Hepatic encephalopathy as a predictor of survival in patients with end-stage liver disease



Hepatic encephalopathy (HE) is an important component of hepatic decompensation, which reduces survival in patients with cirrhosis. The Model for End-Stage Liver Disease (MELD) score has been used to predict survival of patients with cirrhosis. The aims of this study were to determine whether HE is a predictor of survival of patients with cirrhosis and to examine the degree to which HE may add to the survival prediction of MELD. Patients with end-stage liver disease whose data were included in 2 databases were included in the analysis: 223 patients undergoing transjugular intrahepatic portosystemic shunt (TIPS) insertion, and 271 patients hospitalized with hepatic decompensation. In univariate analysis, HE grade 3 or higher was associated with a 3.7-fold (95% confidence interval, 1.9–7.3, P < 0.01) increase in the risk of death in the TIPS patients and HE grade 2 or higher was associated 3.9-fold increase (95% confidence interval [95% CI], 2.6–5.7, P < 0.01) in hospitalized patients. As expected, MELD and Child-Turcotte-Pugh scores (with and without HE included) were also markedly associated with survival. When HE (grade 2 or higher) and MELD were considered together, HE remained strongly statistically significant in the hospitalized patients (hazard ratio = 2.6, 95% CI, 1.7–3.8, P < 0.01). The effect became smaller in the TIPS patients (hazard ratio = 1.1; 95% CI, 0.7–1.6, P = 0.76). In conclusion, this retrospective study demonstrates that HE is an important event in the natural history of cirrhosis that affects subsequent survival of patients. HE may provide additional prognostic information independent of MELD, which warrants prospective validation. Liver Transpl, 2007. © 2007 AASLD.

Hepatic encephalopathy (HE) is a serious complication of decompensated cirrhosis that manifests as a wide range of neuropsychological clinical findings ranging from minimal HE to coma. In patients with cirrhosis, hepatic decompensation, commonly defined by ascites, HE, variceal bleeding and jaundice, is associated with far shorter survival. According to D'Amico et al.,1 the median survival of patients with cirrhosis decreases from >8 years to approximately 2 years.1, 2 Similarly, the development of HE has been associated with shortened survival among patients with end-stage liver disease.

The Model for End-Stage Liver Disease (MELD) score has been adopted as an objective indicator of liver disease severity. It consists only of laboratory data: serum total bilirubin and creatinine, and the international normalization ratio for prothrombin time. One of the reasons that MELD is entirely based on verifiable laboratory results is because it was intended to be used for organ allocation, for which parameters that entail subjective assessment were deemed undesirable. Since MELD was established, questions have risen whether individual complications of portal hypertension, such as refractory ascites and spontaneous bacterial peritonitis, provide additional prognostic information.

In this work, we evaluate the role of HE in predicting the survival of patients with end-stage liver disease. Specifically, the aims were to determine whether HE is a predictor of survival of patients with cirrhosis, and to examine the degree to which HE may contribute to the survival prediction of MELD.


HE, hepatic encephalopathy; MELD, Model for End-Stage Liver Disease; CTP, Child-Turcotte-Pugh; TIPS, transjugular intrahepatic portosystemic shunt; c-statistic, concordant statistic; HR, hazard ratio.


Patient Population

After obtaining approval from the Mayo Foundation Institutional Review Board, 2 databases that included information of 494 patients with cirrhosis were used for this analysis. The databases included information about patients undergoing insertion of a transjugular intrahepatic portosystemic shunt (TIPS) (n = 223) and patients hospitalized with decompensated cirrhosis (n = 271).

The TIPS group consisted of patients who were to have an elective TIPS procedure for the management of recurrent variceal bleeding or refractory ascites. TIPS was inserted in patients for the prevention of variceal bleeding and treatment of refractory ascites, and all patients were hemodynamically stable. These patients have been previously described.3 After TIPS insertion, patients were followed until the date they were last known to be alive, the date of liver transplantation, or the date of death.

Patients from the hospitalized group consisted of all adult patients, 18 years or older, with cirrhosis who were hospitalized at Mayo Clinic Rochester from 1994 to 1999. Patients were identified by a computerized diagnostic index and included in the study once it was ascertained that the diagnosis of cirrhosis was unambiguous. Patients were excluded if they had severe cardiopulmonary disease, sepsis, renal disease, hepatocellular carcinoma, or alcohol use <1 month before hospitalization, or were listed to undergo retransplantation because their comorbidities would have a negative impact on the natural history of their liver disease. These patients have been described in detail elsewhere.3

Patient Evaluation

The grade of HE was obtained for each patient by the treating gastroenterologist or hepatologist,4 and the concurrent laboratory variables were assessed. The severity of HE was graded from 0 to 4 according to Conn and Moilton.4 Patients who had no clinically overt signs of HE were classified as grade 0.

MELD scores were calculated by using serum creatinine, serum total bilirubin, and international normalized ratio (INR) according to the following formula, currently used by the United Network of Organ Sharing:

equation image

The MELD score was calculated at the time patients were assessed clinically for evidence of HE and grade of HE; for TIPS patients, this represented grade of HE within 24 hours of TIPS insertion, and for hospitalized patients the grade of HE at the time of admission.

Components of the Child-Turcotte-Pugh (CTP) score were also collected at the same time as MELD. These included total bilirubin, albumin, prothrombin time, and ascites. Ascites was graded according to 0 (none), 1 (mild), and 2 (moderate to severe). This assessment was conducted by the same gastroenterologist or hepatologist who graded HE. The criteria for bilirubin for cholestatic and noncholestatic disease were those of Child and Turcotte.5

Statistical Analysis

The effect of HE on short-term survival, defined as 3 months, and long-term survival, defined as 12 months, was determined. This was assessed separately in each group because the TIPS group had undergone an intervention that could alter outcome. The probability of survival was estimated by the Kaplan-Meier analysis. The log rank test and the Cox proportional hazard regression analysis were used for survival analysis. The start date for the study was the day of TIPS insertion for the TIPS group and the date of admission for the hospitalized group. The end points that were examined were 3-month and 12-month survival and time to liver transplantation or death. Statistical analysis was performed by SAS6 or S-Plus software.

The accuracy of HE, MELD, and other variables in predicting survival was calculated by the concordant statistic (c-statistic). Values of 0.8–0.9 are considered diagnostically excellent, and a value of 0.7 has good diagnostic value.7


Patient Population

The characteristics of the 2 groups of patients with cirrhosis are listed in Table 1. The median age of the entire group was 58 (range, 47–67) years. The cause of liver disease was alcohol-induced cirrhosis in 221 patients (44.7%), viral hepatitis in 79 (16%), cholestatic liver disease in 62 (12.6%), and other causes of liver disease in 132 (26.7%).

Table 1. Patient Characteristics
CharacteristicTIPS (n = 223)Hospitalized patients (n = 271)Total (n = 494)P value
  1. Abbreviations: MELD, Model for End-Stage Liver Disease; CTP, Child-Turcotte-Pugh score with hepatic encephalopathy; IQR, interquartile range.

Age (yr), mean (range)56 (45.4–64.6)60.4 (49.5–69.2)58.2 (47–67)0.0002
Patients with diagnoses, n (%)   <0.0001
 Alcoholic liver disease138 (62)83 (31)221 (45) 
 Cholestasis19 (9)43 (16)62 (12) 
 Viral hepatitis24 (11)55 (20)79 (16) 
 Other42 (19)90 (33)132 (27) 
Hepatic encephalopathy grade, n (%)   <0.0001
 089 (40)0 (0)89 (18) 
 160 (27)219 (81)279 (56) 
 258 (26)41 (15)99 (20) 
 316 (7)11 (4)27 (5.3) 
 41 (0.4)0 (0)1 (0.2) 
Median MELD score (IQR)15 (10–21)12 (7–17)13 (9–19)<0.0001
Median CTP score (IQR)10 (8–11)8 (7–10)9 (8–10)<0.0001

The 2 groups were very different with respect to mean age; etiology of liver disease, such that alcohol-induced liver disease had a higher frequency in the TIPS group; and the median MELD and CTP scores, which were far higher in the TIPS group than the hospitalized group (Table 1).

After examining the clinical characteristics according to grades of HE, it was found that as the grades of HE increased, there were far more patients with alcohol-induced liver disease, higher MELD scores, and higher CTP scores (Table 2).

Table 2. Clinical Features According to Grade of Hepatic Encephalopathy*
FeatureGrade 0 (n = 89)Grade 1 (n = 279)Grade 2 (n = 99)Grades 3 and 4 (n = 27)Comparison P value
  • Abbreviations: MELD, Model for End-Stage Liver Disease; CTP, Child-Turcotte-Pugh.

  • *

    Values are reported as median (25th–75th percentile).

  • χ2 test.

  • Wilcoxon rank sum test.

  • §

    CTP score calculated excluding hepatic encephalopathy.

Cause of liver disease, n (%)    <0.01
 Alcoholic liver disease60 (67)88 (31.5)56 (56)17 (62.9) 
 Cholestasis6 (6.7)51 (18)5 (5)0 (0) 
 Virus6 (6.7)54 (19)13 (13)6 (22.2) 
 Other17 (19)86 (31)25 (25)4 (14.8) 
MELD score, median (range)12 (8–16)11 (7–17)18 (13–25)20 (15–23)<0.01
CTP score, median (range)§7 (6–8)6 (5–8)8 (7–10)9 (7–10)<0.01

Univariate Survival Analyses

The mean (range) duration of follow-up was 357.6 (0-1,384) days for the TIPS group and 660 (1–2,045) days for the hospitalized group. Median survival was 533 days for the TIPS group and 1,185 days for the hospitalized group. At the end of the follow-up, 110 patients in the TIPS group and 127 in the hospitalized group had died.

Figures 1 and 2 provide Kaplan-Meier survival estimates by HE grade in the TIPS and hospitalized patients, respectively. The corresponding univariate Cox model is provided in Table 3. In the TIPS patients, survival appeared to correlate with HE grade. Compared with patients with HE grade 0, survival in those with grade 1 was not statistically significantly different (hazard ratio [HR] = 1.20; 95% confidence interval [95% CI], 0.73–1.96). Patients with grade 3 HE had far poorer survival than those with grade 0 (HR = 3.68; 95% CI, 1.85–7.30). Survival in patients with grade 2 was intermediate (HR = 1.56; 95% CI, 0.98–2.50, compared with grade 0). In the hospitalized patients, there was a clear separation between those with grade 1 and those with grade 2 (HR = 4.06; 95% CI, 2.67–6.19) or 3 (HR = 3.37; 95% CI, 1.69–6.72). Survival in the latter 2 groups was not different from each other.

Figure 1.

Kaplan-Meier survival estimates of patients undergoing transjugular intrahepatic portosystemic shunt (TIPS) insertion according to grade of hepatic encephalopathy (HE).

Figure 2.

Kaplan-Meier plot of survival estimates of hospitalized patients with cirrhosis according to grade of hepatic encephalopathy (HE).

Table 3. Unvariate Analysis of Impact of Hepatic Encephalopathy on Survival
GroupHazard Ratio95% CIP value
  • Abbreviations: 95% CI, 95% confidence interval; TIPS, transjugular intrahepatic portosystemic shunt; HE, hepatic encephalopathy.

  • *

    The reference group was patients with grade 0 HE.

  • The reference group was patients with grade 1 HE.

 Grades of HE   
  3 and 43.681.85–7.30.0002
 Grades of HE   

Table 4 summarizes univariate analyses evaluating the effect of MELD and CTP. Both MELD and CTP were associated with survival. In the TIPS patients, each unit increase in MELD was associated with a 10% increase (HR = 1.1) in the risk of death. The effect was nearly identical among the hospitalized patients (HR = 1.13). The c-statistics were satisfactorily high (>0.7) in both patient groups. With regard to the CTP score, each unit increase in the score was associated with 31% and 79% rise in mortality in the TIPS and hospitalized patients, respectively. When HE was removed from the CTP score (the “modified CTP score” in Table 4), the HRs and c-statistics did not change very much.

Table 4. Univariate Analysis of Impact of MELD and CTP on Survival
CharacteristicHazard ratio95% CIP valueAUC
3 months1 year
  1. Abbreviations: MELD, Model for End-Stage Liver Disease; CTP, Model for End-Stage Liver Disease; AUC, area under the receiver-operator curve; 95% CI, 95% confidence interval; TIPS, transjugular intrahepatic portosystemic shunt.

 MELD score1.101.08–1.12<0.0010.750.75
 CTP score1.311.19–1.44<0.0010.680.67
 Modified CTP score1.361.22–1.52<0.0010.670.67
 MELD score1.131.10–1.15<0.0010.80.8
 CTP score1.791.60–2.01<0.0010.790.79
 Modified CTP score1.791.6–2.01<0.0010.790.79

Multivariable Analysis

Table 5 summarizes multivariable analyses, evaluating whether HE is predictive of survival independent of MELD. To effectively investigate the relationship between HE and MELD, dichotomized HE (present or absent) was evaluated as well as HE in individual grades. In the former analysis, HE was defined as present if HE grade was 2 or higher. Grades 2 or higher were chosen because it appeared that the survival of patients with HE declined as the grades increased above 2 (Fig. 2).

Table 5. Effect of HE on Survival After Adjusting for MELD
CharacteristicTIPSHospitalizedP valueHR95% CIP value
HR95% CI
  • Abbreviations: HE, hepatic encephalopathy; MELD, Model for End-Stage Liver Disease; TIPS, transjugular intrahepatic portosystemic shunt; HR, hazard ratio; 95% CI, 95% confidence interval; c-statistic, concordant statistic.

  • *

    The reference group was patients with grade 0 HE.

  • The reference group was patients with grade 1 HE.

HE (grade ≥2)1.070.70–1.610.762.551.72–3.78<0.01
c-statistic 0.76   0.81
HE grade      
 3 or higher2.16*1.04–4.470.0382.571.29–5.14<0.01
c-statistic 0.75  0.81 

In the TIPS patients, after adjusting for MELD, presence of HE was associated with a HR of 1.06 (95% CI, 0.70–1.61, P = 0.76). As suggested by Figure 2, the effect of HE was most pronounced in patients with HE grades 3 and 4 in this group of patients. Indeed, HE grade 2 or higher remained a predictor of survival (HR = 2.16, P = 0.04) after adjusting for MELD, with a c-statistic of 0.75 for both 90-day and 365-day survival.

In the hospitalized patients, presence of HE was a strong predictor of survival after adjusting for MELD. The effect size happened to be identical for grades 2 and 3 (HR = 2.57), and the respective c-statistics for 90-day and 365-day survival were 0.81 and 0.80.


We sought to determine whether HE predicts increased short-term and long-term survival of patients with cirrhosis and HE, and whether HE had any effect on the established predictors of survival of patients with cirrhosis by examining the effects of HE only, the effects on MELD and on CTP, and the effects of those predictors on the survival of these patients. Here, we demonstrate that HE is an important landmark in the natural history of cirrhosis in that patients with HE had far worse survival than those without HE. Although the presence and severity of HE correlate with other indicators of disease severity, such as MELD, HE demonstrably affects patient survival independent of MELD.

The impact of HE on the survival of patients with cirrhosis has been demonstrated previously by others.1, 2, 8, 9 Upon reviewing the literature assessing the prognostic factors associated with survival in patients with cirrhosis, D'Amico et al.1 found that the CTP score, albumin, bilirubin, age, ascites, prothrombin time, and the presence of HE were the most common predictors of death. HE occurs relatively late in the course of cirrhosis. Sanyal et al.10 demonstrated a strong association between MELD score and developing HE as well as HE and mortality. It has been postulated that for HE to occur, decreased hepatic function and portosystemic shunting are necessary to allow putative toxic molecules to reach the cerebral circulation. Thus, it makes pathophysiologic sense that HE provides additional prognostic information than biochemical parameters such as MELD.

The comparison between the TIPS and hospitalized patients is instructive. The former group represented a subgroup of patients with hepatic decompensation with refractory portal hypertensive complications, whereas the latter group comprised an heterogenous group, with a varying degree of hepatic decompensation. It is also important to recognize that the TIPS patients were chosen to be well enough to withstand the procedure, despite their severe liver disease. The other obvious difference between the 2 groups is the fact that the TIPS patients all underwent the procedure, which is associated with a number of complications such as fluid overload and congestive failure, renal insufficiency, and infection, in addition to precipitation of HE as a result of increased portosystemic shunting. These differences may explain why the impact of HE was greater in the hospitalized group than the other. For example, occurrence of random complications after TIPS will decrease the relationship between severity liver disease and mortality. Likewise, patients who were given TIPS in spite of HE of grade 2 or higher must have had favorable characteristics that led the clinician to proceed with the intervention. Nonetheless, our data show that advanced HE among TIPS patients did have prognostic significance independent of MELD.

The concordance statistics have been used by us and others as a gauge of accuracy of prognostic systems. It measures the proportion of correct ranking of all possible pairs of subjects with different outcomes (e.g., survival and death). Thus, when MELD was being evaluated for the primary purpose of its ability to rank wait-listed patients according to their risk of death, the c-statistic was an appropriate parameter. However, one of the limitations of the c-statistic is that when the proportion of subjects affected by the particular condition of interest is small, its impact on the c-statistic is small, even when that condition has a large impact on those affected. An example was hyponatremia among patients with decompensated cirrhosis. Hyponatremia has been shown to be an important prognostic indicator in this setting, with a large effect on mortality that correlated with the degree of hyponatremia. However, because it only affected a small proportion of patients, it had minimal impact on the c-statistic.

Similarly, the relatively small proportion of patients affected by advanced HE may explain why the addition of HE resulted in a small increase in the c-statistic. In the TIPS patients, HE grade 2 or more alone (affecting 33%) had a c-statistic of 0.56 for prediction of 3-month survival, which increased to 0.75 when MELD was added to the model. In the hospitalized patients, the c-statistic increased from 0.64 to 0.81 in the prediction of 3-month survival. The latter patients included only 19% with HE grade 2 or higher.

We must point out the limitations of our data, the largest of which was that this was a retrospective analysis of HE. One of the reasons HE was excluded from MELD and as an organ allocation criterion is that it is subjective. Thus, it is difficult enough to assess patients for HE and grade them consistently in a prospective study. This assessment is much less reliable when the study is conducted retrospectively, and this is well illustrated by the distribution of HE grade between the 2 data sets. It is likely that some of the hospitalized patients had a sufficiently minimal degree of HE to have been graded 0 if they had been assessed by the physicians in the TIPS study. Similarly, data represented in Figure 2 may suggest that HE grades 2 and 3 in the hospitalized patients may not have been too distinct from one another. These caveats notwithstanding, even with less than ideally reproducible grading of HE in our data, there is demonstrable correlation between severity of HE and survival.

Another limitation of our study is the relatively small sample size. One may hypothesize that the impact of HE on patient survival may be different depending on the level of MELD. For example, in patients with very high MELD scores, the risk of mortality is already so great that HE may no longer portend a much poorer prognosis. On the other hand, the impact of HE may be much greater in patients with low MELD. A similar relationship was seen in the interaction between hyponatremia and MELD.11, 12

In conclusion, the results of this analysis demonstrate that HE is an important event in the natural history of cirrhosis that affects subsequent survival of patients. In addition, it is strongly suggested that HE may provide additional prognostic information independent of MELD. Although the study was limited by the retrospective assessment of HE, a prospective validation of our data is warranted.