The Model for End-Stage Liver Disease (MELD), which is calculated from 3 biochemical variables (serum bilirubin, prothrombin time, and creatinine), has been shown to be more accurate in predicting survival than the Child-Turcotte-Pugh (CTP) classification for patients with cirrhosis awaiting liver transplantation in the United States.1 The MELD has been shown to predict mortality independently of the etiology of liver disease and the occurrence of complications of portal hypertension.2, 3 Subsequently, the liver allocation system for prioritizing adult patients on the transplantation wait list has been changed from a status-based algorithm to one using a continuous MELD severity scale.4–6
To further improve MELD-based liver allocation, recent studies have found that serum sodium (Na) is an important additional predictor of wait-list mortality.7–9 Hyponatremia is associated with severe complications of cirrhosis, including ascites, hepatorenal syndrome, and liver-related mortality.10–13 It has been suggested that Na should be incorporated into the MELD to further enhance the model's prognostic ability, and so a mathematical equation based on both MELD and Na, known as the Model for End-Stage Liver Disease with the incorporation of serum sodium (MELD-Na), has been developed to predict the 6-month mortality in patients with cirrhosis awaiting liver transplantation from a multicenter study.14 Additionally, 2 recent independent studies also introduced MELD-based prognostic models, the integrated Model for End-Stage Liver Disease (iMELD) score and the Model for End-Stage Liver Disease to sodium (MESO) index, which also contain Na in their equations to improve survival prediction in patients with cirrhosis.15, 16 However, limited data are available for a direct comparison of the performance of the MELD-Na and MELD,17 and the predictive ability of the other models has not yet been confirmed. In addition, it is still uncertain that the strategy of incorporating Na into the MELD is a valid approach. In this study, we compare the short-term prognostic ability of the 4 models—MELD, MELD-Na, iMELD, and MESO index—in a single institute to determine if Na-containing MELD systems have a better predictive accuracy in patients with cirrhosis.
AUC, area under the receiver operating characteristic curve; CI, confidence interval; CTP, Child-Turcotte-Pugh; HBsAg, hepatitis B surface antigen; HCV, hepatitis C virus; iMELD, integrated Model for End-Stage Liver Disease; INR, international normalized ratio; MELD, Model for End-Stage Liver Disease; MELD-Na, Model for End-Stage Liver Disease with the incorporation of serum sodium; MESO, Model for End-Stage Liver Disease to sodium; Na, serum sodium; NPV, negative predictive value; PMN, polymorphonuclear leukocyte; PPV, positive predictive value; PT, prothrombin time; SBP, spontaneous bacterial peritonitis; SD, standard deviation.
PATIENTS AND METHODS
Our hospital, the Taipei Veterans General Hospital, provides primary to tertiary medical care to the residents of northern Taiwan, an area of 11 million inhabitants. From January 1999 to March 2007, patients with liver cirrhosis who visited our hospital were evaluated, and their medial profiles were retrospectively analyzed in this study. The following criteria were used to select eligible patients: (1) an initial Child-Pugh score of 6 or higher, (2) no coexisting hepatocellular carcinoma or human immunodeficiency virus infection at presentation, and (3) a known initial MELD score at the time of evaluation and survival status at follow-up after 6 months. With these criteria, 846 patients were selected from our database. Of these, 21 patients who underwent liver transplantation within 6 months of their first visit were excluded from analysis. The natural history and outcome of the remaining 825 patients were assessed, forming the basis of this study. Some of these patients had been reported in another study that investigated the predictive ability of the MELD in different block distributions.18 The outcome was assessed as the 3- and 6-month mortality. This study was approved by the institutional review board and complies with the standards of the Declaration of Helsinki and current ethical guidelines.
Diagnosis and Definitions
The diagnosis of liver cirrhosis was based on characteristic findings, including physical stigmata of cirrhosis, decreased serum albumin, and increased serum globulin levels, computed tomography or ultrasonography findings of a nodular liver surface, coarsened echogenicity of liver parenchyma, an enlarged spleen, and the detection of esophageal varices by endoscopy. The underlying etiology of cirrhosis was attributed to hepatitis B virus infection if patients were seropositive for hepatitis B surface antigen (HBsAg; RIA kits, Abbott Laboratories, North Chicago, IL), and it was attributed to hepatitis C virus (HCV) infection if patients were seropositive for antibody against HCV (anti-HCV) by a second-generation enzyme immunoassay (Abbott Laboratories).
The presence and treatment of variceal bleeding have been reported in detail in our previous studies.19, 20 Variceal bleeding was diagnosed by (1) clinical signs of hematemesis or coffee-ground vomitus and (2) endoscopic signs of active bleeding or an adherent clot on esophageal or gastric varices. The diagnosis and treatment of spontaneous bacterial peritonitis (SBP) are in line with the recommendations of the International Ascites Club.21 SBP was suspected when the clinical signs of peritonitis and infection were present and was diagnosed when (1) the ascites polymorphonuclear leukocyte (PMN) count was >250/mm3 with or without positive ascites bacterial culture or (2) the ascites PMN count was <250/mm3 but with positive ascites bacterial culture. The West Haven criteria were used to define the severity of hepatic (or portosystemic) encephalopathy.22 The management of encephalopathy included the correction of underlying precipitating factors, administration of enema and oral lactulose, and other supportive treatment as needed.
Calculation of the MELD, MELD-Na, iMELD, and MESO Index
The MELD equation was used to calculate the severity score: 9.57 × loge[creatinine (mg/dL)] + 3.78 × loge[bilirubin (mg/dL)] + 11.2 × loge(international normalized ratio) + 6.43,1 where the minimal values were set to 1.0 for calculation purposes. The maximal serum creatinine level considered within the MELD score equation is 4.0 mg/dL. The MELD-Na equation was based on the MELD and Na, MELD + 1.59 × (135 − Na), with maximum and minimum Na values of 135 and 120 mEq/L, respectively.14 The iMELD equation was based on the MELD score, age (years), and Na (mEq/L): original MELD score + (age × 0.3) − (0.7 × Na) + 100.15 The MESO index was proposed in our previous report,16 and for calculation purposes, the index was defined as [MELD/Na (mEq/L)] × 100 in this study.
The chi-square test or Fisher's exact test (2-tailed) was used for categorical data, and the Mann-Whitney ranked sum test was used for continuous data. To assess the ability of the 4 MELD-based models in predicting the risk of mortality at 3 and 6 months, our analysis was performed by the measurement of the c-statistic equivalent to the area under the receiver operating characteristic curve (AUC). The comparison of the AUCs and the examination of the statistical significance were done with the method of Hanley and McNeil.23 All statistical analyses were conducted with the SPSS for Windows version 14 release (SPSS, Inc., Chicago, IL), SAS version 9.1 (SAS, Cary, NJ), and MedCalc for Windows version 4.2 (MedCalc Software, Mariakerke, Belgium). For all tests, a P value less than 0.05 was considered statistically significant.
The baseline demographics of the patients in this study are shown in Table 1. Patients were predominantly elderly (mean age, 64 ± 13 years) and male (76%), and 70% had evidence of chronic hepatitis B virus infection. More than half of the HBsAg-negative subjects (129 of 247 patients, 52%) had chronic hepatitis C. Alcoholic cirrhosis and cryptogenic cirrhosis were the major etiologies for the remaining 14% of patients. Because of different definitions in the calculations of the scores, the iMELD tended to have the highest scores (38.6 ± 9.2) and was followed by the MELD-Na (17.5 ± 9.2), MELD (15.0 ± 6.5), and MESO index (11.1 ± 5.0) at the time of initial evaluation.
Table 1. Baseline Demographics of the Study Patients
Number of patients
Abbreviations: CTP, Child-Turcotte-Pugh; HBsAg, hepatitis B surface antigen; HCV, hepatitis C virus; iMELD, integrated Model for End-Stage Liver Disease; INR, international normalized ratio; MELD, Model for End-Stage Liver Disease; MELD-NA, Model for End-Stage Liver Disease with the incorporation of serum sodium; MESO, Model for End-Stage Liver Disease to sodium; Na, serum sodium; PT, prothrombin time; SD, standard deviation.
Age (years; mean ± SD)
64 ± 13
Male gender (%)
Etiology of cirrhosis (%)
Serum biochemistry (mean ± SD)
3.4 ± 4.9
INR of PT
1.4 ± 1.2
1.4 ± 0.9
136 ± 5
CTP score (mean ± SD)
8.8 ± 1.7
MELD score (mean ± SD)
15.0 ± 6.5
MELD-Na score (mean ± SD)
17.5 ± 9.2
iMELD score (mean ± SD)
38.6 ± 9.2
MESO index (mean ± SD)
11.1 ± 5.0
Distribution of the MELD-Based Models According to the Na and MELD Score
As shown in Fig. 1, the MELD scores tended to have an inverse relationship with the Na levels; that is, the MELD score tended to be lower with increasing Na levels. The trend reached a plateau when Na was 141 mEq/L or higher. Alternatively, the Na-containing MELD-based models consistently showed increasing scores with increasing MELD score blocks (MELD < 12, 12 < MELD < 18, and MELD > 18; Fig. 2).
Comparison of the AUC and Predictive Accuracy at 3 and 6 Months Between Four MELD-Based Prognostic Models
With the c-statistic and 3- and 6-month mortality as the endpoint, the estimated AUCs for the 4 prognostic models in predicting mortality are given in Table 2 and Fig. 3. Of all patients, 83 (10.1%) and 162 (19.6%) patients died at 3 and 6 months of follow-up, respectively. At 3 months, the iMELD had the highest AUC (0.807) and was followed by the MELD-Na (0.801), MESO (0.784), and MELD (0.773). MESO had a significantly higher AUC in comparison with the MELD (P = 0.013) because of a very small standard error (0.004) in the difference in the areas of the statistical analysis; other comparisons showed no significant differences. At 6 months, the iMELD still had the highest AUC (0.797) and was followed by the MELD-Na (0.778), MESO (0.747), and MELD (0.735). All comparisons showed significant statistical differences between each other (all P < 0.01), with the exception of the iMELD and MELD-Na (P = 0.18).
Table 2. Comparison of the AUC To Predict 3- and 6-Month Mortality Between Four MELD-Based Prognostic Models
Abbreviations: AUC, area under the receiver operating characteristic curve; CI, confidence interval; iMELD, integrated Model for End-Stage Liver Disease; MELD, Model for End-Stage Liver Disease; MELD-NA, Model for End-Stage Liver Disease with the incorporation of serum sodium; MESO, Model for End-Stage Liver Disease to sodium.
Using the cutoffs that had the best discriminative ability to predict mortality derived from the receiver operating characteristic curve, we determined the sensitivity, specificity, positive predictive value, and negative predictive value for each prognostic model (Table 3). The sensitivity and specificity for the 4 models ranged from 61% to 83% and from 70% to 85%, respectively, at 3 and 6 months with the most discriminative cutoffs. The positive predictive value was low, ranging from 24% to 48%, whereas the negative predictive value ranged from 89% to 97% at 3 and 6 months.
Table 3. Sensitivity, Specificity, PPV, and NPV for the Four MELD-Based Prognostic Models Used To Predict Mortality with the Best Predictive Cutoffs at 3 and 6 Months
Abbreviations: iMELD, integrated Model for End-Stage Liver Disease; MELD, Model for End-Stage Liver Disease; MELD-NA, Model for End-Stage Liver Disease with the incorporation of serum sodium; MESO, Model for End-Stage Liver Disease to sodium; NPV, negative predictive value; PPV, positive predictive value.
MELD > 18.6
MELD-Na > 18.6
iMELD > 44.1
MESO > 12.8
MELD > 18.6
MELD-Na > 18.6
iMELD > 44.1
MESO > 12.8
Association of Major Cirrhosis Complications and the MELD-Based Prognostic Models
A total of 411 patients underwent complete endoscopic and diagnostic procedures to detect the presence of varices bleeding, SBP, and hepatic encephalopathy at the time of initial evaluation. We then compared the scores of the 4 prognostic models between patients with and without these complications. As shown in Table 4, patients with SBP consistently had significantly higher scores in all 4 models in comparison with patients without SBP (all P < 0.01). Patients with hepatic encephalopathy at a severity grade of 2 or higher also consistently had higher scores in all 4 models in comparison with patients without encephalopathy, although the statistical significance was established only for the iMELD model (P = 0.037). By contrast, patients with variceal bleeding had scores comparable to or even lower than those of patients without variceal bleeding at a statistically insignificant level in all 4 prognostic models.
Table 4. Association of Major Cirrhosis Complications (SBP, Hepatic Encephalopathy, and Variceal Bleeding) and the MELD-Based Prognostic Models
Abbreviations: iMELD, integrated Model for End-Stage Liver Disease; MELD, Model for End-Stage Liver Disease; MELD-NA, Model for End-Stage Liver Disease with the incorporation of serum sodium; MESO, Model for End-Stage Liver Disease to sodium; SBP, spontaneous bacterial peritonitis.
Yes (n = 23)
18.7 ± 7.0
24.2 ± 10.8
44.5 ± 7.7
14.2 ± 5.4
No (n = 388)
14.9 ± 6.4
17.4 ± 9.4
37.8 ± 9.6
11.0 ± 5.4
Yes (n = 69)
17.4 ± 10.0
20.0 ± 11.4
41.0 ± 11.5
12.8 ± 7.3
No (n = 342)
17.3 ± 10.2
17.3 ± 9.1
37.6 ± 9.1
10.9 ± 4.9
Yes (n = 96)
14.5 ± 7.6
16.6 ± 9.7
35.6 ± 9.5
10.6 ± 5.6
No (n = 315)
15.4 ± 7.0
18.1 ± 9.7
38.9 ± 9.5
11.4 ± 5.4
The MELD scoring system has been widely applied in recent years and shown to predict mortality across a broad spectrum of liver diseases in most studies.5, 24–27 The utilization of the MELD has been demonstrated to have an equal or better ability in short-term or intermediate-term outcome prediction in comparison with the CTP system.1, 28–30 In addition, the application of the MELD system has been shown to be a useful model in predicting the outcome of patients with cirrhosis undergoing surgical procedures for hepatocellular carcinoma or non–hepatocellular carcinoma conditions.31–34
Hyponatremia often indicates a state of hepatic decompensation in patients with liver cirrhosis and is strongly associated with the risk of mortality. By the incorporation of Na into the MELD, the MELD-Na scores of 20, 30, and 40 were associated with 6%, 16%, and 37% risk of death within 6 months of liver transplant listing.14 The predictive ability of the MELD-Na was recently confirmed in another independent report, which showed that its predictive accuracy was better than that of the MELD.17 We and others have also proposed Na-containing MELD-based prognostic models.15, 16 The iMELD was generated on the basis of 3 clinical variables (MELD, age, and Na) derived from a multivariate longitudinal survival analysis,15 whereas the MESO index was created intuitively in patients undergoing portal hemodynamic measurements.16 The feasibility of these strategies is still uncertain because there is evidence that shows that the addition of Na along with the MELD in the AUC analysis did not significantly improve the predictive accuracy of the MELD score alone in a cohort of 308 patients with cirrhosis.35 Our study included a larger patient population and provided further evidence that a combination of Na and MELD may have a better prognostic power in comparison with the MELD alone. Moreover, our analyses are unique in that such a direct comparison of these various models has not been carried out with the same patient cohort.
This study has 2 important clinical implications. First, the predictive ability of the MELD can be improved by the incorporation of Na into the different forms of the equations. Second, we found that the iMELD consistently had the best predictive ability of the 4 models at both 3 and 6 months. However, it should also be mentioned that the statistical difference between the iMELD and MELD-Na was not significant at both time points, and this suggests that these 2 models may be equally accurate for outcome prediction. Interestingly, in addition to the MELD and Na, the iMELD also takes into account the factor of age.15 It is expected that any mortality risk model that incorporates age is likely to have better discrimination in predicting who may survive because advanced age is an apparent mortality risk for every individual, regardless of any other disease states.
One of the important aspects of the MELD and its derived models is that they are continuous variables and account for the spectrum of disease severity. However, using the most discriminative cutoff from the receiver operating characteristic curve for different models may provide additional information in certain clinical settings. Notably, these models typically had a consistently low positive predictive value (48% or lower) but a high negative predictive value (89% or higher); therefore, this might be a useful strategy in selecting transplant candidates. This finding implicates that the chance to survive over a certain period of time (3 or 6 months) for patients with a score lower than the selected cutoff is quite high even without transplantation. For example, a patient with a MELD-Na score of 18 (< 18.6, Table 3) would have a 97% and 92% chance of survival at 3 and 6 months, respectively, without a liver transplant. Although the dichotomous approach is not directly related to the process of organ allocation and may not serve as an absolute standard in clinical practice, this information, which cannot be directly obtained from the AUC analysis, is crucial for transplant physicians in making decisions to avoid futile liver transplantation.
Several cirrhosis complications, such as SBP, hepatic encephalopathy, and variceal bleeding, are commonly seen and are major causes of death in patients with cirrhosis. However, these complications have not been taken into the consideration of all the MELD-derived systems. In this study, we have analyzed the association between these complications and the 4 prognostic models. In the context of SBP, we found that patients with SBP had consistently higher scores for the 4 models in comparison with patients without SBP. The development of SBP in cirrhosis has been reported to be closely associated with increasing MELD scores.36 Furthermore, recent evidence has shown that prevention of SBP with prophylactic antibiotics may improve the survival of patients with cirrhosis.37 These lines of evidence are consistent with the fact that patients with SBP had higher MELD-derived scores and highlight the role of SBP in the natural history of cirrhosis.
Although patients with hepatic encephalopathy (defined as grade 2 or higher) consistently had higher MELD-derived scores than patients without hepatic encephalopathy, the statistical significance was rather weak, with the exception of the iMELD. The occurrence of hepatic encephalopathy is clearly associated with poor survival in patients with cirrhosis and may contribute to an increasing MELD score over time.28, 38 On the contrary, patients with variceal bleeding did not have higher MELD-derived scores. This finding is consistent with our previous study39 and may well be due to the fact that the endoscopic and pharmacologic treatments are effective methods in reducing the chance of liver decompensation. In addition, the amount or magnitude of collateral circulation in portal hypertension does not necessarily correlate with the clinical severity of cirrhosis.40
A potential limitation of this study is that the majority of our patients had chronic hepatitis B, which is common in many Asian countries. Our patient group was also older with more males than the transplant recipients in the United States. Concerns may arise about whether our data are readily applicable in the Western countries in which HCV infection and alcoholism are major etiologies of cirrhosis. Because disease progression might not be the same between HBV-related and HCV-related cirrhosis, more studies are needed to address the potential impact of the viral factor on the outcome. In this respect, independent studies from Hong Kong and Taiwan have indicated that the MELD score is a valid prognostic model in decompensated chronic hepatitis B.41, 42 In addition, our previous studies have shown that the incorporation of the MELD into the staging system for hepatocellular carcinoma is a feasible strategy,43, 44 suggesting that the etiology of cirrhosis should not be considered an obstacle in the MELD-based platform.
In conclusion, our results indicate that incorporation of Na into the MELD may enhance its prognostic accuracy, and both iMELD and MELD-Na are better prognostic models for outcome prediction. Patients with SBP and, to a lesser extent, hepatic encephalopathy are associated with a higher MELD-derived score. Future studies are warranted to define the optimal MELD-based prognostic model for cirrhosis.