Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
End-stage liver disease results from a variety of hepatic insults like viral hepatitis, alcohol and metabolic liver diseases. There is no specific treatment for end-stage liver disease apart from management of its complications. Liver transplantation often offers the only hope to these patients, and has been shown to prolong survival and improve the quality of life.1 Unfortunately, this effective treatment is hampered by the paucity of liver donors worldwide. Therefore, a good prognostic model that identifies patients at highest risk of dying from end-stage liver disease is essential.
Currently, liver graft allocation is prioritized according to the Model for End-Stage Liver Disease (MELD) in the United States and Hong Kong.2 The MELD score was first developed to predict the 3-month mortality in patients undergoing transjugular intrahepatic portosystemic shunt.3 It comprises 3 parameters: serum total bilirubin, international normalized ratio of prothrombin, and serum creatinine. Compared to other prognostic models such as the Child-Turcotte-Pugh score, these parameters are more objective and reproducible, and produce a continuous score that ranks patients in a large population more precisely.4–11
On the other hand, the MELD score does not include ascites in the model. Ascites itself is a strong factor associated with mortality in cirrhotic patients. Among cirrhotic patients who have developed ascites, approximately one-half will die in 2 yr.12 In order to improve the accuracy of the MELD score, investigators have looked into other potential prognostic markers. Serum sodium is another common laboratory test that is objective and reproducible. It is associated with ascites and hepatorenal syndrome, both being predictors of high mortality.13–17 Hyponatremia was also shown to be an independent factor associated with liver-related mortality.14, 15, 17–22 Previous attempts to incorporate serum sodium into the MELD score resulted in better accuracies in predicting mortality than the MELD score alone.23, 24 Among 753 Americans with cirrhosis mainly due to chronic hepatitis C and alcoholic liver disease, those with MELD-Na scores of 20, 30, and 40 had 6-month mortality of 6%, 16%, and 37%, respectively.24 The authors also estimated that 27% of the transplant recipients would be affected if the new MELD-Na score was used for organ allocation.
Previous studies of prognostic models were mainly from the United States and Europe. In Asia, chronic hepatitis B is the main cause of end-stage liver disease. Previous studies in Asia suggest that there may be other important prognostic markers not considered in the MELD score. In 96 Chinese patients with decompensated cirrhosis because of chronic hepatitis B, hepatic encephalopathy and hypoalbuminemia were the most important factors predicting mortality.25 Recently, we also demonstrated that the Child-Turcotte-Pugh score might perform better than the MELD score in patients with chronic hepatitis B.21 In this study, we investigated the performance of a composite score incorporating serum sodium into the MELD score in predicting 3-month and 1-yr mortality in a large cohort of Chinese patients with chronic hepatitis B.
MELD, Model for End-Stage Liver Disease; AUC, area under receiver operator characteristic curve; CI, confidence interval; HR, hazard ratio.
PATIENTS AND METHODS
This is a retrospective study on consecutive adult Chinese patients (aged 18 yr or above) admitted to Prince of Wales Hospital, Hong Kong, between January 1996 and December 2003 because of chronic hepatitis B-related complications.21 Eligible patients were identified by electronic database search based on the International Classification of Disease codes on discharge including chronic hepatitis B, variceal bleeding, spontaneous bacterial peritonitis, hepatic encephalopathy, liver cirrhosis, and ascites. Chronic hepatitis B was defined as positive hepatitis B surface antigen for at least 6 months. We excluded patients with other etiologies of chronic liver disease, clinical admissions for liver biopsy, clinical admissions for non-liver-related investigations and past or current hepatocellular carcinoma.
Clinical Data and Prognostic Models
Baseline demographic, clinical, and laboratory results were retrieved from the medical records. Clinical cirrhosis was defined as ultrasonic features of cirrhosis plus at least 1 of the following: hypersplenism, ascites, varices, or hepatorenal syndrome.26 Ascites was defined as free peritoneal fluid identified by ultrasound or computer tomography scans, or as clinically evident ascites confirmed by paracentesis. Spontaneous bacterial peritonitis was defined as an ascitic fluid polymorph count of 250/mm3 or above with or without positive bacterial culture. Variceal bleeding was defined as hematemesis and/or melena, with esophageal or gastric varices identified during upper gastrointestinal endoscopy, and no other lesion that might explain the bleeding episode. The prognostic models were calculated based on laboratory results obtained on the first day of admission. The MELD score was calculated by the formula (0.957 ln[creatinine] + 0.378 ln[bilirubin] + 1.120 ln[international normalized ratio of prothrombin] + 0.643) × 10.27 To incorporate serum sodium in the prognostic model, we calculated the MELD-Na score according to the formula by Biggins et al.24: MELD + 1.59 (135 − serum sodium), where the minimum value for serum sodium was 120 mmol/L and the maximum was 135 mmol/L.
All analyses were performed using the Statistical Package for Social Sciences (SPSS, version 11.5.1; Chicago, IL). Continuous variables were tested for normal distribution and expressed as mean ± standard deviation or median (range) as appropriate. Categorical variables were compared by Pearson chi-squared test and continuous variables were compared by Student's t-test or Mann-Whitney U test as appropriate. The cumulative transplant-free survival at different categories of MELD scores and MELD-Na scores (at intervals of 10 points) were performed by Kaplan-Meier analysis and compared by log rank test. Patients who received liver transplantation were classified as dead in the survival analysis. Cox proportional hazard model was used to compare the baseline demographic and clinical variables to predict long-term mortality. Multivariate analysis was performed for MELD and MELD-Na scores, both separately and simultaneously, to adjust for prognostic factors including age, gender, variceal bleeding, spontaneous bacterial peritonitis, hepatic encephalopathy, ascites, serum albumin, and alanine aminotransferase. Other prognostic factors including bilirubin, creatinine, international normalized ratio of prothrombin, and sodium were not included in the multivariate analysis to avoid the effect of colinearity as they were components of the MELD or MELD-Na scores. The area under the receiver operator characteristic curve (AUC), was used to measure the performance of MELD and MELD-Na scores for 3-month and 1-yr mortality. The AUC were compared by Delong's test.28, 29 Patients who were lost to follow-up during the first yr were excluded from the AUC analysis. A 2-sided P-value of less than 0.05 was considered statistically significant.
During the study period, 2,073 patients with liver problems were admitted to the hospital. After exclusion of patients without chronic hepatitis B (722), age under 18 (3), non-Chinese ethnicity (9), admissions due to non-liver-related reasons (218), clinical admissions for liver biopsy (425), active malignancies including hepatocellular carcinoma (145) and incomplete case records (188), 363 chronic hepatitis B patients were available for the current analysis. The baseline characteristics of the patients were shown in Table 1. The majority was men (80%) and the age was 51 ± 15 yr. A total of 76% of the patients had clinical liver cirrhosis. During the index admission, 86 (24%) patients received lamivudine treatment. None of the patient had renal failure requiring dialysis.
Table 1. Baseline Characteristics of Patients Admitted with Chronic Hepatitis B–Related Complications
All patients (n = 363)
NOTE: Continuous variables are presented as mean ± SD. Total bilirubin and ALT are presented as median (range).
Abbreviations: SD, standard deviation; ALT, alanine aminotransferase.
Mean ± SD
51 ± 15
Male gender, n (%)
Body mass index (kg/m2)
23.2 ± 3.5
Cirrhosis, n (%)
Encephalopathy, n (%)
Ascites, n (%)
Variceal bleeding, n (%)
Spontaneous bacterial peritonitis, n (%)
Hepatorenal syndrome, n (%)
Lamivudine treatment, n (%)
137 ± 5
102 ± 56
28 ± 8
Platelet count (×109/L)
118 ± 72
International normalized ratio of prothrombin
1.6 ± 0.6
Serum Sodium and Survival
The mean serum sodium level of the chronic hepatitis B patients was 137 ± 5 mmol/L. A total of 106 (29%) patients had serum sodium below 135 mmol/L, the lower limit of normal. A total of 33 (9%) patients had more profound hyponatremia below 130 mmol/L. Patients with ascites had significantly lower serum sodium levels (ascites vs. no ascites: 135 ± 6 vs. 138 ± 3, P < 0.0001).
At a median follow-up of 106 weeks (range 0.1-442 weeks), 134 (37%) patients died and 14 (4%) received liver transplantation. A lower serum sodium level was associated with reduced survival (Fig. 1A). In a Cox proportional hazard model, serum sodium remained an independent factor associated with deaths after adjusting for age, gender, variceal bleeding, spontaneous bacterial peritonitis, hepatic encephalopathy, ascites, serum albumin, alanine aminotransferase, and the MELD score (Table 2). While patients with serum sodium 130 to 135 mmol/L had similar mortality to those with serum sodium above 135 mmol/L (adjusted hazard ratio [HR] 1.6, 95% confidence interval [CI], 0.7-3.3; P = 0.25), the mortality of patients with serum sodium below 130 mmol/L increased by 2.9-fold (95% CI, 1.2-7.0; P = 0.02).
Table 2. Multivariate Analysis of MELD-Na Score and Major Prognostic Factors for Prediction of Long-Term Mortality (Median Follow-Up 106 Weeks, Range 0.1-442 Weeks)
Multivariate model for MELD
Multivariate model for MELD with lamivudine
Multivariate model for MELD-Na
Multivariate model for MELD-Na with lamivudine
NOTE: The upper group of rows show the analysis of the entire study population (n = 363). The lower group of rows show the analysis of patients with clinical liver cirrhosis (n = 277). Both models, including and excluding lamivudine treatment as a covariate, are shown.
Serum sodium level had moderate negative correlation with the MELD score (Pearson's coefficient −0.45, P < 0.0001). After incorporation of serum sodium into the MELD score, patients in this cohort had a mean MELD-Na score of 18.8 ± 9.4 (range 6-76). The majority had MELD-Na scores between 11 and 20 (Fig. 1B). Patients with higher MELD-Na scores were older and had more complications including hepatic encephalopathy, ascites, and spontaneous bacterial peritonitis (P < 0.001 for the comparison of all the above parameters among groups).
The MELD-Na score correlated strongly with short-term in-hospital mortality. No patient with MELD-Na scores below 10 died during the index admission. The corresponding numbers of patients with MELD-Na scores 11-20, 21-30, and above 30 who died during the index admission were 1 (0.5%), 7 (10%), and 11 (28%), respectively (P < 0.0001).
During long-term follow-up, a high MELD-Na score was also associated with reduced survival (Fig. 1B). At a median follow-up of 106 weeks (range: 0.1-442 weeks), the numbers of patients with MELD-Na scores ≤10, 11-20, 21-30, and >30 who died were 6 (11%), 67 (33%), 37 (54%), and 25 (64%), respectively (P < 0.0001). In a Cox proportional hazard model, MELD-Na independently predicted long-term survival after adjusting for age, gender, variceal bleeding, spontaneous bacterial peritonitis, hepatic encephalopathy, ascites, serum albumin, and alanine aminotransferase (Table 2). The results of the Cox proportional hazard model were similar for the entire study population and patients with clinical cirrhosis only, except that serum alanine aminotransferase was not a predictor of mortality independent of the MELD-Na score and other prognostic factors in patients with cirrhosis. Compared to patients with MELD-Na score below 10, the adjusted mortality of patients with MELD-Na scores 11-20, 21-30, and above 30 increased by 2.0-fold (95% CI, 0.9-4.7; P = 0.11), 4.7-fold (95% CI, 1.9-12.7; P = 0.001), and 7.6 fold (95% CI, 2.8-20.6; P < 0.0001), respectively.
Since there have been suggestions that hyponatremia had less impact on mortality in patients with high MELD scores, we compared the HRs obtained from fitting a proportional hazards model using six categories of patients by their serum sodium (<130 or ≥130 mmol/L) and MELD scores (<14, 14-20, and >20) (Table 3). For patients with MELD score 14 to 20, hyponatremia below 130 mmol/L increased the HR of death (with reference to patients with serum sodium above 130 mmol/L and MELD score below 14) from 2.0 (95% CI, 1.1-3.8) to 12.4 (95% CI, 4.4-34.6). For patients with higher MELD score above 20, hyponatremia below 130 mmol/L increased the HR from 4.3 (95% CI, 2.3-8.2) to 19.4 (95% CI, 9.0-41.6). Thus, the effect of hyponatremia on mortality was maintained even in patients with high MELD scores.
Table 3. Effect of Hyponatremia on Mortality During Long-Term Follow-Up in Patients with Different MELD Scores
NOTE: Numbers in each cell represent the HR of death, 95% CIs, and number of patients. The HR is with reference to patients with serum sodium at or above 130 mmol/L and MELD score below 14.
Na ≥ 130
n = 125
n = 141
n = 72
Na < 130
n = 0
n = 7
n = 18
A total of 7 (2%) patients were lost to follow-up during the first yr, leaving 356 patients in the analysis. During this period, 85 (23%) patients died and 14 (4%) patients received liver transplantation. The AUC for MELD as a predictor of 1-yr mortality was 0.72 (95% CI, 0.66-0.78), which was lower than MELD-Na (0.75; 95% CI, 0.69-0.81; P = 0.004). If only patients with clinical cirrhosis were tested, 270 patients remained in the analysis. The AUC for MELD (0.73, 95% CI, 0.67-0.80) remained lower than that for MELD-Na (0.75, 95% CI, 0.69-0.82, P = 0.048).
At 3 months, 2 (0.6%) patients were lost to follow-up, leaving 361 patients in the analysis. During this period, 47 (13%) patients died and no patient received liver transplantation. Similar to the 1-yr data, the AUC for MELD (0.75, 95% CI, 0.67-0.84) as a predictor of 3-month mortality was inferior to that for MELD-Na (0.79, 95% CI, 0.71-0.88; P = 0.004). Among 275 patients with clinical liver cirrhosis, the AUC for MELD (0.76, 95% CI, 0.67-0.85) was also lower than that for MELD-Na (0.79, 95% CI, 0.71-0.88; P = 0.02).
To assess whether patients who were lost to follow-up might affect the calculations, a sensitivity analysis was performed. At both 1 yr and 3 months, MELD-Na outperformed MELD no matter whether patients lost to follow-up were regarded as survived or dead. When all missing cases were considered dead, the AUC for MELD at 1 yr was 0.67 (95% CI, 0.61-0.73) while that for MELD-Na was 0.69 (95% CI, 0.63-0.75; P = 0.003). The AUC for MELD at 3 months was 0.70 (95% CI, 0.62-0.79), which was significantly lower than that for MELD-Na (0.74, 95% CI, 0.67-0.82, P = 0.002). On the other hand, when all missing cases were considered to have survived, the AUC for MELD at 1 yr was 0.72 (95% CI, 0.66-0.78) and that for MELD-Na was 0.75 (95% CI, 0.65-0.81, P = 0.007). The respective AUC at 3 months were 0.75 (95% CI, 0.67-0.84) and 0.79 (95% CI, 0.71-0.87, P = 0.006), respectively.
THE IMPACT OF ANTIVIRAL THERAPY
During the study period, 86 (24%) patients received lamivudine treatment and 277 (76%) patients did not receive any antiviral therapy because of low hepatitis B virus deoxyribonucleic acid, patient refusal or financial reasons. At a median follow-up of 106 weeks, 7 of 86 (8%) patients on lamivudine and 127 of 277 (46%) patients with no lamivudine treatment died (P < 0.0001). While lamivudine did not have major impact on in-hospital mortality (2 of 86 [2%] in lamivudine group vs. 17 of 277 [6%] in no lamivudine group, P = 0.32), patients receiving lamivudine had lower 3-month (3 of 86 [4%] vs. 44 of 277 [16%], P = 0.008) and 1-yr (4 of 86 [5%] vs. 81 of 277 [29%], P < 0.0001) mortality.
When incorporated into the multivariate analysis, lamivudine treatment was a predictor of mortality independent of other prognostic factors and MELD score (Table 2; adjusted HR 0.42; 95% CI, 0.18-0.97). On the other hand, lamivudine was no longer an independent predictor of mortality after adjusting for MELD-Na score (adjusted HR 0.54; 95% CI, 0.24-1.22). Incorporation of lamivudine treatment into the model did not affect the performance of MELD-Na score.
To further assess the impact of antiviral treatment on the prognostic model, MELD-Na score was calculated both at baseline and 3 months after lamivudine treatment for the 86 patients who received lamivudine treatment. At baseline, the median MELD-Na score was 19 (range 9-41). Only 4 (5%) patients had MELD-Na scores below 10, and 29 (34%) patients had MELD-Na scores above 20. Three months after lamivudine treatment, the median MELD-Na score dropped to 10 (range 6-69, P < 0.0001 compared to baseline). At this time point, 53 (62%) patients had MELD-Na scores below 10, while only 9 (10%) had MELD-Na scores remaining above 20. Only 1 of 75 (1%) patients with MELD-Na scores below 20 at 3 months died within 1 yr. In contrast, among the 9 patients with MELD-Na scores above 20 at 3 months after lamivudine treatment, 3 (33%) died at 1 yr.
In this large cohort of Chinese chronic hepatitis B patients with liver-related complications, the MELD-Na score was superior to MELD score in predicting 3-month and 1-yr mortality. The superiority was observed in both the overall population and patients with clinical cirrhosis.
Our study confirms the association between hyponatremia and mortality in patients with end-stage liver disease. In cirrhotic patients with hyponatremia, impaired renal function may be detected by inulin clearance and paraaminohippurate clearance.20 Hyponatremia is also associated with higher plasma renin activity, aldosterone, and norepinephrine concentration.13 Among 262 patients listed for liver transplantation in Argentina, hyponatremia below 130 mmol/L increased the 3-month mortality across all MELD scores.23 Using serum sodium as a continuous variable and incorporating it into the MELD score further increased the accuracy of the prediction model.24 As some patients with low MELD scores also have poor prognosis, Heuman et al.30 further studied a subgroup of cirrhotic patients (mainly due to chronic hepatitis C and alcohol) and found that hyponatremia and ascites were the only 2 independent factors associated with mortality when the MELD score was below 21. In those studies, ascites was absent in 30 to 40% of the patients. Thus, ascites alone could not totally explain the increase in mortality. In our cohort, though patients with ascites had lower serum sodium, hyponatremia remained an independent predictor of mortality even after adjustment for ascites. The majority of our patients had baseline MELD score below 20. In keeping with other studies in Western countries, incorporating serum sodium or ascites variables into the MELD score improved the power to predict mortality.
During long-term follow-up, the MELD-Na score is accurate in predicting mortality. While only just over 10% of patients with MELD-Na score below 10 died during the study period, two-thirds of patients with MELD-Na score above 30 died. Apart from long-term mortality, MELD-Na is also powerful in predicting short-term mortality. In our series, no patient with MELD-Na score below 10 died in the index admission. By contrast, the in-hospital mortality of patients with MELD-Na scores 21-30 and above 30 was 10% and 28%, respectively. This may help clinicians decide whether a patient requires urgent referral to a center with transplant facilities.
Although this is not a randomized controlled trial, lamivudine treatment resulted in improved survival in patients with chronic hepatitis B–related complications. During a median follow-up of 106 weeks, 8% of the patients on lamivudine and 46% of those not on antiviral therapy died. Since it takes a few weeks for lamivudine to work, it is not surprising that lamivudine did not affect in-hospital mortality, while the difference in mortality was apparent from 3 months onward. This may imply that antiviral treatment has strong influence on the mortality of patients with end-stage liver disease due to chronic hepatitis B. Indeed, among patients with chronic hepatitis B and biopsy-proven cirrhosis, lamivudine treatment not only retards the increase in Child-Turcotte-Pugh score, but also reduces cirrhotic complications and hepatocellular carcinoma.31 Even in patients with decompensated cirrhosis secondary to chronic hepatitis B, lamivudine treatment could achieve good viral suppression and improve the Child-Turcotte-Pugh score in some cases.32
In our cohort, patients on lamivudine treatment had dramatic improvement in MELD-Na scores. At baseline, only 5% of the patients had MELD-Na scores below 10, and the percentage rose to 62% by 3 months. The mortality rate also followed the on-treatment MELD-Na score. Using the 3-month on-treatment MELD-Na score, only 1% of the patients with score below 20 died at 1 yr, compared to 33% of those with score above 20. Since oral nucleoside analogs carry minimal side effects and effectively lower the viral load and alanine aminotransferase in most patients, they should be considered in all patients with hepatitis B–related liver cirrhosis and significant viremia. Our data support that it is necessary to recalculate the MELD-Na score after starting antiviral therapy to better reflect the patients' prognosis.
Our study has several limitations. First, because of the retrospective design, missing data were unavoidable. By an extensive search of discharge codes, we managed to limit patients with incomplete case records to 9%. Second, patients who were lost to follow-up might have died in other hospitals. We attempted to minimize the effect by searching a territory-wide interhospital electronic case record for discharges from other hospitals. Using this method, the 1-yr mortality data were only missing in 7 of 363 patients. Third, serum sodium is associated with both ascites and hepatorenal syndrome. As the diagnosis of hepatorenal syndrome might not be accurate in a retrospective study, this was not analyzed in the multivariate model. However, serum creatinine is a component of the MELD-Na and MELD scores, and the effect of hepatorenal syndrome is partially included in the models. Fourth, this study only included patients with chronic hepatitis B and thus the results could not be generalized to patients with other chronic liver disease. Nevertheless, the association between serum sodium and liver-related mortality and the validation of MELD-Na score have been performed in other populations predominantly affected by chronic hepatitis C and alcoholic liver disease.16, 20, 22–24 Since our previous study showed that prognostic models might perform differently in patients with chronic hepatitis B, which remained the most important cause of liver-related complications in Asia,21 we concentrated on this specific population in this study. Last, we did not analyze the use of MELD-Na in patients presented with hepatocellular carcinoma. In fact, MELD score alone is not an accurate tool to predict mortality in patients with hepatocellular carcinoma.21
In conclusion, we have preliminarily validated the MELD-Na score in Chinese chronic hepatitis B patients with liver-related complications. MELD-Na score performed better than the conventional MELD score in predicting 3-month and 1-yr mortality.
V.W. is supported by a research fellowship scheme of the Hong Kong Association for the Study of Liver Diseases.