Introduction
The updated global burden of disease data indicates that the prevalence of cirrhosis increased considerably by 74.5% from 1990 to 2017, with East Asia being one of the regions encompassing the highest number of cases.1 Notably, cirrhosis (and related pathological disorders) has been reported to account for 1.32 million deaths (2.4% of total deaths worldwide).2 It is crucial to develop and validate mortality scoring systems that consider different clinical scenarios, including organ allocation, health resource utilization, prognostication, and patient triage optimization. Among these, accurate prediction of mortality in advanced chronic liver diseases (e.g., cirrhosis) is of utmost clinical relevance and, to some extent, serves as the cornerstone for instigating and refining further therapeutics.3 Accordingly, a range of tools has been proposed and implemented, with the Child-Turcotte-Pugh (CTP) class and model for end-stage liver disease (MELD) being widely applied and validated.
The original MELD score exhibits intrinsic drawbacks, such as a gradually declining accuracy for predicting mortality in certain circumstances, including acute-on-chronic liver failure, hepatocellular carcinoma, and cirrhosis.4–6 Additionally, marked changes in demographics, epidemiology, the inclusion of elderly patients, comorbidities, and evolving treatment strategies for various liver diseases may result in insufficient predictive power of MELD.7–9 In this context, significant efforts have been made to improve prognostic performance by modifying various MELD-based variants and developing other prognostic scores to address sex disparities. More recently, Kim and colleagues substantially adjusted the MELD/MELD-Na scores to generate an iteration designated as MELD 3.0, based on 20,587 liver transplant candidates, showing improved discrimination for 90-day mortality.10 Along similar lines, Rodríguez-Perálvarez et al. derived a novel prognostic score by introducing a more reliable estimation of renal function—namely, the gender-equity model for liver allocation (GEMA), which has been subsequently verified with improved discrimination and reclassification benefit related to 90-day mortality/dropout from the transplantation list in a sizable cohort of 9,320 patients.11 Intriguingly, external validations across different regions and populations have been rapidly conducted in the context of versatile clinical scenarios with advanced hepatopathy. For instance, serial investigations from Spain and Greece have provided strong evidence supporting the better performance and more accurate prediction of GEMA-based scores over MELD family scores.12,13 However, conflicting findings also need to be addressed: Yoo and colleagues reported that MELD 3.0 (with albumin) outperformed the original MELD regarding short-term waitlist survival in East Asia, but this was not the case for MELD-Na; Marrone et al. argued that GEMA-Na exhibited higher accuracy for 90-day dropout prediction relative to MELD, while no superiority was observed with respect to MELD-Na.14,15 Taken together, the generalizability and validity of the aforementioned new prognostic values merit comprehensive investigation, as these upgrades and iterations have been substantially refined in contrast to minor improvements in existing models.16,17 We hypothesized that the implementation of newly developed scores, such as MELD 3.0 or GEMA-Na, would improve discriminative ability in generalized cirrhosis. The main purposes of this study are to: 1) delineate the discriminative and stratified abilities of these novel scores compared to MELD/MELD-Na scores; 2) validate their prognostic utility regarding all-cause mortality at different time points among hospitalized patients with cirrhosis; and 3) verify the main findings in another cirrhotic cohort presenting clinical disparities.
Methods
Study population
The participants in this retrospective study included all patients with cirrhosis hospitalized due to decompensated episodes (i.e., gastroesophageal variceal bleeding, severe jaundice, ascites, hepatic encephalopathy, and infection) from 2018 to 2021 at a single center: the Department of Gastroenterology and Hepatology, Tianjin Medical University General Hospital. A second cohort for validation was recruited from cirrhotic patients experiencing decompensation at a tertiary hospital, the Third People’s Hospital of Chengdu. The definitions were as follows: gastroesophageal variceal bleeding determined by endoscopy, severe jaundice with total bilirubin ≥ 51 µmol/L, ascites clinically evident and documented via ultrasound/physical examination according to the International Ascites Club classification, and hepatic encephalopathy evaluated using the West Haven Criteria (I-IV).18–21 The inclusion criteria were: 1) adult patients aged ≥18 years, and 2) definitive diagnosis of cirrhosis based on a combination of medical history, biochemical examination, endoscopic/elastography findings, and liver biopsy results if available. The exclusion criteria were: 1) concurrent acute-on-chronic liver failure, 2) hepatocellular carcinoma or other extrahepatic malignancies, 3) liver transplantation within 30 days, and 4) refusal to follow up (Supplementary Fig. 1). This study conformed to the Declaration of Helsinki and was approved by the ethics committees of Tianjin Medical University General Hospital and the Third People’s Hospital of Chengdu. Written informed consent was obtained from all participants prior to their involvement.
Evaluation
All scores were calculated upon initial admission.
MELD (original) = 9.57 × ln(creatinine) + 3.78 × ln(bilirubin) + 11.20 × ln(INR) + 6.43.
MELD-Na = MELD (original) + [1.32 × (137 - Na)] - [0.033 × MELD (original) × (137 - Na)].
MELD 3.0 = 1.33 (if female) + [4.56 × ln(bilirubin)] + [0.82 × (137 - Na)] - [0.24 × (137 - Na) × ln(bilirubin)] + [9.09 × ln(INR)] + [11.14 × ln(creatinine)] + [1.85 × (3.5 - albumin)] - [1.83 × (3.5 - albumin) × ln(creatinine)] + 6.
MELD 3.0 (no albumin) = 1.40 (if female) + [4.85 × ln(bilirubin)] + [0.88 × (137 - Na)] - [0.25 × (137 - Na) × ln(bilirubin)] + [9.66 × ln(INR)] + [10.47 × ln(creatinine)] + 6.
Royal Free Hospital Glomerular Filtration Rate (RFH-GFR) = 45.9 × (creatinine−0.836) × (urea−0.229) × (INR−0.113) × (age−0.129) × (sodium0.972) × 0.809 (if female) × 0.92 (if moderate or severe ascites).
GEMA = 3.777 × ln(bilirubin) + 7.883 × ln(INR) – 8.306 × ln(RFH-GFR) + 31.932.
GEMA-Na = GEMA − Na − [0.025 × GEMA × (140 − Na)] + 140.
Outcomes of interest
The primary outcome of this study was to compare the discriminative ability across serial time points assessed at 30 days, 90 days, 180 days, one year, and two years for all-cause mortality based on the 4 MELD-based scores (i.e., original MELD, MELD-Na, MELD 3.0, and MELD 3.0 (no albumin)) and GEMA-Na. The secondary outcome was to determine the optimal threshold for classifying patients into high vs. low mortality risk groups during the two-year follow-up.
Statistical analysis
All continuous data were presented as median ± interquartile range, whereas categorical data were presented as numbers and percentages. The Mann-Whitney non-parametric test was used to detect differences between the groups. The Pearson chi-squared test or Fisher’s exact test was applied to compare proportions between groups. Time-dependent receiver operating characteristic (ROC) analysis and the area under the ROC curve (AUC) were used to compare the predictive power of the prognostic scores. To determine the optimal cutoff for two-year all-cause transplant-free mortality in our study population, two statistical approaches were used: minimally statistical risk analysis and the X-tile project, considering their validity for time-to-event analyses. Univariate and multivariate Cox proportional hazards models were used to determine hazard ratios (HRs) and 95% confidence intervals (CIs) for independent predictors of mortality. Time-to-event data were expressed as Kaplan-Meier curves, and survival conditions were compared using log-rank tests. P < 0.05 was considered statistically significant. Statistical analyses were performed using SPSS (version 23.0, IBM) and R software (version 4.0.2, R Foundation for Statistical Computing).
Results
Baseline characteristics of the study population
The baseline demographics, clinical, and biochemical features of the entire population are presented in Table 1. The median age was 63 (57, 70) years, and 51.0% (158/310) were female. The most common admitted complications were ascites (179/310, 57.7%) and gastroesophageal variceal bleeding (168/310, 54.2%), followed by infection and the onset of hepatic encephalopathy in 12.3% and 8.7% of all participants, respectively. Regarding etiologies, alcohol consumption and chronic viral infection were the two most common causes of liver cirrhosis, each with a prevalence of 24.5%. Hospitalized patients with cirrhosis were categorized into CTP class A/B/C in 99/160/51 subjects. For MELD- and GEMA-based scores, the median MELD, MELD-Na, MELD 3.0, MELD 3.0 (no albumin), and GEMA-Na were 9 (7, 12), 9 (7, 12), 12 (10, 17), 11 (9, 15), and 12 (9, 17), respectively. Upon admission, the median total bilirubin was 21.6 (14.1, 38.5) µmol/L; median creatinine, 59 (48, 72) µmol/L; serum albumin, 28 (25, 32) g/L; and sodium, 140 (138, 142) mmol/L. During the two-year follow-up, a total of 79 patients died, resulting in an overall survival rate of 74.5%. The main causes of death were organ failure (39/79, 49.4%), followed by severe infection/sepsis (20/79, 25.3%), gastrointestinal bleeding (9/79, 11.4%), hepatic encephalopathy (6/79, 7.6%), and other causes. When stratified by survival status, non-survivors had significantly higher levels of total bilirubin, creatinine, white blood cell count, and neutrophil-to-lymphocyte ratio (NLR), but lower levels of hemoglobin, sodium, and albumin. The proportions of patients with combined CTP class B&C and presenting with hepatic encephalopathy, infections, and ascites were higher in the non-survivor group. The cirrhosis etiology attributable to alcohol consumption was significantly higher in deceased patients than in survivors.
Table 1Baseline characteristics of the hospitalized patients with cirrhosis in the investigated cohort
| Variables | Total patients (N = 310) | Survivors (N = 231) | Non-survivors (N = 79) | P-value |
|---|
| Ages (years) | 63 (57, 70) | 63 (57, 69) | 63 (57, 71) | 0.592 |
| Sex (%) | | | | >0.999 |
| Male | 152 (49.0) | 113 (48.9) | 39 (49.4) | |
| Female | 158 (51.0) | 118 (51.1) | 40 (50.6) | |
| Hemoglobin (g/L) | 89 (72, 112.3) | 91 (74, 116) | 67 (59, 77) | <0.001 |
| BMI (kg/m2) | 24.2 (20.8, 27.2) | 24.3 (21.3, 27.1) | 23.1 (19.2, 27.6) | 0.258 |
| Bilirubin (µmol/L) | 21.6 (14.1, 38.5) | 18.9 (12.9, 32.6) | 35.3 (20.2, 103.5) | <0.001 |
| WBC (×109/L) | 3.66 (2.50, 5.20) | 3.46 (2.27, 4.76) | 4.51 (3.16, 6.99) | <0.001 |
| Sodium (mmol/L) | 140 (138, 142) | 141 (138, 142) | 138 (134, 141) | <0.001 |
| Potassium (mmol/L) | 3.9 (3.6, 4.1) | 3.9 (3.6, 4.2) | 3.9 (3.5, 4.1) | 0.325 |
| Platelets (×109/L) | 84 (53, 117) | 82 (51, 115) | 94 (58, 134) | 0.169 |
| Albumin (g/L) | 28 (25, 32) | 30 (27, 34) | 25 (22, 28) | <0.001 |
| Urea (mmol/L) | 5.1 (3.8, 6.8) | 4.8 (3.4, 5.9) | 6.2 (5.2, 9.9) | <0.001 |
| Creatinine (µmol/L) | 59 (48, 72) | 57 (46, 69) | 64 (54, 86) | <0.001 |
| NLR | 3.09 (1.92, 4.89) | 2.68 (1.75, 4.18) | 4.59 (2.99, 9.16) | <0.001 |
| CTP score | 9 (8, 10.5) | 7 (6, 8) | 9 (7, 10) | <0.001 |
| CTP class (%) | | | | <0.001 |
| A | 99 (31.9) | 91 (39.4) | 8 (10.1) | |
| B | 160 (51.6) | 115 (49.8) | 45 (57.0) | |
| C | 51 (16.5) | 25 (10.8) | 26 (32.9) | |
| Complication (%) | | | | |
| GEVB† | 168 (54.2) | 160 (69.3) | 48 (60.8) | 0.165 |
| HE | 27 (8.7) | 14 (6.1) | 13 (16.5) | 0.005 |
| Infection | 38 (12.3) | 21 (9.1) | 17 (21.5) | 0.004 |
| Ascites | 179 (57.7) | 120 (51.9) | 59 (74.7) | <0.001 |
| Etiology (%) | | | | 0.004 |
| Alcohol | 76 (24.5) | 46 (19.9) | 30 (38.0) | 0.001 |
| Viral | 76 (24.5) | 63 (27.3) | 13 (16.4) | 0.054 |
| AILD | 56 (18.1) | 42 (18.2) | 12 (15.2) | 0.609 |
| Cholestasis | 30 (9.7) | 23 (10.0) | 9 (11.4) | 0.675 |
| NAFLD/Cryptogenic | 72 (23.2) | 57 (24.6) | 15 (19.0) | 0.301 |
| MELD score | 9 (7, 12) | 9 (6, 11) | 11 (8, 17) | <0.001 |
| MELD-Na score | 9 (7, 12) | 9 (6, 11) | 12 (8, 18) | <0.001 |
| MELD 3.0 (no alb) score | 11 (9, 15) | 10 (9, 13) | 18 (11, 23) | <0.001 |
| MELD 3.0 score | 12 (10, 17) | 11 (9, 14) | 19 (12, 24) | <0.001 |
| GENA-Na score | 12 (9, 17) | 10 (8, 14) | 18 (13, 24) | <0.001 |
Stratified performance of MELD 3.0/GEMA-Na compared with other MELD-based scores
As an exploratory analysis, we demonstrated the stratification of inpatients at two years between the iterative MELD 3.0 and other MELD-based scores. As depicted in Supplementary Table 1, MELD-Na exhibited increased scores in 87 (45.1%), 23 (33.3%), 15 (62.5%), and zero (0%) patients and decreased scores in zero (0%), two (2.9%), zero (0%), and one (4.2%) patients, with 106 (54.9%), 44 (63.8%), nine (37.5%), and 23 (95.8%) patients showing no shifts, respectively, among the MELD-Na ≤ 10, 10 < MELD-Na ≤ 15, 15 < MELD-Na ≤ 20, and MELD-Na > 20 strata compared with MELD 3.0. Of the 79 non-survivors, increased scores were observed in 39 (49.4%) patients, and the original strata were retained in 40 (50.6%), while no patients had decreased scores. The most notable shift was found among participants with strata of MELD-Na ≤ 10. For a clearer illustration, two separate Sankey diagrams were constructed for the entire population (Fig. 1A) and for deceased patients with cirrhosis (Fig. 1B). On the other hand, the transition from MELD-Na to MELD 3.0 and GEMA-Na, respectively, resulted in a score change of ≥2 points in 196 and 194 patients. The former pattern included 55.5% upgraded and 7.4% downgraded, while the latter pattern included 45.2% upgraded and 17.4% downgraded (Supplementary Fig. 2). Accordingly, MELD 3.0 identified more women (63.6% vs 51%, P < 0.001) and more patients with ascites (66.4% vs 57.7%, P < 0.001) than MELD-Na. GEMA-Na identified more women (59.9% vs 51%, P < 0.001), but a reduced proportion of patients with ascites (42.4% vs 57.7%, P < 0.001) compared with MELD-Na. Additionally, the shift between MELD 3.0 and GEMA-Na/original MELD score is provided in the supplementary files (Supplementary Table 2 and Supplementary Fig. 3).
Prognostication of MELD 3.0/GEMA-Na in hospitalized patients with cirrhosis
To comprehensively evaluate the prognostication of novel prognostic scores, we applied a time-dependent ROC curve to better understand their predictive utility for all-cause mortality. As presented in Figure 2, the MELD 3.0 curve showed a reverse L-shaped relationship with the time course, indicating a steep increase in AUCs at approximately 90 days, followed by a temporally steady relationship between 180 days and two years in predicting all-cause mortality among inpatients with cirrhosis. The model discrimination, assessed by time-dependent AUC, was best for MELD 3.0 and worst for MELD, with MELD-Na and MELD 3.0 (no albumin) showing intermediate results. A similar pattern was observed for GEMA-Na, which reached the best discrimination over other models at around 180 days, without significant fluctuations during the two-year period.
Considering the primary objective of estimating long-term prognostication, comparisons between diverse scoring systems concerning one- and two-year all-cause mortality were made. Figure 3 indicated that MELD 3.0 outperformed MELD-Na, MELD, and MELD 3.0 (no albumin) scores in predicting two-year mortality, as well as one-year mortality. Notably, our preliminary results also indicated that GEMA-Na slightly outperformed MELD 3.0 for predicting two-year all-cause mortality (0.818 vs 0.783, P = 0.038), whereas the predictive power was comparably effective for one-year mortality (0.810 vs 0.786, P = 0.170). Additionally, comparisons at other time points (i.e., 30 days, 90 days, and 180 days) can be found in Supplementary Figure 4.
Determination of optimal cutoff for classifying two-year mortality
Next, we sought to define the best threshold to categorize patients into high vs. low mortality risk groups based on different statistical approaches. Intriguingly, the maximally selected rank statistics and X-tile project consistently pinpointed MELD 3.0 with a cutoff of 18 as the most optimal threshold for categorizing the recruited patients into two groups (Fig. 4A and Supplementary Fig. 5A). Accordingly, patients with a MELD 3.0 > 18 exhibited poor survival outcomes compared to those with a MELD 3.0 ≤ 18 (log-rank test: P < 0.001) (Fig. 4B). Subgroup analysis considering sex differences further demonstrated that MELD 3.0 > 18 was a robust threshold for prognostication among hospitalized cirrhotic patients of both sexes (Fig. 4C & D). For GEMA-Na, the optimal cutoff was identified as 20 by the X-tile project. Patients with a GEMA-Na > 20 also demonstrated poor survival conditions (Supplementary Fig. 6A).
Association between categorized MELD 3.0/GEMA-Na and two-year mortality
Next, we performed a Cox regression model to determine the relationship between a range of demographic, clinical, and biochemical parameters, MELD 3.0/GEMA-Na, and two-year all-cause mortality. Univariate analysis suggested that total bilirubin, white blood cell count, sodium, albumin, creatinine, NLR, CTP class, onset of hepatic encephalopathy, ascites, cirrhosis etiology attributable to alcohol, and all prognostic scores were associated with mortality (Supplementary Table 3). Considering both statistical significance and avoiding parameter collinearity, multivariate Cox regression showed that both GEMA-Na (HR: 1.12, 95% CI: 1.10, 1.17, P < 0.001) and MELD 3.0 (HR: 1.13, 95% CI: 1.10, 1.17, P < 0.001) remained independent risk factors for two-year mortality after adjusting for the onset of hepatic encephalopathy, NLR, ascites, and cirrhosis etiology attributable to alcohol (Tables 2 and 3).
Table 2Multivariate Cox analysis concerning two-year all-cause mortality among hospitalized patients with cirrhosis incorporating MELD 3.0
| Variables | HR (95% CI) | P-value |
|---|
| HE | 1.73 (0.91–3.29) | 0.090 |
| NLR | 1.00 (0.99–1.00) | 0.438 |
| Ascites | 1.52 (0.90–2.59) | 0.121 |
| Alcohol | 1.04 (0.62–1.72) | 0.893 |
| MELD 3.0 score | 1.13 (1.10–1.17) | <0.001 |
Table 3Multivariate Cox analysis concerning two-year all-cause mortality among hospitalized patients with cirrhosis incorporating GEMA-Na
| Variables | HR (95% CI) | P-value |
|---|
| HE | 1.66 (0.89–3.09) | 0.107 |
| NLR | 1.00 (0.99–1.01) | 0.496 |
| Ascites | 1.45 (0.83–2.46) | 0.121 |
| Alcohol | 1.01 (0.61–1.68) | 0.965 |
| GEMA-Na score | 1.12 (1.10–1.17) | <0.001 |
Validation in another cirrhotic cohort
To confirm our findings, an independent cohort enrolling 120 patients from another tertiary hospital was analyzed (median age 67 years, 52.5% female, Supplementary Table 4). Accordingly, both survival analysis (Supplementary Figs. 5B and 6B) and multivariate Cox regression (Supplementary Tables 5 and 6) indicated that GEMA-Na/MELD 3.0, along with their corresponding cutoffs, were closely associated with long-term prognosis in the context of decompensated cirrhosis (GEMA-Na: HR: 4.71, 95% CI: 2.35, 9.44, P < 0.001; MELD 3.0: HR: 3.27, 95% CI: 1.74, 6.16, P < 0.001).
Discussion
In this work, we comprehensively assessed the discrimination and predictive ability of two novel prognostic scores, MELD 3.0 and GEMA-Na, in predicting all-cause mortality among hospitalized patients with cirrhosis. Compared to their predecessors, both models consistently demonstrated strong discriminative ability, particularly in relation to long-term mortality, without a dramatic decline, as reflected by smooth curves in time-dependent ROC analyses. This effect was more pronounced for GEMA-Na, which was the only score exhibiting an AUC greater than 0.8 up to two years. Furthermore, the stratified performance, prognostication, and independent relationship with transplant-free mortality have been elucidated in the present study. Validation in another cohort suggested that the proposed cutoffs could capture an increased likelihood of death in the context of decompensated cirrhosis, paving the way for their implementation in clinical practice.
The construction and introduction of novel prognostic scores can be attributed to the ever-changing epidemiology of cirrhosis and therapeutic advances in recent decades. In fact, Godfrey and colleagues reported that the prediction accuracy of MELD and MELD-Na represented a temporal decline in predicting short-term mortality, a major outcome in organ allocation systems.7 In that paper, it was highlighted that the etiology-related differences accounted for the decreased relationship between MELD/MELD-Na scores and mortality. Contrary to the sufficient discrimination observed in the context of HCV or cholestatic liver diseases for liver transplantation, the prognostic uncertainty of MELD/MELD-Na scores was notably addressed in populations with alcoholic liver disease and NASH-related cirrhosis. Intriguingly, these findings align with our results, where the major cirrhosis etiologies shifted to alcohol and, to a lesser extent, NAFLD. NAFLD is characterized by the disruption/dysregulation of multiple system, accompanied by comorbidities such as hypertension, dyslipidemia, diabetes, and chronic kidney disease. These cannot be easily captured or graded solely through hepatic function parameters. Additionally, patients with aggressive alcohol-related liver disease are more likely to have greater transplant priorities and better post-transplant recovery due to a lower chronic disease burden, resulting in overestimation and low concordance when using MELD/MELD-Na to predict outcomes against true survival status. Therefore, the improvement in discriminative accuracy of MELD 3.0 and GEMA-Na in this study highlights several advantages. For instance, MELD 3.0 reduces the impact of creatinine, as abnormal creatinine levels often serve as surrogates for chronic nephropathy in patients with NAFLD rather than indicators of acute kidney damage, which is more clinically relevant in daily practice.10,22 On the other hand, GEMA-Na applied RFH-GFR, a cirrhosis-dictated mathematical metric, to replace serum creatinine, aiming to eliminate creatinine-triggered bias.23 Additionally, the refined strategy of reweighting INR and bilirubin variables in developing GEMA-Na also contributed to its improved performance.16
As demonstrated in the present study, GEMA-Na, which was derived from 9,320 candidates in the UK, exhibited improved discrimination and reclassification to predict 90-day mortality or delisting, with a more pronounced benefit in females, compared to several MELD-based scores.11 Notably, applying GEMA-Na instead of MELD-Na could avoid one in eight deaths in female patients. This is reflected in our investigation, where GEMA-Na showed significantly improved discrimination over MELD 3.0 for predicting two-year all-cause mortality (AUC: 0.807 vs 0.779, P = 0.038).11 Another study conducted by Sealock and colleagues addressed pervasive sex differences in MELD-Na components and proposed a simulated sex-adjusted MELD-Na score that could increase the rate of female transplantation while modestly decreasing overall mortality rates.24 Most recently, in a unique population of Italian patients with elderly donors and recipients, Marrone and colleagues confirmed that these sex-adjusted scores demonstrated a higher discriminative ability compared to MELD.15 In that study, GEMA-Na showed the best performance in predicting 90-day dropout from the list, with a 4.4% improvement in prediction, potentially avoiding one in nine dropouts.
Another topic of this work was to elucidate the potential role of MELD 3.0 in long-term prognostication. Surprisingly, we observed that this new score represented an increase in predictive power at 90 days, although it slightly dropped to a lesser extent, maintaining approximately 0.780 in time-dependent AUC analysis. Furthermore, the discrimination concerning MELD 3.0 outperformed all other MELD-based scores across a spectrum of time points, extending up to two years. As stated in the original manuscript, MELD 3.0 updates its iteration by deliberately considering the interaction terms between different parameters, rather than simply adding female sex or serum albumin variables. A recent study in patients with alcohol-associated hepatitis (71% with underlying cirrhosis) revealed that MELD 3.0 served as the best predictor of dialysis and was superior to MELD-Na in identifying subjects with an increased likelihood of death at 90 days.25 Similarly, Yoo and colleagues corroborated that MELD 3.0 with albumin showed better predictive power regarding short-term prognosis in Eastern Asian populations on the liver transplantation waitlist.14 For predicting mortality in Chinese patients with cirrhosis, a population with a mean MELD 3.0 of 11.0 comparable to our cohort, Song and colleagues also confirmed the robust impact of MELD 3.0 on short-, medium-, and long-term prognosis, indicating its considerable superiority in prognostic value.26 However, that study concentrated on a specific group of cirrhotic patients receiving transjugular intrahepatic portosystemic shunts, so there is still a knowledge gap in generalized inpatient populations experiencing decompensating episodes. Accordingly, to some extent, we clarified the clinical implementation and expanded the scope of MELD 3.0 to more common scenarios. However, critics may raise concerns about the unexpected steady performance of MELD 3.0 in prognostication in the present study. We believe that the beneficial features and additional merits of MELD 3.0 compared to other MELD-based scores strongly reflect the pathophysiological complexity and perturbations associated with various complications during the advancement/progression of cirrhosis, which have been accurately captured in the rescaled MELD 3.0 version. Notably, it has been suggested that MELD 3.0 is closely linked to malnutrition and other pathological entities associated with metabolic disruption, such as sarcopenia and frailty. Our prior report clearly showed that these debilitating conditions could lead to inferior outcomes, and their clustering in a fraction of inpatients with cirrhosis underpins synergistically detrimental effects.27 Taken together, we believe that all the tailored scores mentioned above are clinically meaningful in diverse scenarios, offering prognostic ability beyond prioritization status.11,15,24
This work does have some limitations. First, its retrospective nature may have introduced bias and uncertain causality. For instance, the study population consisted of hospitalized patients due to decompensating events, which limits its generalizability to the broader waitlist population. Moreover, we excluded subjects experiencing acute-on-chronic liver failure to avoid survival rate fluctuations, given its high short-term mortality.28 However, this exclusion criterion was consistent with contextual studies.11,15 Second, the sample size was determined by the study period and the number of admissions, rather than being based on a formal power calculation. Third, although we illustrated a steady prognostic performance of both MELD 3.0 and GEMA-Na across multiple time points, their discriminative accuracy was suboptimal, without an AUC ≥ 0.9. Indeed, other indices, such as nutritional status and frailty, which have been shown to be detrimental or contributory pathogenic drivers, may improve the predictive performance of the latest iterations. Fourth, muscular abnormalities were not included in the final analyses, which significantly impacted serum creatinine levels and entailed sex disparities, as not all participants had undergone radiological examinations. Lastly, there was a lack of dynamic changes or longitudinal data regarding MELD- and GEMA-based scores to evaluate their discriminative accuracy in relation to inferior outcomes in our study cohort. In fact, one outstanding article implicates inherent weaknesses in these models, with a gradual decline in their accuracies due to epidemiological and demographic changes.3 Furthermore, specific conditions, such as hepatic malignancy, and detrimental complications like spontaneous bacterial peritonitis and hepatorenal syndrome, also negatively impact the usefulness of these scores. It should be kept in mind that the calculation of the preferred models should be updated periodically, particularly whenever the patients’ clinical status markedly changes. Taken together, further prospective studies with sizable populations are warranted to validate our preliminary findings.
Supporting information
Supplementary Fig. 1
A flowchart of the study population.
ACLF, acute-on-chronic liver failure.
(TIF)
Supplementary Fig. 2
Transition from MELD-Na to MELD 3.0 and GEMA-Na.
MELD, model for end-stage liver disease; GEMA-Na, gender-equity model for liver allocation-sodium.
(TIF)
Supplementary Fig. 3
Sankey diagrams to directly visualize and quantify the shifts in the numbers of patients between MELD 3.0 and GEMA-Na/original MELD score.
MELD, model for end-stage liver disease; GEMA-Na, gender-equity model for liver allocation-sodium.
(TIF)
Supplementary Fig. 4
Comparison of GEMA-Na/MELD 3.0 and other scores according to the area under the receiver operating characteristic curves in predicting all-cause mortality among hospitalized patients with cirrhosis.
MELD, model for end-stage liver disease; GEMA-Na, gender-equity model for liver allocation-sodium.
(TIF)
Supplementary Fig. 5
X-tile project to identify a cutoff of MELD 3.0 of 18 and survival analysis in the validation cohort.
MELD, model for end-stage liver disease.
(TIF)
Supplementary Fig. 6
X-tile project to identify a cutoff of GEMA-Na of 20 and survival analysis in the validation cohort.
GEMA-Na, gender-equity model for liver allocation-sodium.
(TIF)
Supplementary Table 1
Comparison of MELD-Na and MELD 3.0 in the entire population and deceased patients with cirrhosis.
(DOCX)
Supplementary Table 2
Comparison of MELD and MELD 3.0 in the entire population and deceased patients with cirrhosis.
(DOCX)
Supplementary Table 3
Univariate Cox regression analysis of independent risk factors for 2-year all-cause transplant-free mortality.
(DOCX)
Supplementary Table 4
Baseline characteristics of the hospitalized patients with cirrhosis in the verified cohort.
(DOCX)
Supplementary Table 5
Multivariate Cox analysis concerning 2-year all-cause mortality among hospitalized patients with cirrhosis incorporating categorized GEMA-Na.
(DOCX)
Supplementary Table 6
Multivariate Cox analysis concerning 2-year all-cause mortality among hospitalized patients with cirrhosis incorporating categorized MELD 3.0.
(DOCX)