universitario Ramón y Cajal, IRYCIS, University of Alcalá, CIBEREHD, Madrid, Spain; 13University of Leuven, Conflict of interest Leuven, Belgium; 14Hospital General Universitario Gregorio None of the authors have conflicts of interest for the reported Marañón. Facultad de Medicina (Universidad Complutense of study. Madrid), CIBERehd, Madrid, Spain; 15Pavol Jozef Safarik Univer- Please refer to the accompanying ICMJE disclosure forms for sity in Kosice, Kosice, Slovakia; 16Munich University Hospital, further details. Munich, Germany; 17Medical University of Vienna, Vienna, Austria; 18Derriford Hospital, Plymouth Hospitals Trust, Ply- Authors’ contributions mouth, UK; 19Internal Medicine PO Ostuni, ASL Brindisi, Italy; 20 JT, JF, WL, JC, RJ, RM, PG, PA, VA: study concept and design, King’s College Hospital, London, United Kingdom; 21University acquisition of data, analysis and interpretation of data, drafting of Hospital Halle-Wittenberg, Halle(Saale), Germany; 22University the manuscript, funding recipient, administrative, technical and Hospital Leipzig, Leipzig, Germany; 23Jena University Hospital, material support, study supervision; EG, AA, AC, CP, MP, CS, AC, Jena, Germany; 24Gastrounit, Medical Section, Hvidovre Hospital AM, FA: acquisition of data, analysis of data, technical and and Department of Clinical Medicine, University of Copenhagen, material support; TT, MB, PA, CA, FEU, CJ, MST, TG, AA, WL, ES, RB, Copenhagen, Denmark; 25Leiden University Medical Center, Lei- MJ, CS, TR, JA, PG, WB, SZ, CR, TB, AS, LLG, MC, OR, RS, HZ, AC, GSP, den, Netherlands; 26Universitá Sapienza Roma, Roma, Italy; 27 AdG, HG, FS, CT, OCÖ, FS, SR, RA, MRG, HVV, CF, MM, MP, PC, SP, Medical University of Graz, Graz, Austria; 28Medical University IG, MP, VV, RM, ZV, MB, EB: acquisition of data, interpretation of of Innsbruck, Innsbruck, Austria; 29CHTMAD Vila Real- Journal of Hepatology 2020 vol. 73 j 842–854 853 Research Article Cirrhosis and Liver Failure Blueclinical, Vila Real, Portugal; 30Hospital de la Santa Creu i Sant [7] Moreau R, Jalan R, Gines P, Pavesi M, Angeli P, Cordoba J, et al. Acute-on- Pau and CIBERehd, Barcelona, Spain; 31UCL Medical School, Royal chronic liver failure is a distinct syndrome that develops in patients with acute decompensation of cirrhosis. Gastroenterology 2013;144:1426–1437. Free Hospital, London, United Kingdom; 32Aarhus University [8] Moreau R, Claria J, Aguilar F, Fenaille F, Lozano JJ, Junot C, et al. Blood Hospital, Aarhus, Denmark; 33AP-HP Hôpital Paul Brousse, Centre metabolomics uncovers inflammation-associated mitochondrial Hépato-Biliaire, Universite Paris Saclay, INSERM Unit 1193, Ville- dysfunction as a potential mechanism underlying ACLF. J Hepatol 2020 juif, France; 34Aachen University Hospital, Aachen, Germany; Apr;72(4):688–701. 35 [9] Wiest R, Garcia-Tsao G. Bacterial translocation (BT) in cirrhosis. Hep- AP-HP, Hôpital Jean Verdier, Service d’Hépatologie, Bondy; atology 2005;41:422–433. Université Paris 13, Sorbonne Paris Cité, “Equipe labellisée Ligue [10] Fernandez J, Claria J, Amoros A, Aguilar F, Castro M, Casulleras M, et al. Contre le Cancer”, Saint-Denis; Inserm, UMR-1162, “Génomique Effects of albumin treatment on systemic and portal hemodynamics and fonctionnelle des tumeurs solides”, Paris, France; 36Hospital systemic inflammation in patients with decompensated cirrhosis. Universitari Vall dHebron, Barcelona, Spain. Gastroenterology 2019;157:149–162. [11] Jalan R, Saliba F, Pavesi M, Amoros A, Moreau R, Gines P, et al. Develop- ment and validation of a prognostic score to predict mortality in patients with acute-on-chronic liver failure. J Hepatol 2014;61:1038–1047. [12] Jalan R, Pavesi M, Saliba F, Amoros A, Fernandez J, Holland-Fischer P, et al. Supplementary data The CLIF Consortium Acute Decompensation score (CLIF-C ADs) for Supplementary data to this article can be found online at https:// prognosis of hospitalised cirrhotic patients without acute-on-chronic doi.org/10.1016/j.jhep.2020.06.013. liver failure. J Hepatol 2015;62:831–840. [13] Smith C, Kosten S. Multiple Imputation: a Statistical Programming Story. PharmaSUG; 2017. Paper SP01. References [14] Wolbers M, Koller MT, Witteman JC, Steyerberg EW. Prognostic models Author names in bold designate shared co-first authorship with competing risks: methods and application to coronary risk predic- tion. Epidemiology 2009;20:555–561. [1] Arroyo V, Moreau R, Kamath PS, Jalan R, Gines P, Nevens F, et al. Acute-on- [15] Arroyo V, Moreau R, Jalan R. Acute-on-chronic liver failure. N Engl J Med chronic liver failure in cirrhosis. Nat Rev Dis Primers 2016;2:16041. 2020;382:2137–2145. [2] Fernandez J, Acevedo J, Wiest R, Gustot T, Amoros A, Deulofeu C, et al. Bac- [16] Schrier RW, Arroyo V, Bernardi M, Epstein M, Henriksen JH, Rodes J. terial and fungal infections in acute-on-chronic liver failure: prevalence, Peripheral arterial vasodilation hypothesis: a proposal for the initiation of characteristics and impact on prognosis. Gut 2018 Oct;67(10):1870–1880. renal sodium and water retention in cirrhosis. Hepatology 1988;8: [3] Arvaniti V, D’Amico G, Fede G, Manousou P, Tsochatzis E, Pleguezuelo M, 1151–1157. et al. Infections in patients with cirrhosis increase mortality four-fold and [17] Ruiz-del-Arbol L, Urman J, Fernandez J, Gonzalez M, Navasa M, should be used in determining prognosis. Gastroenterology Monescillo A, et al. Systemic, renal, and hepatic hemodynamic derange- 2010;139:1246–1256. 1256.e1–5. ment in cirrhotic patients with spontaneous bacterial peritonitis. [4] Gines P, Quintero E, Arroyo V, Teres J, Bruguera M, Rimola A, et al. Hepatology 2003;38:1210–1218. Compensated cirrhosis: natural history and prognostic factors. Hepatol- [18] Delano MJ, Ward PA. The immune system’s role in sepsis progres- ogy 1987;7:122–128. sion, resolution, and long-term outcome. Immunol Rev [5] Claria J, Stauber RE, Coenraad MJ, Moreau R, Jalan R, Pavesi M, et al. 2016;274:330–353. Systemic inflammation in decompensated cirrhosis: characterization and [19] Malik R, Mookerjee RP, Jalan R. Infection and inflammation in liver failure: role in acute-on-chronic liver failure. Hepatology 2016;64:1249–1264. two sides of the same coin. J Hepatol 2009;51:426–429. [6] Trebicka J, Amoros A, Pitarch C, Titos E, Alcaraz-Quiles J, Schierwagen R, [20] Wasmuth HE, Kunz D, Yagmur E, Timmer-Stranghoner A, et al. Addressing profiles of systemic inflammation across the different Vidacek D, Siewert E, et al. Patients with acute on chronic liver clinical phenotypes of acutely decompensated cirrhosis. Front Immunol failure display “sepsis-like” immune paralysis. J Hepatol 2019;10:476. 2005;42:195–201. 854 Journal of Hepatology 2020 vol. 73 j 842–854