Fulminant hepatic failure (FHF) is a complex multisystem illness that results after a catastrophic insult to the liver, manifesting in the development of a coagulopathy and encephalopathy within a short period of time.1 It carries high mortality rate with short-term transplant-free survival rate of 43%.2 Moreover, it accounts for about 7% of liver transplants among adults.3 Therefore, the determination of prognosis for accurately predicting the outcome in FHF is of immense value in establishing the need for referral to specialist centers. Liver transplantation (LT) has significantly improved the survival in these patients and remains the cornerstone of treatment in selected patients with FHF.4, 5 It is vital that irreversible FHF be recognized early, in order to avoid life threatening complications of worsening cerebral edema and multiorgan failure.6 While early and accurate identification of those who will die without LT is crucial, a decision not to transplant is equally important as it avoids a liver replacement and the need for life-long immunosuppression.
Among various prognostic criteria proposed, the “Clichy criteria”7 and the “King's College Hospital (KCH) criteria”8 are used worldwide. According to the Clichy criteria, LT is recommended in the grade 3 or 4 encephalopathy with a factor V level <20% in patients under 30 yr of age or <30% if over 30 yr of age.7, 9 The Clichy criteria is in use in much of Northern Europe; however, their more widespread application has been hindered by 2 main factors, namely limited availability of factor V level measurement and derivation of criteria from a cohort of FHF patients due to hepatitis B virus infection.10 KCH criteria8 have been more widely used and include the clinical and biochemical data such as age, duration of jaundice, bilirubin, prothrombin time, arterial pH, and serum creatinine, which are routinely available in clinical practice and due consideration was also given to the etiology; i.e., acetaminophen vs. nonacetaminophen. These parameters are available on admission with in a few hours of the patient's arrival at the emergency room, which helps early listing and referral to specialist center. The criteria are highly predictive of poor outcome when fulfilled; a lack of fulfillment does not assume survival.11–14 Serial measurements of Acute Physiology and Chronic Health Evaluation III scores might be helpful in identifying nonsurvivors among patients who do not fulfill the KCH criteria.14 As a result, several other indices and measurements have been proposed as prognostic marker, such as, plasma coagulation factor V and factor VIII factor V ratio,15 arterial ketone body ratio,16 galactose elimination capacity,17 Gc globulin concentration,18 blood lactate,19, 20 blood phosphate,20, 21 and most recently arterial ammonia,22 but none have been validated in large prospective studies.
The Model for End-Stage Liver Disease (MELD) is a survival model based on a composite of 3 laboratory variables: serum creatinine, serum bilirubin, and international normalized ratio for prothrombin time.23 This model has been validated in several independent cohorts of patients with cirrhosis.24, 25 MELD more accurately stratifies patients according to mortality risk than the Child-Turcotte-Pugh score; hence the former has replaced the latter to prioritize and rank organ allocation of cadaveric livers for transplantation on the United Network for Organ Sharing liver waiting list.24, 25 The variables used in MELD (international normalized ratio, bilirubin, and creatinine), have also been used for the assessment of mortality in patients with FHF.8, 26, 27 Elevated serum creatinine has been shown to be associated with poor outcome in patients with FHF.8, 28
The present study was done to compare MELD and KCH criteria with other early clinical prognostic indicators (CPI) in a cohort of patients with FHF seen at our center.
KCH, King's College Hospital; MELD, Model for End-Stage Liver Disease; LT, liver transplantation; CPI, clinical prognostic indicators; FHF, fulminant hepatic failure; JEI, jaundice-encephalopathy interval; ROC, receiver operating characteristic curves; PPV, positive predictive value; NPV, negative predictive value; IgM, immunoglobulin M.
PATIENTS AND METHODS
In the retrospective analysis, a total of 160 consecutive patients with FHF admitted to the Emergency Medical Ward between January 1996 and June 1998 were included in the study. For the purpose of this study, acute liver failure was defined according to the criteria of O'Grady1; i.e., onset of hepatic encephalopathy occurring within 12 weeks of onset of jaundice and was further subclassified into hyperacute (interval 0–7 days), acute (interval 8–28 days), and subacute (interval 29 days–12 weeks) liver failure. A total of 7 (4.4%) patients were excluded because of incomplete record and 9 (5.6%) patients with drug induced FHF (all antitubercular drugs-related) were also excluded in order to study a homogenous group of acute viral hepatitis–induced FHF. Hence a total of 144 patients with FHF due to acute viral hepatitis were included in the study. None of the patients consumed alcohol in a significant amount; i.e., more than 50 gm/day of alcohol for 5 yr and none had the evidence of chronic liver disease based on clinical, biochemical, or abdominal ultrasound examination or at autopsy. A viral etiology was presumed when a history of exposure to drugs or toxins was absent and a typical prodromal illness was present. Patients with a history of intake of hepatotoxic drugs and absence of viral markers were diagnosed as having drug-induced liver failure.
Hepatic encephalopathy was graded from grade 1 to 4 according to the West-Hevan criteria.29 The grade of encephalopathy at the time of admission and the peak grade during hospitalization were noted. The interval between detection of jaundice to the onset of hepatic encephalopathy was defined as the “jaundice-encephalopathy interval” (JEI).27
The other clinical parameters noted on admission were the presence of ascites and cerebral edema. The laboratory parameters studied on admission were hemoglobin, white blood cell counts, prothrombin time, international normalized ratio, serum bilirubin, serum alanine aminotransferase, serum aspartate aminotransferase, serum alkaline phosphatase, serum albumin, serum creatinine, and serum sodium. The amount of fresh-frozen plasma infusion required for bleeding complications was also noted.
The diagnosis of cerebral edema was based on presence of any of the following: bradycardia, hypertension (150/90 mm Hg), increased muscle tone, unequal or abnormally reacting pupils, neurogenic hyperventilation, myoclonus, and spontaneous decerebrate posturing.27, 30, 31
The clinical diagnosis of ascites was made by the presence of shifting dullness of fluid in abdomen or by the presence of fluid thrill. Ultrasound examination of the abdomen was performed when clinical signs for ascites were ambiguous or where there was a suspicion of underlying chronic liver disease. Abdominal paracentesis was performed if any evidence of peritonitis developed, such as fever, peripheral leukocytosis, or ileus. The diagnosis of spontaneous bacterial peritonitis was based on the presence of an ascitic fluid polymorphonuclear cell count of >250/mm.3, 32
Renal failure was diagnosed if the patient developed a reduced urine output of <400 mL in 24 hours, with serum creatinine of >1.5 mg/dL in the absence of hypovolemia.
During the period of the study, LT was not available in the Institute. Therefore, all patients were managed conservatively. All patients were managed with a standard protocol of protein restriction, bowel decontamination, lactulose, rehydration, intravenous ranitidine or proton pump inhibitors, and bowel wash. All the patients were monitored regularly for hypoglycemia. Standard treatment for clinical evidence of cerebral edema included: elevation of head end of the bed to 15–30°, restriction of intravenous fluids according to central venous pressure, maintaining normotension, hydrotherapy for fever control, limiting all nonessential physical examination, correction of hypercapnia and hypoxemia, and parenteral mannitol (mainstay of the treatment) in standard doses. All patients in grade 3 or 4 hepatic encephalopathy were intubated electively for the respiratory support and mechanical ventilation provided for those with inadequate respiratory effort. Renal replacement therapy formed a part of standard care given to all patients having renal failure. This therapy consisted of peritoneal dialysis or continuous veno-venous hemofiltration. We did not use any bioartificial liver support devices. Fresh-frozen plasma was infused for bleeding manifestations as and when required. Appropriate antibiotics were given for treating infection. Spontaneous bacterial peritonitis was treated with intravenous cefotaxime for 10 days.32
The statistical endpoint was death due to any cause during the hospital admission or complete recovery. Univariate logistic regression or Yates corrected χ2 test was used to screen the variables reported in Table 1. Variables that were statistically significant formed a pool of potential independent predictors, which are reported in Table 2, after they were dichotomized for best discrimination between survivors and nonsurvivors by the construction of receiver operating characteristic curves (ROC) curves (duration of jaundice >5.5 days, bilirubin ≥20 mg/dL, prothrombin time ≥35 seconds, creatinine ≥1.5 mg/dL, grade of encephalopathy >2) or by taking the cutoff established previously (age ≥50 yr, JEI >7 days).27 For cerebral edema, the criteria was presence or absence. A MELD score of ≥33 was found to be best discriminant between survivors and nonsurvivors by the construction of ROC curves. The predictors were then entered into a backward elimination variable selection procedure by multivariate logistic regression analysis. The predictors with a P value less than 0.05 were retained (Table 3). The sensitivity, specificity, positive predictive value (PPV), and negative predictive values (NPV), diagnostic accuracy, and area under the ROC curve for each predictor identified on multivariate analysis and the combination thereof was then assessed. Concordance (c) (range 0.0–1.0) was equivalent to the area under the ROC curve. The c-statistic was used to compare MELD with other clinical prognostic criteria.
Table 1. Univariate Analysis of Variables on Admission in Patients With Fulminant Hepatic Failure
Variables on admission
NOTE: Normal values: alanine aminotransferase 2–15 IU/L, aspartate aminotransferase 2–20 IU/L, alkaline phosphatase 3–13 IU/L, albumin 3.5–5.5 g/dL, and creatinine 0.6–1.2 mg/dL.
Abbreviations: HBsAg, hepatitis B virus surface antigen; JEI, jaundice-encephalopathy interval; MELD, Model for End-Stage Liver Disease.
Age (mean ± SD)
28.6 ± 12.3
33.4 ± 15.8
Duration of jaundice (days)
7.8 ± 10.3
12.1 ± 13.5
5.9 ± 9.4
9.6 ± 12.7
Grade 3 or 4 encephalopathy
Presence of ascites
Fresh frozen plasma infusion
Presence of cerebral edema
13.2 ± 3.0
11.9 ± 2.4
White blood cells (×103/mm3)
3.3 ± 8.1
14.6 ± 7.6
16.2 ± 23.2
23.2 ± 11.3
Alanine aminotransferase (IU/L)
88.2 ± 76.5
89.0 ± 82.4
Aspartate aminotransferase (IU/L)
117.3 ± 77.9
115.9 ± 1.2.9
Alkaline phosphatase (IU/L)
20.6 ± 20.5
24.4 ± 44.6
2.9 ± 0.6
2.8 ± 0.5
Prothrombin time (seconds)
33.5 ± 14.2
51.8 ± 28.7
1.2 ± 0.6
2.2 ± 2.1
136.3 ± 6.0
134.6 ± 7.5
Presence of HBsAg
29.7 ± 6.4
38.8 ± 8.1
Table 2. Univariate Analysis of Dichotomous Variables Influencing the Outcome
Abbreviations: JEI, jaundice-encephalopathy interval; MELD, Model for End-Stage Liver Disease.
Statistical analysis was performed by SPSS software for Windows (version 10.0; SPSS, Chicago, IL).
A total of 144 patients with FHF due to acute viral hepatitis were analyzed.
Prodromal symptoms such as anorexia, vomiting, or diarrhea were present in all patients with viral hepatitis, hepatitis B virus was implicated in 61 (42.4%) patients. In order to study a homogenous group only 144 patients with FHF complicating viral hepatitis were analyzed. A total of 52 (36.1%) patients survived and the remaining 92 (63.9%) patients died. There were 98 (68.0%) patients with hyperacute, 39 (27.1%) with acute, and 7 (4.9%) with subacute liver failure. Of 98 patients with hyperacute liver failure, 43 (43.8%) survived, compared to 9 (19.6%) of 46 patients with combined acute and subacute liver failure (P = 0.005).
Demographic and Clinical Profile
Demographic, clinical, and laboratory variables on admission to the hospital retrieved from the patient record and analyzed with univariate analysis are shown in Table 1. The patient cohort consisted of 144 patients with a mean age of 31.7 ± 14.7 yr (range 12–82 yr). There were 62 (43.1%) males and 82 (56.9%) females.
Univariate analyses showed that a higher percentage of patients had age ≥50 yr (P = 0.026), duration of jaundice > 5.5 days (P = 0.006), JEI > 7 days (P = 0.005), grade 3 or 4 encephalopathy (P = 0.11), presence of cerebral edema (P < 0.0001), bilirubin ≥20 mg/dL (P = 0.007), prothrombin time ≥35 seconds (P = 0.002), and creatinine ≥1.5 mg/dL (P = 0.001) among nonsurvivors (Table 2). Presence of ascites, fresh-frozen plasma infusion, hemoglobin levels, white blood cell counts, serum levels of alanine aminotransferase, and aspartate aminotransferase, alkaline phosphatase, albumin, sodium, and hepatitis B surface antigen positivity were similar among survivors and nonsurvivors.
Table 2 showed the dichotomous variables that were significantly associated with mortality on univariate analysis. In multivariate logistic regression, a backward elimination selection procedure selected 6 independent predictors of outcome on admission, age ≥50 yr, JEI >7 days, grade 3 or 4 encephalopathy, presence of cerebral edema, prothrombin time ≥35 seconds, and creatinine ≥1.5 mg/dL. These factors adversely affected the outcome in our patients (Table 3).
The sensitivity, specificity, PPVs, NPVs, and diagnostic accuracy depending on the number of adverse prognostic markers are shown in Table 4. Presence of any 3 CPI was optimum in identifying survivors and nonsurvivors. The mortality increased with an increasing number of adverse prognostic factors.
Table 4. Assessment of Clinical Prognostic Indicators
Mean MELD score was significantly higher among nonsurvivors than survivors (38.8 ± 8.1 vs. 29.7 ± 6.4; P < 0.0001). There was also considerable overlap in values of MELD score among nonsurvivors and survivors (Fig. 1). Table 5 showed comparisons between any 3 CPI, MELD score, and KCH criteria. A MELD score of ≥33 was inferior to the presence of any 3 CPI adversely affecting the outcome (c-statistic [95% confidence interval], 0.717 [0.636–0.789] and 0.802 [0.726–0.878] for MELD and any 3 CPI, respectively; P = 0.045). The specificity, PPV, NPV, and diagnostic accuracy were lower for MELD than any 3 CPI. Addition of other clinical variables, such as presence of cerebral edema, age ≥50 yr, or JEI did not significantly enhance the accuracy of MELD (Table 5). A MELD score of ≥33 was, however, better than the KCH criteria (c-statistics [95% confidence interval], 0.717 [0.636–0.789] vs. 0.676 [0.588–0.764]; sensitivity [76.1% vs. 46.7%], NPV [61.4% vs. 48.4%], and diagnostic accuracy [72.9% vs. 61.8%] for MELD and KCH criteria, respectively). The presence of any 3 CPI was far superior to KCH criteria. These data suggest that the presence of any 3 CPI is superior to MELD and KCH criteria in predicting the outcome in our patients (Table 5).
Table 5. Comparison Between Clinical Prognostic Indicators, MELD and King's College Hospital Criteria
Positive predictive value (%)
Negative predictive value (%)
Diagnostic accuracy (%)
c-statistics (95% CI)
Abbreviations: CI, confidence interval; CPI, clinical prognostic indicators; JEI, jaundice-encephalopathy interval; MELD, Model for End-Stage Liver Disease; KCH, King's College Hospital.
Any 3 CPI
MELD ≥33 + presence of cerebral edema
MELD ≥33 + age ≥50 yr
MELD ≥33 + JEI >7 days
In this retrospective study, we examined the prognostic significance of MELD score and KCH criteria on admission in a cohort of patients with FHF and compared with other early clinical indicators of prognosis. This study demonstrated that any three of 6 independent predictors of outcome (age, JEI, encephalopathy, presence of cerebral edema, prothrombin time, and serum creatinine) identified by multivariate analysis were more useful than a MELD score or KCH criteria.
The MELD score, originally derived to estimate short term survival of patients undergoing transjugular intrahepatic postoperative shunts,23 is used nowadays to prioritize patients with chronic liver disease for LT.25 MELD is a severity score derived from total serum bilirubin, international normalized ratio, and creatinine, which is also found to be independent predictor of survival by multivariate analysis in our cohort of patients with FHF. This study further showed that MELD had comparable sensitivity, but lower specificity, PPV, NPV, and diagnostic accuracy to any 3 CPI. This could be related to the absence of 3 other independent prognostic indicators (age, JEI, and presence of cerebral edema), which are not a part of MELD. However, addition of any of the remaining 3 variables, namely presence of cerebral edema, age ≥50 yr or jaundice, encephalopathy interval >7 days to MELD improves sensitivity at the expense of specificity. The KCH criteria as prognostic indicators are in widespread use. There was similarity between KCH criteria and 6-factor model identified by us; 3 factors (age, JEI, and prothrombin time) were common in both models. However, when KCH criteria were applied to 144 FHF patients seen at our center, we found high specificity (88.5%) and PPV (87.8%) but lower sensitivity (46.7%) and diagnostic accuracy (61.8%) than could be expected from the original report, 91% and 90%, respectively.8 Similar observations have also been reported earlier.12 The area under ROC curve for any 3 CPI was higher than the area under ROC curve for MELD score or for KCH criteria, indicating superiority of any 3 CPI to MELD score and KCH criteria (Fig. 2).
Our study group represents a homogeneous population of patients with acute viral hepatitis. While results of hepatitis B surface antigen were available in all patients, viral markers for other hepatitis viruses were not available in all patients. However, results in a similar population of patients were available in 73 consecutive patients with FHF due to acute viral hepatitis.33 This study showed presence of acute hepatitis B virus marker in 38 (52.1%) patients, anti-hepatitis C virus in 3 (4.1%), anti-hepatitis E virus (immunoglobulin M [IgM]) in 12 (16.4%), anti-hepatitis A virus (IgM) in 3 (4.1%), and dual infection in 2 patients (anti-hepatitis B virus nucleocapsid protein [IgM] + anti-hepatitis E virus [IgM] in 1 and anti-hepatitis C virus + anti-hepatitis E virus [IgM] in another). The remaining 18 (24.7%) patients did not show positivity for any of the above markers. It is possible that there could be potential for error in calculating KCH criteria because of nonavailability of anti-hepatitis A virus (IgM) in all patients, although it is unlikely to affect calculation significantly because of rarity of hepatitis A virus–related FHF in our population.33 Hepatitis A virus accounted only 1.7% of 423 adult FHF cases at a large tertiary health care center in North India,27 which was related to almost universal exposure to hepatitis A virus during childhood in North Indian population.34
The possible prognostic value of the MELD score has been the subject of few recent reports. For the first time Kremers et al.35 demonstrated that the MELD score may be useful in prioritizing FHF without acetaminophen toxicity candidates for liver allocation within the United Network for Organ Sharing status 1 designation. They evaluated the MELD score at listing as a predictor of pretransplant and posttransplant survival in United Network for Organ Sharing 1 patients and compared survival among 4 diagnostic groups, namely, FHF due to acetaminophen, FHF without acetaminophen toxicity, primary graft nonfunction within 7 days of transplantation, and hepatic artery thrombosis within 7 days of transplantation, within the United Network for Organ Sharing status 1 designation using Kaplan-Meier and Cox regression methodology. They demonstrated that the FHF without acetaminophen toxicity had the poorest survival probability while awaiting LT. This was negatively correlated with MELD score (P = 0.0001), which translated into best survival benefit associated with LT.
Three additional studies prognosticating MELD score in patients with FHF are published in an abstract form.36–39 A study from Pittsburgh assessed the applicability of MELD in prognosticating patients with FHF.36 Viral hepatitis was the most frequent etiology (31%). It was demonstrated that the mean MELD score at admission was significantly higher among nonsurvivors compared to survivors and transplanted patients (P = 0.0001). However, there was no significant difference between spontaneous survivors and transplanted patients. The MELD score tended to stay high in patients who were likely to die than patients who were likely to recover. MELD score was therefore, considered to provide a complementary tool additional to other prognostic criteria. In another recent study, Yantorno et al.,37 compared the KCH and Clichy criteria with the MELD score in patients presenting with FHF of various etiologies, excluding acetaminophen toxicity. They demonstrated that the MELD score was an excellent predictor of outcome in both adults and children with a diagnostic accuracy of 95% and a c-statistic score of 0.96. Compared to the KCH and Clichy criteria, a MELD score >30 was associated with lowest rate of false-positive and false-negative results (KCH criteria: PPV = 71%, NPV = 84%; Clichy criteria: PPV = 91%, NPV = 72%; MELD >30: PPV = 91%, NPV = 100%). The data showed that the MELD score is a better predictor of mortality. Our results, however, did not support these observations. Rossaro et al.38 investigated the prognostic value of MELD in 729 adult patients in the U.S. acute liver failure study group cohort; they could not confirm the favorable results as reported in previous reports.35, 38 The patients were divided into 2 groups, namely acetaminophen and nonacetaminophen. In the acetaminophen group, MELD <30 had the greatest applicability in predicting spontaneous survival, while MELD ≥30 did not accurately predict death or LT (NPV = 82%, PPV = 52%). In the nonacetaminophen group, MELD ≥30 had a high PPV (81%), but low NPV (41%). More recently, Pelaez-Luna et al.39 compared MELD with KCH criteria and a new acute liver failure in-hospital mortality score; they found MELD to be inferior to the in-hospital mortality score but superior to KCH criteria.
Inclusion of patients undergoing LT as live positives introduces a potential bias in several reported studies, since it cannot be proved that every patient undergoing LT would have died with continued medical therapy. Our study does not have this kind of bias, since an ideal model must not only be sensitive and specific but must also use those variables that are objective and easy and cheap to measure. The model should be able to identify early in the course of the illness, with high accuracy, not only those who will die without an LT but also those who will survive with medical treatment. The results of our study demonstrated that MELD ≥33, KCH criteria, and any 3 CPI have high PPV (80.5%, 87.8%, and 90.7%, respectively) but low NPV (58.7%, 48.4%, and 65.2%, respectively), which indicates that these models have the greatest applicability in predicting death rather than spontaneous survival. Clearly, future prospective studies are needed to identify additional prognostic indicator(s) to predict the outcome accurately for both spontaneous survival and death.
There are few limitations to this study. First, this is a retrospective study and 7 patients had to be excluded because of incomplete data. The clinical characteristics of these patients were not different from the rest of the patients. Thus exclusions did not appear to introduce bias that would compromise the applicability of the results. Second, results of all viral markers were not available in all patients. Last, a high NPV could not be achieved. Although we assured a large number of clinical variables, there may yet be many variables, such as lactate, pH, phosphates, α-feto protein, etc., which could not be assessed due to the retrospective nature of the study. Thus a prospective collection of data in these patients may be more accurate.
In summary, MELD is less useful in accurately predicting the outcome of FHF patients due to acute viral hepatitis and is inferior to a combination of other clinical indicators of prognosis. KCH criteria has a very low sensitivity, NPV, and diagnostic accuracy. Future prospective studies should identify additional prognostic indicator(s) to predict the outcome accurately, not only those who will die without LT but also for those who will survive with medical treatment.