The unwell patient with advanced chronic liver disease: when to use each score?

Abstract

Background

Prognostication in chronic liver disease and the implementation of appropriate scoring systems is difficult given the variety of clinical manifestations. It is important to understand the limitations of each scoring system as well as the context and patient group from which it was developed to allow appropriate application. This review seeks to explore the optimal clinical uses of different predictive scores developed for compensated and decompensated chronic liver disease, acute on chronic liver failure, and hepatocellular carcinoma. We will also review future areas of research for each score and current gaps in the literature.

Main body

The Child–Pugh score is the pre-eminent prediction score for liver disease that was developed through empiric selection of relevant variables. It is useful for selection of patients for surgical resection of hepatocellular carcinoma but is inferior to other scores for other clinically relevant endpoints such as survival in acute decompensations. The Model for End-Stage Liver Disease (MELD) score and subsequent variants (MELD-Na, MELD 3.0) were developed to predict mortality following elective transjugular intrahepatic portosystemic shunt (TIPS) insertion. An alternative is the Frieberg Index of Post-TIPS Survival (FIPS) score, which has been externally validated for TIPS populations. Organ allocation for liver transplantation is also currently prioritised using the MELD score, with the MELD 3.0 reducing waitlist gender discrepancies. The Chronic Liver Failure Consortium (CLIF-C) acute decompensation (AD) and acute-on-chronic liver failure (ACLF) scores are used for predicting mortality in cirrhotic patients with acute decompensation of liver disease and acute-on-chronic liver failure, respectively. Both scores were developed from retrospective analyses of an observational European cohort with external validation. Understanding of ACLF presentation of advanced liver disease remains in the preliminary stages. Improving collective understanding is important to optimise prognostication. The albumin-bilirubin score is a non-invasive predictor of survival in patients with hepatocellular carcinoma. Incorporating artificial intelligence to personalise predictive algorithms may provide the most effective prognostication for all clinical phenotypes.

Conclusion

We summarised key prognostic scores used in advanced liver disease and make recommendations for the optimal uses. Nuances in the development and implementation of each are discussed to help guide effective use.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12916-025-04185-w.

Background

In the management of chronic liver disease, survival and prognosis are used to guide treatment decisions and discussions with patients and their caregivers. For a clinician, it is important to have appropriate tools to provide an accurate and realistic assessment of the natural course of a patient’s disease. Multiple scores have been developed to model survival outcomes in advanced chronic liver disease (ACLD). The first widely utilised score, the Child–Pugh score (CPS), remains one of the preeminent models for cirrhosis despite having been developed through empirical selection of relevant variables [1, 2]. More recently, contemporary scores have been derived from statistical analysis for variables independently predictive of mortality: the Model for End-Stage Liver Disease (MELD) score has historically represented the most used alternative to the CPS [2, 3]. However, the landscape of ACLD has evolved since the development of these scores and will continue to change [4, 5]. Aetiological causes of ACLD have transitioned from predominantly viral hepatitis to alcoholic-associated liver disease and metabolic dysfunction-associated fatty liver disease (MAFLD) [6]. Whilst the optimal diagnostic criteria for steatotic liver disease remain unclear, extended definitions now include patients that may not have been considered to have liver disease when index scores were derived [7]. ACLD may also manifest in a variety of phenotypes including acute decompensations (AD) of ACLD, acute on chronic liver failure (ACLF), stable ACLD in patients awaiting orthotopic liver transplant (OLT) and those with hepatocellular carcinoma (HCC). There are varied outcomes for each of these conditions [8]. Moreover, the collective understanding of ACLF is incomplete, with a unifying definition not yet established [9]. The advancement of contemporary understanding of ACLD requires re-assessment of prognostic tools. The aim of this review is to summarise, appraise, and compare current scores assessing disease in ACLD and highlight when each may be most useful.

Main text

Child–Pugh score

In 1964, the CHILD score was developed as a stratification tool for surgical risk in patients with liver cirrhosis. Bilirubin, albumin, ascites size, encephalopathy grade, and nutritional status were empirically selected based on observed association [1]. The qualitative variables, ascites, encephalopathy, and nutritional status were each stratified into three groups based on severity, whilst continuous variables, bilirubin, and albumin were stratified by empiric selection of cut-off values. This enabled the manufacturing of a score between 5 and 15 for each patient. However, literature regarding surgical risk during this period did not support malnutrition as an independent predictor of survival [10]. The Child–Pugh score (CPS) was subsequently proposed in 1973, wherein nutritional status was replaced with prothrombin time [11]. On the surface, the score now effectively represented synthetic (albumin, international normalised ratio (INR)) and metabolic (bilirubin, encephalopathy) functions of the liver, with cut-off points for prothrombin time guided by clinical experience. Patients were then able to be distributed across grades of disease (A: 5–6, B: 7–9, C: 10–15) based on their total score from each of the 5 components.

The CPS was the first widely studied score for ACLD [2]. It was initially verified for patients being evaluated for surgery for portal hypertension [11], with evidence that its use in selecting surgical candidates could be expanded. Small meta-analyses of patients undergoing both elective oesophageal (12 included studies) and pancreatic (8 included studies) resections found that outcomes were better for Child–Pugh A patients [12, 13]. However, there are no large meta-analyses examining surgical outcomes for patients based on CPS, or indeed validating it as superior to other prognostic scores. Current expert consensus in the American Gastroenterological Association guidelines suggests cautious consideration of its use as part of a pre-surgical risk assessment [14].

Integration of the CPS, as a measure of compensation status of ACLD, helps refine the Barcelona Clinic Liver Cancer (BCLC) strategy for treatment selection in HCC[15]. It has been shown to be useful for guiding patient selection for surgical resection[16] and systemic therapy[17, 18], as well as established as a predictive factor for mortality following transarterial chemoembolisation (TACE)[19]. Recognising the importance of the CPS in making treatment decisions in HCC, a meta-analysis aimed at evaluating first-line treatments for patients with CPS 8 or higher is currently underway (PROSPERO, CRD42024528193).

The CPS score has failed to be validated in several other contexts. Early studies demonstrated that the CPS may have prognostic value in patients with AD [20–23], however, in a meta-analysis by Wu and colleagues, it did not reach the threshold value for predicting mortality (p = 0.28) [24]. 16 studies were incorporated from countries across Europe, South America, and Asia attempting to predict mortality outcomes following hospitalisation for AD, but significance was not met. Similarly, it has not demonstrated good discriminatory capacity for patients with ACLF [25–28]. A meta-analysis by Zheng and colleagues showed that the CPS performed worst out of all considered scores, with an AUROC of 0.71 across 16 studies considered [28]. Further, whilst a retrospective study by Pohl and colleagues suggested its use in post-transjugular intrahepatic portosystemic shunt (TIPS) mortality [29], meta-analysis finds it non-superior to the MELD score [30]. There are currently two registered meta-analyses designed to assess predictive factors for survival post-TIPS, of which CPS is considered a variable (PROSPERO, CRD420251007253, and CRD42024502288).

The intrinsic advantage of the CPS is that it is easy to interpret through the provision of distinct risk groups. When compared to other continuous scores, it provides a simple and clear output that is useful for conveying information to patients that can be translated into overall survival (Additional File 1: Table S1) [3]. Each variable has also been shown to be an independent predictor of mortality. D’Amico and colleagues conducted a systematic review to identify significant predictors of survival in cirrhosis. Across 118 studies, the most common were the five variables in the CPS as well as the score itself and age [2]. A recent meta-analysis of biochemical predictors of decompensation in ACLD by Gananandan and colleagues validates that INR, albumin, and bilirubin are independent predictors [31]. However, whilst univariate models can identify clinical parameters as independent predictors of survival, multivariate analysis is required to confirm that when combined, they all still have an independent impact [32]. Given the complex interactions between various organs for synthesis and metabolism, expert opinion suggests the CPS is more appropriate as a multiorgan assessment in patients with ACLD rather than a reflection of liver function [3]. A retrospective audit of 1080 patients by Charif and colleagues demonstrated that whilst all elements were predictors in univariate analysis, they did not remain independent predictors when incorporated into multivariate analysis [33]. The other limitation of empirically selecting variables is that other important predictors of survival may be omitted. It is now well understood that renal function [34, 35] and hyponatremia [36, 37] are important predictors of survival, neither of which are accounted for by the CPS. These biochemical abnormalities are characteristic of the hyperdynamic circulatory phenotype of portal hypertension, a key pathological process that the CPS does not consider [38, 39]. Simplifying the range of scores to 10 (across 3 grades) limits how discriminatory it can be. The conversion of continuous variables to categorical groups may achieve greater simplicity, but it is at the loss of statistical power and makes the score more susceptible to confounders [40]. Ultimately, despite being the index score developed in the field, clinical applicability remains limited.

MELD score and subsequent variants (MELD-Na, MELD 3.0)

With the development and increasing utilisation of novel rescue procedures to manage bleeding oesophageal varices, the MELD score was developed in 2000 to predict 3-month mortality following TIPS insertion [41]. The initial score was developed in a retrospective cohort of 231 patients undergoing elective TIPS insertion between 1991 and 1995. The study excluded those with severe cardiopulmonary or intrinsic renal disease, whilst OLT was not considered a disease endpoint, which may lead to observation bias. The derivation study internally verified the MELD scores’ capacity to predict mortality in patients following TIPS insertion [41], which was quickly expanded to patient selection for liver transplantation (Additional File 1: Table S2) [42, 43].

The combination of serum sodium level and MELD score soon became of interest given that hyponatremia is an independent predictor of risk of complications following liver transplantation [44, 45]. Kim and colleagues retrospectively analysed a cohort of liver transplant patients in order to determine the predictive ability of the MELD score and the MELD score plus sodium for 3-month mortality (Additional File 1: Table S3) [37]. From 2016, the MELD-Na score was adopted by UNOS for deceased donor liver allocation [46]. Objectivity, however, was not pursuant to the implementation of the MELD or MELD-Na scores for donor allocation as there were higher rates of waitlist mortality in female patients [47, 48]. The MELD score predicts lower overall mortality in female patients, possibly due to reduced creatinine production [49]. Kim and colleagues again used a large cohort of 29,410 waitlist candidates to amend the MELD score, re-weighing the importance of creatinine whilst introducing albumin and sex as covariates in the MELD 3.0 [50]. The derivation cohort demonstrated re-stratification of 8.8% of patients to a higher score, thereby increasing the effectiveness of organ allocation with improved discrimination in comparison to MELD-Na (concordance statistic 0.869 vs. 0.862, p < 0.01). It has been adopted as the current allocation standard by UNOS whilst European guidelines support its use [51]. An external retrospective study engaging 6071 transplant candidates identified that female sex is not an independent predictor when incorporated into Cox regression analysis with the MELD 3.0 [52]. Whilst it was powered to detect significant differences, the design does not effectively test the impact of the MELD 3.0 and prospective studies are needed to assess whether it truly accounts for gender discrepancies.

The MELD 3.0 has been externally validated by retrospective studies in Korea and Singapore for mortality prediction for inpatients with cirrhosis. Both studies demonstrated better discrimination compared to prior iterations at 3 and 6 months [53, 54]. A 2018 meta-analysis by Wu and colleagues demonstrated that the MELD score met the threshold value for significance in predicting mortality for patients with AD at 3 (p = 0.010) but not 6 (p = 0.760) or 12 months (p = 0.139) [24]. Subgroup analysis found that across 4 studies the MELD score was significant in predicting mortality following an upper gastrointestinal (GI) bleed (p < 0.001). A systematic review by D’Amico and colleagues assessing the MELD score’s ability to predict all-cause mortality for patients with cirrhosis is currently underway [55]. Secondary outcomes of all-cause mortality for patients with TIPS, alcoholic liver disease or OLT are also being examined.

There are other potential uses for the MELD score. Guidelines still support the use of the MELD score as a tool to grade the severity of alcoholic hepatitis, with the STOPAH trial showing superiority to Maddrey’s discriminant function [56, 57]. A registered trial by Tschaen and colleagues is underway performing a meta-analysis to compare the two (PROSPERO, CRD42024562395). A small meta-analysis of 5 studies did not find a statistical relationship between MELD score and mortality in MAFLD patients undergoing OLT; however, it was the opinion of the authors that, based on qualitative findings, a higher score conveys worse prognosis [58]. The BOMASH trial (Clinical trial ID: NCT06813508) is a prospective analysis attempting to establish predictive factors for OLT and HCC in patients with MAFLD that may help establish this relationship.

The intrinsic strengths of the MELD scores come from their objective construction as a product of multivariate analysis, representing the best method to minimise confounders and increase statistical power [32]. Being continuous scores, they also possess higher statistical power [59]. The MELD scores are also independent of disease aetiology and decompensated sequelae (ascites, HE, acute variceal bleeding). The increasing burden of metabolic associated fatty liver disease and the decreasing incidence of ACLD due to hepatitis B virus (HBV) and hepatitis C virus (HCV) infection indicates that the dominant liver disease aetiologies have changed, and disease courses are likely to be different[5, 60, 61]. As the MELD score is not affected by disease aetiology, it is more resistant to the changing etiological landscape of liver disease [60, 61].

Whilst the components of the MELD scores are independent predictors of mortality, each may have varied significance due to the maladaptive neurohormonal and physiological responses in ACLD. Rosenstengle and colleagues conducted a large analysis of transplant-listed candidates, stratifying them by dominant biochemical parameter: bilirubin (MELD-Br, n = 13,658), INR (MELD-INR, n = 13,809), and creatinine (MELD-Cr, n = 12,412). Survival was lower for females across each group, emphasising the importance of prospectively validating the MELD 3.0 score, whilst the MELD-Cr had significantly lower 1-year intention-toto-treat rates (p < 0.001) [62]. Creatinine is a variable that may be affected by many patient-specific characteristics including sarcopenia and myosteatosis, which are themselves independent predictors of survival [63]. Lattanzi and colleagues demonstrated in a prospective trial of 249 cirrhotics that radiologically identified sarcopenia and myosteatosis were independently associated with mortality, and that amalgamating them with the MELD score resulted in improved prediction of 3- and 6-month mortality [64]. Meta-analyses have supported the prognostic importance of sarcopenia [65, 66]. The limitations of each variable in the MELD scores including hyponatremia are further discussed in the supplement.

The initial proposed MELD-Na score by Biggins and colleagues prioritised lower sodium cut-off values (120 mmol/L to 135 mmol/L) than the widely accepted score later proposed by Kim and colleagues (125 mmol/L to 140 mmol/L) [37, 67]. The perennial challenge of donor allocation is balancing the dichotomy between transplant list dropout and post-operative mortality. Allocations using the lower values may result in the capture of more vulnerable patients, with hyponatremia conferring increased waitlist mortality; however, it also has post-operative risks, most prevalently the risk of osmotic demyelination syndrome. Dawwas and colleagues conducted a retrospective study of 5150 patients demonstrating that hyponatremia (sodium < 130 mmol/L) was associated with a 55% increased 3-month mortality following OLT [68]. Prioritisation discrepancies between the two approaches need refinement to balance waitlist mortality with transplant complications effectively.

United Kingdom End-Stage Liver Disease (UKELD) score

Around the same time as the development of the MELD-Na score, Barber and colleagues aimed to produce a prognostic score for allocation of liver transplantation in a population specific to the United Kingdom (UK). From a cohort of 1103 patients awaiting liver transplantation at 7 centres across the UK, statistical analysis was conducted using Cox regressions to identify bilirubin, INR, and sodium as independent predictors of survival[69]. Creatinine was empirically included as a variable in the development of a predictive score, which was validated in a prospectively recruited cohort of 452 patients. The derivation study found the UKELD to be a superior predictor of mortality for patients on the transplant list when compared to the MELD and MELD-Na scores. In 2008, the UKELD score was incorporated into the UK guidelines for the allocation of liver transplants, with a score ≥ 49 established as the cut-off for transplant eligibility, as this equated to an equal risk of mortality (Additional File 1: Table S4) [70]. It currently remains an important part of the allocation process in determining survival on the wait list; however, donor allocation has largely been replaced by a more nuanced scheme designed to maximise transplant benefit [71, 72]. The Gender Equity Model for Liver Allocation (GEMA) score was also developed from 5762 patients on the UK transplant registry to enable equitable organ allocation. It was internally validated in 1920 Australian transplant patients and externally in 6071 Spanish patients [52, 73]. Whilst guidelines consider its use, it has not been widely adopted [51].

An ideal score for liver transplantation should allocate patients on the wait list with the goal of minimising mortality and maximising survival post-transplant, the latter of which is not accounted for by the UKELD score [74]. Although the UKELD is employed as a score for stratifying patients for donor allocation, it does not have any prognostic capacity for survival following transplantation [69]. By contrast, the MELD score has been shown to be inversely related to post-transplantation mortality [75], most analysis is low-level evidence [76]. Further investigation is required.

CLIF-C ACLF score

Acute on chronic liver failure (ACLF) is a clinical syndrome characterised by the development of multiorgan failure in the context of liver cirrhosis [77]. It is a unique variant of AD associated with high short-term mortality [9, 77]. The CANONIC study is a prospective, multi-centre European observational study of 1349 AD patients that helped establish the definition of, and ultimately predictive algorithms for, ACLF. Nearly two thirds of the chronic liver failure (CLIF) consortium derived cohort had aetiology relating to alcohol intake [77]. In addition to the presence of AD and high 28-day mortality rate (15%), organ failure measured by sepsis organ failure assessment (SOFA) was required for a diagnosis of ACLF. The SOFA score (comprising liver, kidney, brain, coagulation, circulatory and respiratory system failure) was consensus developed for population-level assessments of mortality in intensive care unit patients but was subsequently accepted at the individual level for evaluation of sepsis [78, 79]. The use of the SOFA score to define outcomes in clinical trials is acceptable as a measure of organ failure, and the authors increased the validity by reporting sub-scores for each type of organ failure [80].

Given the empirical approach to derivation, it was of interest if other predictive values could improve the discriminatory performance of the index SOFA-based clinical score (titled CLIF-OF score, Additional File 1: Table S5). Jalan and colleagues performed a retrospective regression analysis on 275 CANONIC patients with ACLF to incorporate CLIF-OF parameters with additional white cell count (WCC) and age parameters in a newly derived CLIF-C ACLF score. Internal validation in a prospective cohort of 225 French ACLF patients showed superiority to MELD, MELD-Na, and Child–Pugh scores for 28- and 90-day mortality [81]. A retrospective study of 74,790 patients from the Veteran’s health administration in America confirmed the improved discrimination capability of the CLIF-C ACLF score[82]. However, discrepancies between definitions of ACLF persist across geographical regions, with the Asian Pacific Association for Study of the Liver (APASL) including cut-off variables for bilirubin and INR as a requirement for diagnosis [83]. A meta-analysis examining 23 studies from Asian countries (total 26 studies; 3672/4732 patients from Asian countries; 77.60%) suggested that MELD (AUROC 0.82) was superior to the CLIF-SOFA score (AUROC 0.73) [28, 84]. Further, a study of 18,416 UNOS transplant candidates with ACLF found that a combination of MELD score and ACLF grade showed excellent calibration (goodness of fit p = 0.98) and correlation (R = 0.99) for waitlist mortality[85]. A retrospective study of 433 patients using the APASL definition found that a score incorporating sarcopenia and MELD score had an AUROC of 0.86, but no comparison to the CLIF-C ACLF score was made [86]. Establishing a unified definition and further prospective studies are needed. There are currently two prospective, observational studies underway in China (NCT05393453) and India (NCT06069284) designed to identify predictors of mortality in ACLF.

Nearly two thirds of the cohort from the derivation study had an aetiology relating but not limited to alcohol, whilst excessive alcohol intake has been identified as one of the main triggers of ACLF [77]. Systemic inflammation is over-represented in patients with alcoholic ACLF [87]. The CLIF-C ACLF score, and the selection of WCC as a variable, may only be optimal for measuring mortality in patients with alcohol-related disease. However, a subsequent study by Li and colleagues identified the CLIF-C ACLF score maintains its predictive capacity in patients with hepatitis B virus [88]. Further research should be conducted based on etiological sub-groups to confirm that patients with alcoholic ACLF are being fully accounted for.

CLIF-C AD score

Following the success of the CLIF-C ACLF score, the CLIF consortium became interested in developing a score for AD without ACLF. Regression analysis was run on 1016 CANONIC study patients without ACLF, identifying age, creatinine, INR, WCC, and sodium as characteristics associated with mortality at 90, 180, and 365 days [89]. The derivation study externally verified the superiority of the CLIF-C AD in predicting 90-day mortality (AUROC 0.76) compared to the CPS (AUROC 0.65, p < 0.001), MELD (AUROC 0.66, p < 0.001), and MELD-Na (AUROC 0.70, p = 0.024) scores. It is used primarily for predicting survival in hospitalised patients with an acute decompensation of cirrhosis [90]. It has also been shown to be superior to the CPS, MELD, and albumin-vilirubin (ALBI) scores for patients undergoing TIPS insertion, but not the MELD-Na score [91].

The scores used to measure mortality in AD we have previously discussed were developed without excluding patients with ACLF. It is possible that patients with ACLF may have skewed mortality statistics given their high rates of short-term mortality [9, 77]. Furthermore, the CLIF-C AD score was developed measuring survival as the primary outcome in patients being hospitalised for AD. Other scores such as the CPS and MELD score were developed in different contexts and later verified for use in predicting survival. The CLIF-C AD score is the first major score we are aware of that was developed for survival in patients with AD. Unlike the development of previous scores, the authors sub-distributed liver transplantation as a competing risk to mortality during derivation. This enabled the study to adjust for the effect of liver transplantation, therefore replicating real-life situations where a patient’s future chance of transplant may not be known. However, when competing risks are used, the one-to-one correspondence between hazard and cumulative incidence (i.e. between risk and rate) is lost [92]. The variables selected through regression analysis may not be as strongly associated with mortality as the analysis suggests.

The CLIF-C AD score stratifies patients into high (> 60) and low (< 45) risk groups. However, patients with a score between 45 and 60 do not qualify for either risk group. The main advantage of a score that stratifies risk groups is its ability to simply convey prognostic information to patients. In the derivation study, 57% of the participants did not qualify for a risk group, indicating there is a significant proportion of patients who are provided little prognostic information from the CLIF-C AD score. Further research into quantifying the risk of these intermediate patients should be conducted to increase the value of the CLIF-C AD score.

Whilst the MELD score imposes a minimum value for the number of points that can be assigned to a patient’s score when they are on dialysis, the CLIF-C AD score does not take this into consideration. Patients that are on regular haemodialysis therapy may have erroneously low scores due to a low creatinine level, despite a higher disease burden. ACLD is common in dialysis patients, and as such this represents a significant sub-population for whom the CLIF-C AD score may overestimate survival [93].

A similarly novel score for mortality prediction in cirrhosis, the modified Liaoning score, was recently developed in a prospective enrolled cohort of 474 patients in China [94]. Utilising age, ascites, presence of upper GI bleed, platelet count, bilirubin, albumin, and creatinine, the score was validated in three prospective cohorts from Fuzhou, China (AUROC 0.881); Jinan, China (AUROC 0.722); and Brazil (AUROC 0.759) for 2-year survival, with each having numerically higher AUROCs than the MELD and CPS. Prospective studies comparing new predictive scores such as the CLIF-C AD, MELD 3.0, and modified Liaoning scores to contemporary ones such as the MELD and Child–Pugh are indicated.

Albumin-bilirubin (ALBI) score

HCC is a common endpoint for patients with chronic liver disease, with prognosis determined by tumour burden and underlying liver function. The tumour-node-metastases (TNM) staging system is well established in predicting survival in patients based on tumour burden [95], underlying liver function has historically been arbitrarily assessed using the CPS. Johnson and colleagues determined to develop a model that predicted mortality in HCC patients specifically focusing on the impact of liver function on survival [96].

Patients for the study were recruited from major HCC centres in Japan, Hong Kong, Spain, England, and the USA, with all patients having advanced disease by Barcelona Clinic Liver Cancer (BCLC) staging and being treated with sorafenib. Univariate analysis was conducted on the Japanese cohort as this was the largest and most complete database. Independent predictors of mortality were male sex, log₁₀(bilirubin), albumin, tumour size, tumour number, the presence of vascular invasion, and TNM stage. TNM staging and tumour size are more representative of tumour burden rather than underlying liver function. When these were removed from the model, the variables that remained consistently associated with survival were log₁₀(bilirubin), albumin, vascular invasion, and tumour number. The latter two were adjudged to be more correlated with tumour burden than underlying liver function, and as such, log₁₀(bilirubin) and albumin were elected as the variables. Cox regression linear predictors were used to determine three grades of the ALBI score: ≤ − 2.60 (ALBI grade 1), > − 2.60 to ≤ − 1.39 (ALBI grade 2), and > − 1.39 (ALBI grade 3). The study then validated the ALBI score in the cohorts of patients from other centres to show that it retained its discriminatory ability.

In contrast to the specific patient population selected for the derivation study, HCC encompasses a heterogenous population, perhaps weakening the versatility of the ALBI score. It has been validated as a predictor of survival in patients with advanced HCC (Additional File 1: Table S6) [96, 97] and expanded to patients with all levels of BCLC staging and in palliative conditions [98, 99]. Guidelines currently recommend the use of atezolizumab and bevacizumab as first-line therapy for patients with advanced HCC, with a post-HOC analysis of 336 patients from the foundational phase III IMBrave150 study finding progression-free survival was lower for grade 2 patients (5.6 months (95% CI, 5.3, 7.0)) compared to grade 1 patients (8.8 months (95% CI: 6.7, 10.0)) treated with atezolizumab/bevacizumab [15, 100]. Eight studies (6538 patients) were included in a meta-analysis by Mishra and colleagues that found pre-treatment ALBI grade corresponded with mortality outcomes following transarterial chemoemobilisation [101].

A meta-analysis comprising 5 Chinese and 2 Italian cohort studies of predominantly viral HCC with cirrhosis (5377 patients; 90.3–100% Child–Pugh A) found that ALBI grade 2 and 3 patients had a 2.572-fold (95% CI, 1.825 to 3.626, p < 0.001) higher chance of post-hepatectomy liver failure when compared to grade 1 patients [102]. It may be a useful tool to augment risk stratification prior to surgery in Child–Pugh A patients.

The ABLI score utilises two variables in its prediction of mortality risk—albumin and bilirubin. Albumin levels can be influenced by therapeutic interventions such as paracentesis and albumin infusions [103, 104]. With simple interventions, half the variables in the ALBI score may be altered, resulting in significantly altered survival predictions. Continually, there are many independent predictors for survival, such as renal function, which are not accounted for by the ALBI score.

Frieberg Index of Post-TIPS Survival (FIPS) score

Transjugular intrahepatic portosystemic stent (TIPS) is a non-surgical therapy utilised in patients with portosystemic hypertension suffering from recurrent ascites or secondary prophylaxis of variceal bleeding [105]. The MELD score was initially developed for post-TIPS survival whilst the CPS has also been validated in this context [29, 41, 106]. Since the introduction of these scores however, developments such as the expanding polytetrafluoroethylene-covered stent grafts (ePTFE-SGs) has resulted in higher survival rates and fewer complications [107]. With consideration of these differences, a retrospective German cohort of 835 patients was used to examine the evolution in indications for TIPS insertion. When the MELD score was first developed most patients underwent TIPS as an emergency procedure for variceal bleeding, however in contemporary practice the main indication is diuretic resistant ascites [108]. In 2021, Bettinger and colleagues retrospectively developed the Freiberg Index of Post-TIPS Survival (FIPS) from a German cohort of 1871 patients with de novo TIPS insertion, to better characterise short term post-TIPS mortality risk [109]. Multivariate regression determined age, the logarithm of bilirubin, albumin, the inverse of creatinine and pre-procedure history of hepatic encephalopathy to be the most salient prognostic variables. A high and low risk group, stratified by the 85th percentile, showed distinctly different overall survival on Kaplan–Meier analysis. External validation has occurred in large retrospective cohorts from America and China [110, 111]. However, a recent meta-analysis by Zhao and colleagues confirmed the MELD score to be consistently the best predictor of short term survival (AUROC 0.72 for 3 month survival); in particular a prospectively collected Chinese cohort of 855 patients demonstrated the MELD 3.0 to be superior for 3 month mortality (AUROC 0.727 vs 0.582 for FIPS, p = 0.015) [112, 113]. Zhao’s meta-analysis demonstrated the FIPS score had an AUROC of 0.75 for overall transplant free survival; however, the Bettinger’s initial derivation study did not have enough statistical power to determine whether it was predictive of liver transplantation in post-TIPS patients.

The derivation study had no control group assigned to help identify which patients may benefit from TIPS insertion based on the FIPS score. Alternative therapies for recurrent ascites and secondary prophylaxis of variceal bleeding are available to patients, and it is important to differentiate clinically who would benefit from TIPS instead of these. Stockhoff and colleagues performed retrospective propensity score matching on 784 German patients to compare treatment with TIPS (n = 256) and paracentesis (n = 528). In the low-risk FIPS group, there was significantly better survival when treated with TIPS compared to paracentesis (p = 0.037), no difference was apparent in the high-risk group (p = 0.557) [114]. The conclusion drawn by the authors was that high-risk FIPS grouping should not preclude TIPS insertion; however, further analysis into the ability of the FIPS score to provide clinical information beyond prognosis should be pursued.

Given the clinical parameters of bilirubin, albumin, creatinine, and age are all known to be associated with decompensated liver disease, the FIPS score also has since been shown to predict mortality in patients with acute decompensation of ACLD [114, 115]. Inclusion in future meta-analyses would be beneficial.

Conclusions

The use of statistical analysis to assess prognosis has resulted in the development of multiple scores measuring mortality and prognostic outcomes in ACLD. Table 1 represents possible clinical scenarios and the recommended score to use. Alternatives are provided but should be used cautiously. The possible use of each score is represented in Fig. 1. For patients with cirrhosis, the CLIF-C AD and CLIF-C ACLF scores are indicated to predict mortality for their respective clinical scenarios. The MELD 3.0 score is the optimal score for allocation of patients for liver transplant in those patients with cirrhosis and HCC. For patients with diagnosed HCC, the CPS is useful for selecting patients for surgical resection while the ALBI is a non-invasive predictor of survival. The FIPS score may be a useful adjunct for survival estimation following TIPS insertion. Mortality following TIPS insertion is best assessed through the MELD 3.0 score.

Table 1.

Clinical presentations with optimal (bold) and alternative (italics) scores that may be used

Clinical situation for patients with ACLD
Pre-surgical risk assessment Child–Pugh
Stratifying treatment selection in HCC Child–Pugh (as part of BCLC recommendations) ALBI Grade
Mortality following TIPS insertion MELD 3.0 MELD, MELD-Na, FIPS, CLIF-C AD Prioritisation of candidates for liver transplantation MELD 3.0 MELD, MELD-Na, UKELD, GEMA
Mortality following acute decompensation CLIF-C AD Modified Liaoning score, FIPS
Mortality in ACLF CLIF-C ACLF MELD, MELD with ACLF grade
Mortality risk in patients with HCC ALBI Grade Outcomes in alcoholic hepatitis MELD Maddrey’s discriminant function

Fig. 1.

Possible clinical contexts under which each reviewed score may be considered for use

We have outlined the strengths and weaknesses (Table 2) of each scoring system. Although most scores employ the use of independent predictors, they are weakened by the susceptibility of individual variables to alteration. Interactions with other variables and therapeutic interventions all contribute to their malleability. For example, creatinine is influenced by a range of factors, such as age, muscle mass, and dietary protein intake, with sarcopenia being particularly prevalent in ACLD [116, 117]. These factors, particularly sarcopenia and myosteatosis, are not accounted for by current scores. It has been shown that the MELD score underestimates mortality in patients with renal disease and it would not be unreasonable to expect similar results with other scores [118]. Continually, therapeutic interventions such as diuretics or large volume paracentesis can cause short-term variations in creatinine and albumin levels. These fluctuations result in a changed MELD score despite the risk of death being unchanged [119]. Hypoalbuminemia is also associated with high INR [120]. There is a risk of valuing synthetic capacity at double the other parameters in scores such as the CPS [3].

Table 2.

Strengths and weaknesses of each score

	Strengths	Weaknesses
Child–Pugh score	Easy to use and interpret Well established in clinical practice Involves independent predictors of survival in advanced chronic liver disease	Empirical selection of variables Arbitrary selection of cut-off values for categorical groups Subjective elements Inherent malleability of variables with therapeutic intervention
MELD score and variants	Multivariate analysis used for selection of statistically significant prognostic variables • Variable significance weighted based on regression coefficient Independent of aetiology and sequalae of decompensation (ascites, hepatic encephalopathy, acute variceal bleeding)	Continuous scale is harder to interpret Performs worse in specific sub-populations (patients with MAFLD, alcoholic cirrhosis) Patient selection in derivation study favoured more healthy individuals Inherent malleability of variables with therapeutic intervention
UKELD score	Targeted score—developed specifically in cohort of liver transplant patients Multivariate analysis used for selection of statistically significant prognostic variables Prospective external validation in derivation study	Only accounts for liver transplantation waitlist survival Post-transplant survival not captured prognostically by score Hard to interpret and does not provide much valuable information Does not account for patients with HCC Inherent malleability of variables with therapeutic intervention
CLIF-C ACLF score	Targeted score—developed in cohort of patients experiencing ACLF Accounts for independent risks factors derived through statistical analysis Distinct survival outcomes obtained	No statistical analysis was conducted to identify markers of organ failure in the context of ACLF—was based on pre-existing organ failure models (SOFA) Subjective elements Only verified in acute setting—no evidence on chronic outcomes ACLF as a syndrome is still poorly defined Inherent malleability of variables with therapeutic intervention
CLIF-C AD score	Targeted score—first to exclude patients with distinct syndrome of ACLF Multivariate analysis used for selection of statistically significant prognostic variables Sub-distribution to account for patients who undergo liver transplantation Distinct survival outcomes derived Completely objective score	• Does not account for dialysis patients unlike other scores (MELD, MELD-Na) Large proportion of patients not accounted for by thresholds set for high and low risk groups • Selection of cut-offs not statistically based Not validated in cohorts containing combination of AD and ACLF patients • On initial presentation these can be difficult to distinguish Inherent malleability of variables with therapeutic intervention
ALBI score	Targeted score—developed specifically for advanced stage HCC patients Simple and objective score Distinct survival outcomes derived	Potentially may be overly simplistic—does not account for many known predictors of mortality Developed in cohort of patients with advanced HCC treated with sorafenib • Excludes intermediate/early-stage HCC and those undergoing more recent novel treatment regimens Inherent malleability of variables with therapeutic intervention
FIPS score	Targeted score—developed specifically for patients considered for TIPS insertion Multivariable analysis used for selection of statistically significant prognostic variables Completely objective score	Does not account for patients with HCC Derivation cohort primarily consisted of patients with alcohol associated cirrhosis • May be less predictive for those with more metabolic risk factors No control group in derivation study Inherent malleability of variables with therapeutic intervention

Other weaknesses to consider include the empirical selection of variables and subjectivity of measurement in scores such as the CPS and CLIF-C ACLF. Being continuous scores in nature, the MELD and MELD-Na scores can be difficult to communicate survival outcomes to patients. Furthermore, the UKELD score is hard to interpret and does not account for post-transplant survival [74]. The CLIF-C AD score fails to account for a large proportion of patients, assigning them to an arbitrary intermediate risk group. Finally, there is still ambiguity in the definitions associated with ACLF. For further discussion regarding the weaknesses of scores, including the issues with designs of derivation cohorts and real-world interactions between variables, see the Additional File 1.

Table 3 summarises some key areas of research whilst acknowledging ongoing studies we have discussed in this narrative review. There are many factors known to be predictive of mortality that are not yet incorporated into predictive scores. There is also limited external validation of novel scores such as the MELD 3.0 and CLIF-C AD scores. In the setting of liver transplantation, it would be beneficial to establish if gender disparities are addressed by the MELD 3.0 with prospective cohorts and meta-analyses. Unifying the definitions of ACLF is important in ensuring the CLIF-C ACLF score can be used optimally in clinical practice. Prospective trials validating predictive factors are currently being undertaken. Finally, as our understanding of the complexities of metabolic dysfunction associated with liver disease increases, identification of previously excluded patients with disease is occurring [7]. Scores developed specifically for this disease are needed to account for a larger, heterogeneous group of patients.

Table 3.

Suggestions for future research with current areas of ongoing research

Future areas for research
Addressing the absence of known predictive factors in currently used score
Sarcopenia, myosteatosis and gender are all predictive factors that are under accounted for
A Cochrane protocol for a meta-analysis evaluating the predictive capacity of current scores is available55
Scores developed over recent years such as the MELD 3.0, CLIF-C AD and modified Liaoning scores should be included in updated meta-analyses
The ongoing under prioritisation of females for liver transplantation
Prospective studies evaluating new allocation scores such as the MELD 3.0 and GEMA score are needed
Novel predictors of outcomes in patients with ACLF
The derivation study of the CLIF-C ACLF score recruited a cohort with aetiology largely related to alcohol misuse. It would be of use to determine if different causes of ACLF are as effectively captured
There are two current prospective studies registered investigating predictive factors - NCT05393453 and NCT06069284
Improved predictive models for mortality following TIPS insertion
There are two meta-analyses currently registered on PROSPERO by Chi et al. (CRD42024528193) and Huang et al. (CRD420251007253) that will provide further prognostic information on each score
With the novel inclusive diagnostic criteria for metabolic dysfunction associated liver disease increased granularity on outcomes for these patients needs to be sought
The BOMASH study (NCT06813508) has been designed to investigate outcomes in patients with MAFLD
Incorporating the use of machine learning and AI models into predictive algorithms

Current prognostic models in advanced liver disease rely on predictive analytics; the use of historical data to model future events. Prescriptive analytics takes predictive analyses and applies them to a continually evolving model to better predict outcomes [121]. The use of artificial intelligence through neural networks and machine learning in personalising survival modelling in patients with liver disease is of future interest [122]. The use of machine learning based on dynamic changes is highly effective in patients admitted to the intensive care unit [123, 124]. Future studies investigating the use of machine learning and neural networks in predicting mortality and other relevant outcomes are warranted. The evolving heterogeneity of our understanding of liver disease may be best addressed by the personalised medicine AI promises.

Supplementary information.

Additional File 1: Table S1-6. Table S1—Mortality rates for CPS Classes. Table S2—3-month mortality in patients undergoing liver transplantation by MELD score. Table S3—3-month mortality in patients undergoing liver transplant by MELD-Na Score. Table S4—1-year mortality without transplantation as per the UKELD Score. Table S5—CLIF-OF Score Rubric. Table S6—Median survival based on ALBI score.

Abbreviations

ACLD

Advanced chronic liver disease

ACLF

Acute-on-chronic liver failure

AD

Acute decompensation

ALBI

Albumin-bilirubin score

APASL

Asian Pacific Association for Study of the Liver

AUROC

Area under the receiver operating characteristic curve

BCLC

Barcelona Clinic Liver Cancer

CLIF-C

Chronic Liver Failure Consortium

CPS

Child-Pugh score

ePTFE-SG

Expanding PolyTetraFluoroEthylene-covered tent Graft

FIPS

Frieberg Index of Post-TIPS Survival

GEMA

Gender Equity Model for Liver Allocation

GI

Gastrointestinal

HBV

Hepatitis B virus

HCC

Hepatocellular carcinoma

HCV

Hepatitis C virus

INR

International normalised ratio

MAFLD

Metabolic dysfunction-associated fatty liver disease

MELD

Model for End-Stage Liver Disease

OLT

Orthotopic liver transplant

SOFA

Sepsis Organ Failure Assessment

TACE

Transarterial chemoembolisation

TIPS

Transjugular intrahepatic portosystemic shunt

TNM

Tumour lymph node metastasis

UK

United Kingdom

UKELD

United Kingdom End-Stage Liver Disease

WCC

White cell count

Authors'contribution

OM, WM and GA review conception; OM first draft composition; WM, SR, JG, GA substantial revisions and editing; all authors final review.

Funding

This work was supported by the Ainsworth bequest to Western Sydney University.

Availability of data and materials

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

N/A.

Consent for publication

N/A.

Competing interests

The authors declare no competing interests.