Prognosis algorithms for acute decompensation of cirrhosis and ACLF

Abstract

Accurate prediction of survival in patients with cirrhosis is crucial, as patients who are unlikely to survive in the short‐term need to be oriented to liver transplantation and to novel therapeutic approaches. Patients with acute decompensation of cirrhosis without or with organ dysfunction/failure, the so‐called acute‐on‐chronic liver failure (ACLF), have a particularly high short‐term mortality. Recognizing the specificity of this clinical situation, dedicated classifications and scores have been developed over the last 15 years, including variables (e.g. organ failures and systemic inflammation) not part of the formerly available cirrhosis severity scores, namely Child‐Pugh score or MELD. For patients with acute decompensation of cirrhosis, it led to the development of a dedicated score, the Clif‐C‐AD score, independently validated. For more severe patients, three different scoring systems have been proposed, by European, Asian and North American societies namely Clif‐C‐ACLF, AARC score and NASCELD‐ACLF respectively. These scores have been validated, and are widely used across the world. The differences and similarities between these scores, as well as their validation and limitations are discussed here. Even if these scores and classifications have been a step forward in favouring homogeneity between studies, and in helping making decisions for individual patients, their predictive value for mortality can still be improved as their area under the ROC curve does not exceed .8. Novel scores including biomarkers reflecting the pathophysiology of acute decompensation of cirrhosis might help reach that goal.

Abbreviations

AARC

APASL‐ACLF Research Consortium

ACLF

acute‐on‐chronic liver failure

AD

acute decompensation of cirrhosis

AUROC

area under the receiver operating characteristic curve

CLIF‐C ACLFs

CLIF Consortium ACLF score

CLIF‐C AD

chronic liver failure Consortium acute decompensation

CLIF‐C OFs

CLIF Consortium organ failure score

COSSH

Chinese group on the study of severe hepatitis B

GWASs

Genome‐wide association studies

ICU

intensive care unit

MELD

model for end‐stage liver disease

NACSELD

North American Consortium for the study of end‐stage liver disease

Key points.

• CLIF‐C AD score is a score developed and validated for patients with acute decompensation of cirrhosis without ACLF.

• For patients with ACLF, three different scoring system are widely used, namely the CLIF‐C ACLF score, the AARC score and the NACSELD score.

• Predictive value of currently available scores can still be improved.

• Multiomics analyses may help develop novel tools to improve prediction of patients outcome.

1. INTRODUCTION

The natural history of cirrhosis is classically divided into two distinct clinical stages: a long asymptomatic phase, called compensated cirrhosis, characterized by preserved quality of life and a median survival exceeding 12 years, and the decompensated stage, marked by the occurrence of complications, including ascites, gastrointestinal bleeding, hepatic encephalopathy ¹ or jaundice or bacterial infection, ² with median survival dropping to 2–4 years. ² More recently, decompensation of cirrhosis has been further divided into acute decompensation and non‐acute decompensation, as discussed elsewhere in this issue of Liver International. Acute decompensation (AD) is defined as the recent development of ascites, encephalopathy, gastrointestinal haemorrhage or any combination of these disorders in patients with cirrhosis. ³ , ⁴

Accurate prediction of survival in patients with cirrhosis is of paramount importance. Indeed, liver transplantation can significantly enhance survival in patients who are unlikely to survive in the short term with medical management alone. ⁵ Moreover, patients with a poor predicted outcome are the population of choice for inclusion in clinical trials testing novel therapeutic approaches.

In patients with severe cirrhosis in a stable conditions (no recent acute events in the last weeks), the model for end‐stage liver disease (MELD) score has been instrumental for prioritization of liver transplant candidates. ⁶ As its predictive power has diminished in recent years, ⁷ refinements such as MELD‐Na or MELD 3.0 have been proposed. ⁸ In patients with AD, short‐term mortality is high, ⁴ , ⁹ , ¹⁰ and MELD or MELD‐Na ¹¹ have a limited prognostic accuracy. There has thus been a growing interest over the last 15 years in precisely characterizing this subgroup of patients and improving the ability to predict their mortality. This interest has led to the development of scores dedicated to AD and to the term acute‐on‐chronic liver failure (ACLF). ¹² , ¹³

In this review, we will analyse prognostic scores used across the world to prognose mortality in patients with acute decompensation of cirrhosis (AD) without or with ACLF.

2. PROGNOSTIC SCORE OF AD WITHOUT ACLF

2.1. Initial description and performance

The EASL‐Clif Consortium utilized the CANONIC study, a prospective investigator driven study, to develop the chronic liver failure Consortium acute decompensation (Clif‐C AD) score. This score is based on objective and readily available clinical parameters, aiming at accurately predicting the risk of mortality in patients with AD of cirrhosis without ACLF. ¹⁴ Indeed, some patients with AD without ACLF at admission are at high risk of short‐term mortality and require closer monitoring and interventions. The CANONIC study population consisted of 1349 patients admitted electively for AD of cirrhosis across 29 centres in Europe, with 1016 not having ACLF at inclusion. The majority (80%) of patients had alcohol or HCV‐related cirrhosis. Age, serum sodium, log‐transformed white cell count, creatinine and INR were identified as the best predictors of mortality. The C‐Index of Clif‐C AD score for 90 days, 6 months and 1 year mortality (.74, .71 and .67) was significantly superior to that obtained with Child‐Pugh score, MELD score, and MELD‐Na score (Table 1). Clif‐C AD score below 45 identified a group of patients with a mortality rate five times lower than the entire series of patients. Conversely, Clif‐C AD score above 60 identified a group of patients with a high mortality rate, similar to patients with ACLF grade 1 reported in the CANONIC study by Moreau et al. ⁹

TABLE 1.

Scores predicting outcome of patients with acute decompensation of cirrhosis.

Reference	N	Causes of cirrhosis	Mortality (90 days)	C‐Index or AUROC of Clif C AD score for 90‐day mortality	Comparison of Clif‐C AD score with other scores	p Value (C‐index vs. other score)
Jalan et al., J Hepatol. 2015 14	1016	Alcohol (49%) HCV (22%) Alcohol+HCV (10%)	13%	.74	Child‐Pugh: .65 MELD score: .65 MELD‐Na: .68	p < .001 p < .001 p < .001
Jalan et al., J Hepatol. 2015 14	225 (external validation)	Alcohol (49%) HCV (18%) Alcohol+HCV (15%)	10%	.74	Child‐Pugh: .63 MELD score: .65 MELD‐Na: .71	p = .043 p = .013 p = .299
Shi et al., Hepatol Res. 2017 15	1245	HBV (67%)	10%	.74	Child‐Pugh: .69 MELD score: .66 MELD‐Na: .70	p < .05 p < .01 p < .05
Li et al., Sci Rep. 2016 51	590	HBV 92%	4.6%	.73	MELD score: .65 MEL D‐Na: .71
Baldin et al., Dig Dis Sci. 2021 16	266	HCV (32%) Alcohol (20%)	Low‐CLIF‐C‐AD: 10% Intermediate – CLIF‐C‐AD: 22% High CLIF‐C‐AD: 62%	Unavailable	Unavailable	Unavailable
Picon et al., World J Gastroenterol. 2017 17	95	HCV (46%) Alcohol (36%)	43%	.70	Child‐Pugh: .73 MELD score: .71 MELD‐Na: .73	Unavailable
Teerasarntipan et al., PLoS One. 2022 52	301	Alcohol (51%) HCV (23%)	33%	.59	Child‐Pugh: .60	p > .05
Alexopoulou et al., Scand J Gastroenterol. 2017 18	104	Alcohol (57%)	31%	.68	Child‐Pugh: .66 MELD score: .66 MELD‐Na: .67	p = .668 p = .799 p = .676
Antunes et al., Gastroenterol Hepatol. 2017 19	577	Alcohol (65%)	31%	.67	Child‐Pugh: .65 MELD score: .64 MELD‐Na: .69	p = .442 p = .296 p = .423

In a subset of 344 patients with AD at enrolment and post‐enrolment follow‐up, the dynamic evaluation of Clif‐C AD score over time (48 h, 3–7 days, 8–15 days) improved predictive performance, with significant results at 1 year.

2.2. Validation

In the initial study, an external validation analysis was conducted on 225 patients from the Royal Free Hospital in London and the Hospital Clinic in Barcelona, with clinical characteristics similar to the CANONIC cohort. Clif‐C AD score improved predictive ability over Child‐Pugh score, MELD score, and MELD‐Na score, although this difference was statistically significant only for 90‐day mortality. ¹⁴

Clif‐C AD score was validated in the Ningbo cohort in China, that included all patients with AD of cirrhosis between 2009 and 2014. Out of 1454 patients in the cohort, 1245 did not have ACLF at inclusion. The main aetiology of cirrhosis was HBV‐related (67%). Baseline and sequential Clif‐C AD score had a higher C‐index than MELD scores and Child‐Pugh scores for predicting mortality at day‐28, 90, 180 or 365. Importantly, the study confirmed that the predictive value of this model was not affected by the cause of cirrhosis. ¹⁵ Authors also validated that a Clif‐C AD score above 60 identified a group of patients with a high susceptibility of progression to ACLF and death. A Brazilian team prospectively enrolled 266 patients with AD of cirrhosis without ACLF. Using the cut‐off values previously described (i.e. 45 and 60), patients with high CLIF‐C AD score had an increased risk of new complications during hospitalization (90% vs. 70% vs. 49% for high, intermediate and low Clif‐C AD score). Patients with high CLIF‐C AD score also had significantly higher 90‐day mortality rate (62%) than patients with low‐ CLIF‐C AD score (10%). ¹⁶

But the superiority of Clif‐C AD score over other scores has not been systematically observed. In another Brazilian prospective cohort study, Clif‐C AD score had an AUROC of .75 for predicting mortality at 28 days, not significantly different from the AUROC of Child‐Pugh score, MELD score, MELD‐Na score and CLIF Consortium organ failure score (CLIF‐C OFs). The lack of significance persisted at 90 days. These results should be interpreted cautiously due to the low number of included patients (n = 95), and the significant number of lost to follow‐up (15%). ¹⁷ Similar findings were observed in a prospective monocentric Greek study (n = 104) where the predictive discrimination ability of Clif‐C AD for mortality at 28, 90, 180 or 365 days was not significantly better than that of MELD, MELD‐Na and Child‐Pugh scores. However, the Greek study was affected by limited access to ICU and transplant facilities, since only two patients of the whole cohort underwent liver transplantation and only three patients were admitted to the intensive care unit. This may explain the high mortality rate observed in the study (30.6% at 90 days in patients without ACLF vs. 12.6% in CANONIC). ¹⁸ In a similar manner, Antunes et al. retrospectively analysed 557 consecutive patients with AD of cirrhosis without ACLF. The 30‐ and 90‐day mortality rates were 10.3% and 31.1%, respectively, higher than those reported in CANONIC. Clif‐C AD score did here again not demonstrate superiority over traditional models in predicting patient prognosis at Day 30 and Day 90. ¹⁹

The PREDICT study, a European multicentre prospective observational study included 1071 patients hospitalized with AD without ACLF at inclusion validated the prognostic value of changes in Clif‐C AD score over the time. ²⁰ Indeed, Clif‐C AD score significantly worsened during the follow‐up of patients who developed ACLF, while it improved in the other patients.

Some studies attempted to extend the utility of Clif‐C AD score to other settings. A large study in China used pooled individual data of 608 patients from two published studies to investigate the prognostic value of Clif‐C AD scores in patients with Child‐Pugh B cirrhosis and acute variceal bleeding. Authors found that Clif‐C AD score outperformed active bleeding at endoscopy and other prognostic models such as MELD, and Child‐Pugh score for predicting 6‐week and 1‐year mortality. However, 19% of patients presented with shock at admission, suggesting potential inclusion of patients with ACLF in the analysis. ²¹ The same team used these two cohorts and a retrospective cohort including all consecutive patients with cirrhosis admitted for acute variceal bleeding at six tertiary academic hospitals in China from June 2016 to December 2019, gathering a total of 1021 patients. After adjusting for potential confounding factors, preemptive TIPS was associated with a reduction in the risk of death, with a hazard ratio of .62 compared to medical treatment alone. When stratifying according to CLIF‐C AD score, this relative risk reduction of death was found in patients with Clif‐C AD score between 48 and 56 and above 56 but not in patients with a Clif‐C AD score below 48. The relative risk reduction of death was more pronounced in patients with CLIF‐C AD score above 56 than patients with Clif‐C AD score between 48 and 56. Clif‐C AD score may thus identify, among patients with cirrhosis and active variceal bleeding, a group of patients with a high risk of mortality in whom preemptive TIPS may be beneficial. ²² In the same setting, Zhu et al. retrospectively included 235 consecutive patients with cirrhosis and acute variceal bleeding without ACLF at admission from a single centre. The area under the receiver operating characteristic curve (AUROC) of Clif‐C‐AD was significantly higher than that of Child‐Pugh, MELD, and MELD‐Na for predicting 6‐week mortality. Authors proposed thresholds of 48 and 60 to stratify into low, moderate and high death risk groups. Limitations of this study included the absence of data on changes in Clif‐C AD over time and on ACLF development. Moreover, patients had a median serum bilirubin of 18.9 mg/dL, which is unusually high. ²³

The utility of Clif‐C AD score was also extended to evaluate the beneficial effect of preoperative TIPS on the development of postoperative ACLF in a retrospective single‐centre study. Forty five patients with cirrhosis undergoing surgery without TIPS were compared with 45 patients with cirrhosis undergoing surgery, but with preoperative TIPS placement (either for refractory ascites or acute variceal bleeding), matched 1:1 using a propensity score. In the multivariate analysis, Clif‐C AD score, type of surgery and surgery without TIPS were associated with ACLF at 90 days. In the no‐TIPS group, a cut‐off of 45 for Clif‐C AD score was able to identify a patient population at high risk of development of ACLF. Patients with TIPS and Clif‐C AD score above 45 had a significantly lower rate of ACLF development after surgery than patients without TIPS. ²⁴

2.3. Limitations

Despite the improvement in prognosis allowed by the Clif‐C AD score, even in the cohort where this score has been built, a notable proportion of incorrect predictions was observed (26% for 90‐day mortality). ¹⁴ This suggests that additional studies and tools are required to enhance the prognostication of patients with AD of cirrhosis. ¹⁴ Another limitation of the Clif‐C AD score, is that up to 50% of patients fall into the ‘grey zone’ with a Clif‐C AD score between 45 and 59. ¹⁴

3. PROGNOSTIC SCORES OF ACLF IN ASIA

3.1. Initial description and performance

In 2009, an Asian Pacific Association for the Study of the Liver [APASL] consensus described ACLF as a clear‐cut acute liver insult, presenting as jaundice and coagulopathy, followed by ascites or encephalopathy within a time frame of 4 weeks in a patient with previously diagnosed or undiagnosed chronic liver disease. ²⁵ The APASL‐ACLF Research Consortium (AARC) database was set up, comprising 5228 patients from 43 centres in 15 countries prospectively enrolled and followed, forming the basis for the last consensus of 2019. ⁴

Using AARC database, AARC score was derived from a cohort of 480 patients and validated in 922 patients. ²⁶ Clinically relevant characteristics and laboratory parameters associated with mortality at Days 4,7 and 28 were included in the score. Total bilirubin, hepatic encephalopathy, INR, serum creatinine and serum lactate were identified as independent predictors of mortality. The AARC ACLF score assigns a subscore between 1 and 3 to each variable according to two cut‐off values. The AARC ACLF score ranges from a minimum of 5 to a maximum of 15 (Figure 1A,B). AUROC of AARC ACLF score was .80 for 28‐day mortality in derivation cohort (Table 2). Three grades were defined according to the score: Grade I for a score of 5–7, Grade II for a score of 8–10 and Grade III for 11–15 with a respective 28‐day mortality of 12.7, 45 and 86%. Change from one grade to another at Day 4 or Day 7 correlated with mortality.

FIGURE 1.

AARC, COSSH, EASL‐CLIF and NACSELD scoring system of ACLF. (A) Component of the AARC scoring system. (B) Grading of ACLF according to AARC scores. (C) ACLF classification according to COSSH ACLF II score. (D) CLIF organ failure score system. The dark grey cells indicate each organ failure and the light grey each organ dysfunction. (E) ACLF classification according to EASL‐CLIF and scores developed by EASL (CLIF‐C ACLF score). (F) NACSELD organ failure score system and NACSELD classification of ACLF. AARC, Asian Pacific association for the study of the liver ACLF research Consortium; ACLF, acute‐on‐chronic liver failure; CLIF‐C, chronic liver failure‐Consortium; CLIF‐C‐OFs, chronic liver failure Consortium organ failure score; COSSH, Chinese group on the study of severe hepatitis B; FIO2, fraction inspired of oxygen; HE, hepatic encephalopathy according to West Haven criteria; INR, international normalized ratio; MAP, mean arterial pressure; NACSELD, North American Consortium for the study of end‐stage liver disease; PaO2, partial pressure of arterial oxygen; RRT, renal replacement therapy, SpO2, oxygen saturation measured with pulse oximetry.

TABLE 2.

Main ACLF prognostic scores used across the world.

	AARC score	COSSH‐ACLF II	Clif‐C ACLF score a	NACSELD‐ACLF
Initial description	Choudhury et al. Hepatol Int. 2017: 480 patients 26	Li et al., J Hepatol. 2021: 954 patients 28	Jalan et al., J Hepatol. 2014: 275 patients 33	O'Leary et al., Hepatology. 2018: 1605 patients 45
C‐index or AUROC in the initial study	.8 for 28‐day mortality	.826 and .809 for 28‐day and 90‐day mortality respectively	.76, .73, .72 and .71 for 28‐day, 90‐day, 6‐month and 1‐year mortality respectively	.81 for 30‐day mortality
Cause of cirrhosis	Alcohol (59%), HBV (17%), NASH (7%)	HBV (100%)	Alcohol (52%), Hepatitis C virus infection (20%)	Alcohol (46%)
Validation	Internal validation in the initial study: 922 patients Verma et al., Hepatol Int. 2021. (n = 3692) 30	External validation in the initial study: 321 patients Yu et al., Front Microbiol. 2022. (n = 919) 31 Dong et al., Sci Rep. 2020. (n = 799) 53	External validation in the initial study: 225 patients Picon et al., World J Gastroenterol. 2017. (n = 42) 17 Silva et al. Liver Int. 2014. (n = 46) 38 Dhiman et al., World J Gastroenterol. 2014 (n = 50) 39 Shi et al., Hepatol Res. 2017 (n = 209) 15 Shi et al., Hepatology. 2015 (n = 405) 41 Kuo et al., J Pers Med. 2021 (n = 135) 54 Ramzan et al., Cureus. 2020 (n = 75) 55 Engelmann et al., Crit Care Med. 2018 (n = 202) 56 Maipang et al., PLoS One. 2019. (n = 343) 57 Li et al., Ann Palliat Med. 2020 (n = 529) 58 Zhang et al., PeerJ. 2020 (n = 102) 59 Chen et al., J Clin Med. 2020 (n = 249) 60 Chirapongsathorn et al., JGH Open. 2022 (n = 346) 61 Karvellas et al., Crit Care Med. 2018 (n = 867) 62 Teerasarntipan et al., PLoS One. 2022 (n = 301) 52 Dong et al., Sci Rep. 2020. (n = 799) 53 Barosa et al., Rev Esp Enferm Dig. 2017. (n = 49) 63 Li et al., Sci Rep. 2016. (n = 300) 51	Internal validation in the initial study: 1070 patients Rosenblatt et al., Liver Transpl. 2020 (n = 106 634) 46
Study not validating	Chirapongsathorn et al., JGH Open. 2022 (n = 346) 61		Alexopoulou et al., Scand J Gastroenterol. 2017 (n = 78) 18 Antunes et al., Gastroenterol Hepatol. 2017 (n = 222) 19 Dupont et al., Dig Liver Dis. 2015 (n = 281) 43
Limitation	Mainly based on the presence on liver dysfunction Hepatic encephalopathy evaluation may be possibly subjective Same score to all patients with a serum lactate above 2.5 mmol/L and a serum creatinine level above 1.5 mg/dL which may underestimate mortality in patients ‘very sick’ with very high serum creatinine or lactate level		Liver failure, is defined by increased serum bilirubin level, which may be questionable. Renal dysfunction is defined by elevated serum creatinine above 2 mg/dL. Serum creatinine is affected in patients with cirrhosis by several factors, overestimate glomerular filtration rate in patients with cirrhosis. Moreover, the value of 2 mg/dL may be questionable	Criteria used in the NASCELD‐score are representative of advanced circulatory, brain, respiratory and renal failure. These criteria should be used to better predict the futility of care

Abbreviations: AARC, APASL‐ACLF Research Consortium; CLIF‐C, CLIF Consortium; COSSH, Chinese group on the study of severe hepatitis B; NACSELD, North American Consortium for the study of end‐stage liver disease.

^a Studies evaluating CLIF‐SOFA are not included. In a same manner, studies evaluating only CLIF‐OFs were not included.

The Chinese group on the study of severe hepatitis B (COSSH) conducted a prospective study at 13 liver centres in Chinese university hospitals, enrolling 1322 patients with chronic hepatitis B from June 2013 to October 2016. They proposed a definition of HBV‐related ACLF, expanding on the definition of EASL‐CLIF by including patients with chronic hepatitis B and single liver failure (INR ≥1.5 and total bilirubin ≥12 mg/dL) (Figure 1C,D). These criteria were found to be more sensitive in diagnosing HBV‐related ACLF than EASL‐CLIF criteria. ²⁷ Based on this study, they proposed the COSSH‐ACLF, and then the COSSH‐ACLF II score, including INR, hepatic encephalopathy, total bilirubin, neutrophils, urea and age (Figure 1D, Table 3). The C‐index of the latter for 28 and 90‐day mortality was .826 and .809, respectively (Table 2), which was higher than CLIF‐C‐ACLF, MELD and MELD‐Na. ²⁸

TABLE 3.

Variables used in each scoring system.

3.2. Validation

3.2.1. AARC ACLF score

In the initial study, a validation was performed using data from 922 patients from AARC database. ²⁶ The AUROC of AARC ACLF score was .78 for 28‐day mortality in the validation cohort. ²⁶

A Chinese team conducted a retrospective analysis of 786 patients, of whom 196 had ACLF according to APASL criteria, from the Medical Information Mart for Intensive Care III database. This open database consists of more than 40 000 ICU patients who stayed at the ICU of Beth Israel Deaconess Medical Center (USA) between 2001 and 2012. The AARC score had an AUROC of .75 for 28‐day mortality and was superior to the Child‐Pugh score, but there were no significant differences compared to other scores, including MELD, Clif‐SOFA, Meld‐NA and Clif‐C ACLFLact. In 212 patients with laboratory examinations available at Days 4–7, a change from Grade 1 to Grade 2 or Grade 2 to Grade 3 was associated with increased mortality. When Grade 2 changed to Grade 1, the mortality rate decreased. However, it is important to note that this study included all patients with cirrhosis in the critically ill unit, regardless of ACLF status. Moreover, only patients with cirrhosis were included, unlike the initial study where patients with chronic liver disease without cirrhosis were also included. ²⁹

A prospective cohort of 3692 patients with a diagnosis of APASL‐ACLF recruited from the AARC Consortium showed that Day 7 AARC model had highest c‐index (.872) for 30‐day mortality compared to MELD‐lactate, MELD, Clif‐C ACLF and NASCELD‐ACLF score. ³⁰ Patients with Day 7 AARC score > 12 had the highest 30‐day mortality rate (94%). ³⁰ Conversely, at baseline, Clif‐C ACLF score had higher c‐index (.820) than AARC score (.815). ³⁰

3.2.2. COSSH‐ACLF score

In the initial study, an external validation was included using data from 321 patients with HBV‐ACLF. ²⁸ C‐index of COSSH‐ACLF II score was .835 for 90‐day mortality, higher than that of MELD and MELD‐Na, but not to CLIF‐C ACLFs. Moreover, a Chinese team conducted a retrospective analysis of two prospective cohort and focused on patients with HBV‐ACLF according to COSSH‐ACLF criteria. Among 919 patients with HBV‐ACLF, COSSH‐ACLF II had a C‐index of .739 for 90‐day mortality and was superior to CLIF‐C ACLF, MELD and MELD‐Na. ³¹

3.3. Limitations

Studies using AARC criteria have shown that HBV reactivation was the most frequent trigger of ACLF. Other precipitating event of ACLF were HEV infection and drug‐induced liver injury. ³² These triggers are not commonly observed in Western countries. Therefore, the applicability of the AARC score in Western countries still needs to be proven.

It is essential to note that the AARC score is primarily based on the presence of liver dysfunction, and extrahepatic organ failures are not included in the score, except for serum creatinine. The evaluation of hepatic encephalopathy may involve a degree of subjectivity. Additionally, the AARC score assigns the same score to all patients with a serum lactate above 2.5 mmol/L and a serum creatinine level above 1.5 mg/dL, which may underestimate mortality in patients who are ‘very sick’ with very high serum creatinine or lactate levels. This highlights the potential limitations of the AARC score, and further research is needed to assess its performance in diverse patient populations and healthcare settings, especially in Western countries where the aetiology of liver diseases and precipitating events may differ.

4. PROGNOSTIC SCORES OF ACLF IN EUROPEAN COUNTRIES

4.1. Initial description and performance

The Clif Consortium used the CANONIC study to develop the European definition of ACLF and grade its severity. Authors initially developed a modified Sequential Organ Failure Assessment score (the Clif‐SOFA), ⁹ a score based on organ function, which has been 1 year later simplified into the CLIF‐C OFs ³³ (Figure 1C). In order to improve prediction of death, CLIF‐C OFs score has been combined with age and white blood cell count to develop a specific prognostic score for ACLF, called CLIF Consortium ACLF score (CLIF‐C ACLFs) ³³ (Figure 1D,E) (Table 3). The C‐index of the Clif‐C ACLF for 28‐day, 90‐day, 6‐month and 1‐year mortality (.76, .73, .72 and .71) was significantly better than that obtained with the Clif‐C OFs, Child‐Pugh Score, MELD and MELD‐Na ³³ (Table 2). The Clif‐C ACLF computed at 3–7 days and 8–15 days after diagnosis of ACLF, predicted 28‐day and 90‐day mortality significantly better than Clif‐C ACLF at diagnosis. ³⁴

4.2. Validation

An external validation was included in the initial study, using data from 225 patients with ACLF consecutively admitted to the ICU at Paul Brousse hospital, Villejuif, France and followed up during 90 days. Clif‐C ACLF had a better ability to predict mortality at Day 28 and Day 90 than Child‐Pugh score, MELD and MELD‐Na. ³³

In the CANONIC population, the clinical course of ACLF patients during early follow‐up (28 days) was investigated. Interestingly, the 28‐ and 90‐day mortality rates in patients with initial ACLF‐2 or ‐3 who achieved ACLF resolution during hospitalization were not significantly different compared to rates observed in those with ACLF‐1. Clif‐C‐ACLF at ACLF diagnosis was the only predictor independently and significantly associated with severe early course (final grade of ACLF‐2 or 3). ³⁴

In hospitalized patients with cirrhosis, acute kidney injury is associated with an increased risk of death. ³⁵ , ³⁶ In the CANONIC population, ACLF grade, as proposed by the EASL‐CLIF‐Consortium, had greater prognostic accuracy compared to acute kidney injury classification in the prediction of 28‐day and 90‐day mortality. ³⁷

In a prospective study conducted in a tertiary Brazilian centre, including 192 consecutive patients with acute decompensation of cirrhosis between December 2013 and November 2013, the Clif‐SOFA score, had higher AUROCs than MELD and Child‐Pugh scores for predicting 30‐day mortality and 90‐day mortality. ³⁸ A prospective study in India included patients with cirrhosis and acute decompensation from July 2013 to December 2013. ACLF was defined as per the EASL‐CLIF criteria and compared to ACLF defined as per the APASL criteria. Patients with prior decompensated cirrhosis were not considered as having ACLF as per the APASL criteria. Out of the 50 patients recruited, ACLF was present in 38 (76%) patients as per the Clif‐SOFA and 19 (38%) as per APASL criteria. Twenty‐eight‐day and 90‐day mortality between no ACLF and ACLF group using the APASL definition was not significantly different, contrary to patients defined as per Clif‐SOFA. AUROC of Clif‐SOFA for 28‐day mortality was better than AUROC of Child‐Pugh and MELD scores. This difference may be explain by the ‘Western‐type’ patients, as the main cause of cirrhosis was excessive alcohol consumption (58%), and as infection was present in 66% of the patients at admission, while 56% had prior history of decompensated cirrhosis. ³⁹

Using a very large dataset (n = 48 547) of adult patients who were listed for liver transplantation, Li and Thuluvath demonstrated a clear discordance in the prevalence of ACLF and in 30‐day mortality rates of patients diagnosed with ACLF by Clif criteria and NASCELD criteria. The AUROC for 30‐day all‐cause mortality based on ACLF criteria, as defined by the EASL‐Clif and the NASCELD, was significantly higher for Clif‐criteria (.768) than NASCELD criteria (.597). ⁴⁰

In a prospective cohort study conducted in Brazil in a tertiary care centre, Clif‐C ACLF had an AUROC of .71 for predicting mortality at 28 days, better than that of Child‐Pugh, MELD, MELD‐Na and CLIF‐C OFs scores. The difference was not significant at 90 days. These results should be interpreted carefully given the low number of patients included (n = 42). Moreover, among patients with ACLF at inclusion or developing ACLF, mortality was particularly high (79%) despite the fact that more than 70% had ACLF grade‐1. ¹⁷

The Ningbo cohort included all patients with AD of cirrhosis between 2009 and 2014. Out of 1454 patients, 209 had ACLF at inclusion. The main cause of cirrhosis was HBV‐related (67%). Baseline and sequential evaluation of Clif‐C ACLF performed better than MELD, MELD‐Na and Child‐Pugh score for predicting mortality at Days 28, 90, 180 and 365. Interestingly, sequential evaluation of CLIF‐C ACLF at Days 4–7 had a C‐index of .869 for predicting mortality at Day 28. Most included patients had HBV‐related cirrhosis confirming that the predictive value of this model was not affected by the cause of cirrhosis. Using a cut‐off of 45, authors were able to differentiate two groups: a group of patients with a high 28‐day mortality (66%) who may require immediate liver transplantation, and a group of patients with a lower risk (27%) who should be treated with organ support first and then evaluated for liver transplantation in the absence of improvement of Clif‐C‐ACLF score over time. ¹⁵

The same team analysed retrospectively 405 patients with ACLF in the same cohort (266 ACLF at inclusion and 139 patients developing ACLF during hospitalization). ⁴¹ Acute decompensation was defined as ascites, hepatic encephalopathy, upper gastrointestinal bleeding, bacterial infection (i.e. the European definition) and acute severe hepatic damage according to APASL criteria. ACLF was diagnosed according to Clif‐SOFA score. ACLF (322) patients with definite precipitating events were categorized into two groups according to types of acute insults: hepatic‐ACLF or extrahepatic‐ACLF. Patients with hepatic‐ACLF had HBV‐related cirrhosis (90%). Liver and coagulation failure were highly prevalent in hepatic‐ACLF patients, whereas extrahepatic‐ACLF patients developed higher frequency of extrahepatic organ failure. Ninety‐day and 1‐year mortality was significantly higher in patients in the extrahepatic‐ACLF group. In patients with extrahepatic‐ACLF, Clif‐C‐ACLF had the highest predictive value for predicting mortality at 28 days, 90 days and 1 year, whereas in the hepatic‐ACLF patients, the so‐called ‘integrated MELD’ (iMELD) model, a score proposed for ACLF patient, ⁴² had the highest AUROC. Patients in hepatic‐ACLF group had mostly HBV‐related cirrhosis and precipitating events were mostly HBV flares or superimposed infection of other hepatitis viruses. This presentation is not typical for ACLF in Western countries, where patients with prior decompensation, alcohol or HCV‐related cirrhosis and ACLF precipitated mostly by non‐viral insults, such as bacterial infection, variceal bleeding or hepatic insults like active drinking, are more common. This difference in aetiology and presentation may explain the lower predictive value of CLIF‐C ACLF in the population with hepatic‐ACLF.

Some studies did not find a superiority of Clif‐C ACLFs over other scores. A Greek and a Brazilian study did not find a higher predictive discrimination ability of Clif‐C ACLF for mortality at 28, 90180 or 365‐days than MELD, MELD‐Na and Child‐Pugh scores and for mortality at 30 and 90 days than MELD, MELD‐Na and Child‐Pugh scores respectively. ¹⁸ , ¹⁹ Another retrospective analysis of patients admitted to an intermediate care unit, upstream of ICU in Caen, France between 2009 and 2010 did not show superiority of Clif‐C‐ACLF score over SOFA, MELD, MELD‐Na, Child‐Pugh, SAPS II and SAPS III. ⁴³ However, this retrospective study included all patients admitted for AD of cirrhosis, regardless of ACLF status and with a high proportion of gastrointestinal bleeding at admission (47%).

4.3. Limitations

Clif‐C ACLF score includes CLIF‐C OFs score to diagnose and grade the severity of ACLF. Each organ system in CLIF‐C OFs is assigned a subscore between 1 and 3 based on two cut‐off values. Some of these items can be less reliable in certain circumstances. Liver failure is defined by an elevated serum bilirubin level; yet elevated serum bilirubin levels can occur in other conditions such as bile duct obstruction or sepsis, leading to potential confounding factors. Renal dysfunction in CLIF‐C OFs is defined by serum creatinine level above 2 mg/dL. However, according to the international Club of Ascites, acute kidney injury is defined by an increased in serum creatinine ≥.3 mg/dL or a percent increase of creatinine ≥50% from baseline; moreover patients with chronic kidney disease may have elevated serum creatinine level without acute kidney injury.

5. PROGNOSTIC SCORES OF ACLF IN NORTH AMERICA

5.1. Initial description and performance

Bajaj et al. retrospectively analysed the NACSELD (North American Consortium for the study of end‐stage liver disease) database, which consists of prospectively collected data from patients with cirrhosis hospitalized in hepatology centres across the USA and Canada. Initially, authors focused on 507 patients with cirrhosis and infection, and defined organ failure based on specific criteria: (i) hepatic encephalopathy grade III or IV by West Haven criteria, (ii) shock defined as mean arterial pressure < 60 mmHg or a reduction of 40 mmHg in systolic blood pressure from baseline despite adequate fluid resuscitation, (iii) need for mechanical ventilation and (iv) need for dialysis or other forms of renal replacement therapy (Table 3). Thirty‐day mortality in patients with one organ failure was 27%, and 48% in patients with two or more organ failures. ACLF was defined as two or more organ failures. ⁴⁴ Later, the definition of ACLF was extended to patients without infection. Using a larger database of 2675 patients with cirrhosis across 14 centres in North America, the cohort was split into a 60/40 ratio for training and validation cohorts. Patients who met the criteria for NASCELD‐ACLF (≥2 organ failures) had an overall 41% 30‐day mortality (48% if they were infected vs. 24% if they were non‐infected) compared to 7% in patients without NASCELD‐ACLF ⁴⁵ (Figure 1F). A multivariable logistic regression model was fitted with NACSELD‐ACLF, age, white blood count, serum albumin, MELD and infection status. The strongest predictor of 30‐day mortality remained NACSELD‐ACLF for infected and uninfected patients. The AUROC of the NACSELD‐ACLF model was .807 and .853 for 30‐day mortality in training and validation set respectively (Table 2).

5.2. Validation

An external validation of the NACSELD‐ACLF score was conducted using a large national cohort database of hospital discharges. The study included patients with cirrhosis and a decompensating event. Out of 1 523 478 patients included, 106 634 met the threshold for the NACSELD‐ACLF score. The ACLF score was significantly associated with a marked reduction in inpatient survival. The AUROC for the NACSELD‐ACLF score was .77 (95% CI, .77–.78), indicating good discriminatory ability for predicting outcomes). ⁴⁶

5.3. Limitations

None of the above‐mentioned studies was prospective. The validation study used a large national database which is prone to administrative coding errors. Moreover, some important laboratory data were missing such as MELD, Child‐Pugh score and white blood cell count. Criteria used in the NASCELD‐score are representative of advanced circulatory, brain, respiratory and renal failure. These criteria might be better suited to predict the futility of care.

6. CONCLUSION AND PERSPECTIVES

Several scores have been developed in recent years to predict the outcome of patients with AD and ACLF (Table 3). As the genetic background, causes of cirrhosis and type of precipitating events vary considerably around the world, we can speculate that each score is best suited to the population from which it was derived, namely CLIF‐C‐ACLF for Western type patients, AARC score for Asian patients and COSSH for HBV infected patients. The criteria used in the NACSELD‐ACLF score are representative of advanced disease and may be more useful in predicting futility of treatment. The Clif‐C AD score is the only one proposed so far for AD without ACLF and can be used to stratify patients according to their risk of 90‐day, 6‐month and 1‐year mortality.

Further improvements in scoring systems are yet still needed. Indeed, whatever the score and definitions used across the world, AUROC for predicting mortality does not exceed .8 (Figure 2). This is likely explained by the large inter‐individual variability of precipitating events and clinical presentations, but also by the fact that important factors involved in the pathophysiology of acute decompensation of cirrhosis and ACLF are not taken into account in currently available prognostic tools. ⁴⁷ These processes include (a) systemic inflammation, identified as the cornerstone in AD and ACLF, ⁴⁸ that can be captured by whole blood transcriptomics, plasma concentrations of cytokines and lipid mediators; (b) major disorders of energetic metabolism, that can be reflected by metabolomics. Indeed, ACLF is characterized by (i) increased systemic catabolic metabolism, which leads to intense lipolysis, glycogenolysis and proteolysis and release of fatty acids, glucose and amino acids to the immune system and peripheral organs; (ii) anabolic and increased energetic metabolism (ATP synthesis) by innate immune cells; (iii) reduced mitochondrial respiration and energy production by peripheral non‐immune cells ⁴⁸ ; (c) genetic predispositions favouring specific gene‐by‐environment interactions that can be captured by genomics; (d) biological age reflecting the actual functional state of the organism, that can been captured by DNA methylation ⁴⁹ ; (e) multiple organ involvement, as decompensation of cirrhosis can be associated with injury or altered function of several organs (gut, heart, kidneys, brain and splanchnic vascular bed). These impaired organs release mediators, including extracellular vesicles and microRNAs, which act remotely on other organs and promote their dysfunction, but can also be measured in the blood. ⁵⁰

FIGURE 2.

AUROC of the main AD and ACLF prognostic scores used across the world for mortality at 90 days and 28 days, respectively. The size of the circles represents the number of patients included in each study. For ACLF score (AARC, COSSH, CLIF‐C ACLF and NACSELD), AUROC at 28 days are represented. AUROC for 90‐day mortality are represented for CLIF‐C AD. AARC, APASL‐ACLF research Consortium; CLIF‐C AD, chronic liver failure Consortium acute decompensation score; CLIF‐C ACLF, chronic liver failure Consortium ACLF score; COSSH, Chinese group on the study of severe hepatitis B; NACSELD, North American Consortium for the study of end‐stage liver disease.

Integrating the large amount of data generated by omics approaches poses challenges that can be addressed by system approaches. The system approach is a method of problem solving and decision making that focuses on understanding and analysing complex systems as a whole rather than focusing on individual components or parts. It is a holistic approach that recognizes that the behaviour of a system is determined by the interactions and relationships between its parts, rather than by the parts themselves. In the system approach, a system is viewed as an interconnected set of elements or components that work together to achieve a common goal. By studying and integrating large sets of clinical data together with omics data, system medicine aims to gain a comprehensive understanding of disease mechanisms and develop personalized treatment strategies.

It is precisely the objective of the European Union's Horizon 2020 research and innovation programme DECISION (https://decision‐for‐liver.eu/) to better understand the pathophysiology of decompensated cirrhosis leading to ACLF at the systems level by taking advantage of already existing large and clinically well‐characterized cohorts of patient, and by performing on those cohorts multiomics analyses to develop novel tools to improve prediction of patients outcome (Figure 3).

FIGURE 3.

Pathophysiological concepts in acute decompensation of cirrhosis, basing the DECISION research program. https://decision‐for‐liver.eu/.

FUNDING INFORMATION

P‐E.R's research laboratory is supported by the Fondation pour la Recherche Médicale (FRM EQU202303016287), ‘Institut National de la Santé et de la Recherche Médicale’ (ATIP AVENIR), the ‘Agence Nationale pour la Recherche’ (ANR‐18‐CE14‐0006‐01, RHU QUID‐NASH, ANR‐18‐IDEX‐0001, ANR‐22‐CE14‐0002) by ‘Émergence, Ville de Paris’, by Fondation ARC, by the European Union's Horizon 2020 research and innovation programme under grant agreement No 847949 and by France 2030 RHU LIVER‐TRACK.

CONFLICT OF INTEREST STATEMENT

P‐E.R. has received research funding from Terrafirma and acted as consultant for Hemostod, Mursla, Genfit, Boehringer Ingelheim and Abbelight, and received speaker fees from Tillots pharma and AbbVie.

Valainathan SR, Xie Q, Arroyo V, Rautou P‐E. Prognosis algorithms for acute decompensation of cirrhosis and ACLF . Liver Int. 2025;45:e15927. doi: 10.1111/liv.15927