Prediction of survival for patients with fulminant hepatic failure


  • See Article on Page 26

  • Conflict of interest: Nothing to report.

The mortality of patients with advanced fulminant hepatic failure (FHF) is high, and although liver support devices may in time have a place in treatment, liver transplantation remains the only therapeutic option that has been shown to improve outcome. Both selection of patients and timing of transplantation remain areas of clinical practice where, without being overly melodramatic, literally life and death decisions have to be made often on the basis of clinical judgment and usually with inadequate information and too little time.


FHF, fulminant hepatic failure; AFP, alfa-fetoprotein; INR, international normalized ratio; ALT, alanine aminotransferase.

The decision to offer liver transplantation to a patient with FHF cannot be based solely on a balance of survival probabilities with and without transplantation. Both the quality and length of life after transplantation are excellent, but neither is normal. In those with advanced FHF who recover with just medical support, the liver almost always returns to normal, both structurally and functionally. In contrast, those who are given an orthotopic graft are likely to require lifelong and life-limiting immunosuppression with well-recognized consequences, both physical and psychological. For obvious reasons the patient usually cannot participate in the discussions and give informed consent for the procedure, so decisions must be made on their behalf. Not all recipients welcome the survival after a transplant, and a few either stop immunosuppression or repeat their overdose. Therefore, the decision to offer transplantation must be made not just on the balance of survival with and without transplantation; there must be a significant benefit in survival and an expectation of a quality of life acceptable to the recipient.

The use of auxiliary (orthotopic or heterotopic) liver transplantation has many theoretical benefits because this approach does not require the removal of the native liver; if the host liver does recover, the donor liver can be removed or allowed to atrophy and the recipient is spared the consequences of lifelong immunosuppression. This option may be valuable, especially when there is uncertainty regarding whether the liver will recover. However, relatively few such procedures are undertaken, in part1 because of the poor results. Combination of auxiliary orthotopic transplantation with a subtotal hepatectomy may prove of value in those with a toxic cause of liver failure.2

There are self-evident benefits of a robust model that will identify those patients with FHF who will require liver transplantation at a stage which leaves adequate time to find a suitable liver (from either a living or deceased donor) and before the onset of irreversible complications that will preclude a successful outcome. Any such model must not only be sensitive and specific but must also use those variables that are objective and easy and cheap to measure. Ideally, any model would be able to identify early during the course of the illness, with high accuracy, not only those who will die without a transplant but also those who will survive without surgical intervention. Furthermore, since the patient's prognosis will change with the onset of complications such as sepsis, cerebral edema, or vascular instability, such models should be time dependent.

For the statisticians, FHF offers an almost unlimited number of clinical and laboratory variables to include in the derivation of a clinical model: most analytes either increase or decrease in those with FHF and often the sicker the patient, the worse the outcome and the more abnormal the analyte. Not all laboratories measure the components of the prognostic models in the same way, and the choice of laboratory may affect the estimate of one patient's prognosis.

Application of prognostic models to populations is usually associated with satisfactory accuracy, but application to the individual is more problematic because confidence intervals are usually wide. The limitations of the clinical utility of prognostic models has been discussed by Christensen,3 who identified a number of factors including the fact that statistical models do not use the full information in the prognostic variables (a restriction that could be overcome with the use of neural networks) and that the prognostic variables used are not sufficiently informative. Selection of variables used in prognostic models is, in practice, based on those variables that are routinely collected, and some subjectivity is used in selection of those variables identified in the initial analysis as being of statistical significance. The analysis will be based on historical controls and, even if validated prospectively, may not take into sufficient account changes in clinical care. The recent advances in the care of patients with fulminant hepatic failure include better intensive care, possible use of hypothermia, and liver support devices.

It must also be remembered that the decision to proceed to transplantation involves a balance not only of the probability that the patient will die from liver failure but also an estimation of the probability of whether the patient will survive after transplantation. While there are several models that predict survival after transplantation, these are usually derived for those with chronic rather than acute liver disease. These models do show that while the patient's condition is a relatively weak factor in predicting outcome, factors such as the cold ischemia time and degree of steatosis in the graft are important determinants. Again, most prognostic models give estimates for survival in the short-term—say, 1-year survival— but for a 20-year-old with liver failure, 30- or 40-year survival estimates are also needed.

The first model to identify patients who are likely to die as a consequence of their FHF was developed at King's College Hospital and published in 1989.4 Based on a retrospective analysis of 588 patients treated between 1973 and 1985, the model identified that there were important prognostic differences between those with acetaminophen-associated liver failure and those with other aetiologies. Other prognostic markers identified by the multivariate analysis included serum bilirubin, creatinine, and prothrombin time. The model has been remarkable in that it has withstood the test of time and most units still use the King's criteria as a basis for selection of patients for liver transplantation. The model has been refined and both the specificity and sensitivity have been improved, for example, with inclusion of lactate.5 Estimates suggest that the King's criteria have a positive predictive accuracy of around 0.8.6, 7 Others have produced models of broadly similar sensitivity and specificity: for example, Dabos and colleagues8 evaluated 59 patients with FHF not associated with acetaminophen and showed using nuclear magnetic resonance spectroscopy that a model using albumin, lactate, valine, and pyruvate when applied within 6 hours of admission would predict survival with reasonable accuracy (a positive predictive value of 91%, a negative predictive value of 94%, sensitivity of 94%, and specificity of 86% for transplantation). Use of the APACHE II model is of similar sensitivity and specificity as the King's Criteria,9 but such a model cannot be applied as soon after admission as the King's criteria. The MELD score, originally derived to estimate short-term survival of patients undergoing transjugular intrahepatic shunts, is derived from transformation of the serum bilirubin, international normalized ratio (INR), and creatinine: this model is well validated to predict survival for patients with cirrhosis and may also predict outcome for those with FHF.10

The King's Criteria use etiology, age, serum creatinine, bilirubin, degree of encephalopathy, and arterial pH as prognostic variables. Other markers are listed in Table 1.11–19 Some may be of little importance and merely reflect that the sicker the patient, the worse the outcome and the more abnormal the measurement, so it is not surprising there would be a correlation with outcome. Some markers, such as the degree of liver cell necrosis and liver size, may well have the necessary statistical power to show a significant correlation with survival, but they are either impractical to measure or subject to significant observer variation.

Table 1. Variables That Correlate With Outcome in Patients With Fulminant Hepatic Failure
 Degree of encephalopathy
 Rate of development of encephalopathy
 Intracerebral hypertension
 Clotting variables, such as INR, Factor V19
 Ketone Body ratio18
 Gc protein16
 MELD score10
FunctionalGalactose elimination capacity15
PhysiologicalAPACHE II9
 Hepatic artery resistance index changes13
HistologicalDegree of liver cell necrosis14
MorphologicalLiver volume17

The prognostic value of serum alfa-fetoprotein (AFP) in patients with FHF was suggested nearly 3 decades ago,12 where it was shown that in a very small series of patients, survivors had a higher serum AFP (>50 ng/mL) and a rising serum AFP and this could help identify patients with a poor prognosis. A subsequent study20 confirmed these observations but suggested that the higher levels, which reflected hepatic regenerative activity, may reflect merely a temporal effect. Other studies have drawn attention to the changes in serum AFP during the course of the illness.22–24

The reason for the correlation between serum AFP and outcome is not clear: Sakurai et al. showed there was a statistically significant correlation between serum AFP and liver volume21 and serum levels of AFP are likely to relate to hepatocyte regeneration.

In this issue, Schmidt and Dalhoff22 report their prospective study of 239 patients with fulminant hepatic failure following acetaminophen overdose and show that AFP is a useful marker of prognosis: a level greater than 4 μg/L occurred in more than three quarters of those who survived compared with one third who died, and the increase in AFP levels tended to occur sooner after the transaminases peak in survivors than in those who died. Thus, using as a cutoff a serum AFP level of 3.9 μg/L one day after the peak alanine aminotransferase level was highly sensitive at identifying those who were likely to survive.

How useful serial measures of AFP will be in selecting patients for transplantation is uncertain, but these data suggest that this is a promising approach: clearly, it will be important to validate these observations in another cohort of patients, preferably in another center. For practical purposes, it will be necessary to have the ready availability of a highly sensitive method for measuring AFP; here the authors used enzyme-linked immunoassay. Relating the AFP level to the peak ALT level has an attractive simplicity and obviates the variability in relating the AFP to either the time after overdose (which may not be known, may be unreliable, or may be irrelevant in the case of a staggered overdose) or after admission (this time point, used for application of the King's criteria, will vary within and between centers); however, knowing when the peak level has been achieved will require a delay of at least 24 hours. The authors make the point that there was a fall in the INR observed in the survivors but not in those who died, and this contrasts with our own experience where one third of those who died with FHF did so in the context of a falling INR.

Current US 1-, 3-, and 5-year survival rates for patients grafted for fulminant hepatic failure are 80.3%, 73.7%, and 69.8%. These rates are lower than those for patients grafted for chronic liver disease (5-year patient survival for cholestatic cirrhosis and noncholestatic cirrhosis are 80.2% and 70.6%, respectively).23 Those factors that predict survival without a transplant are not the same as those that predict survival after transplantation. Using multivariate analysis of variables on the UNOS OPTN Database, Barshes and Lee24 identified four factors that predicted 5-year survival: recipient age greater than 50 years, body mass index above 29 kg/m2, a history of life support, and serum creatinine greater than 2.0 mg/dL. Five-year survival ranged from 47% for those with all four factors and 83% for those with none. Others have drawn attention to the importance of the quality of the graft in affecting the outcome.25

The underlying assumption of the treatment of patients with FHF is that physiological functions should be supported until the liver regenerates sufficiently to maintain vital function.26 There is an assumption that the liver will regenerate given sufficient time, and this may not be the case in all patients. Treatment is therefore supportive, aimed at replacing the defective liver function (whether synthetic, metabolical, or immunological), and treating the consequences of the necrotic liver. Although treatment for patients with liver failure is continuing to improve, there is no highly effective liver support device: current treatments are showing promise but act primarily to stabilize the patient for a short period. The use of liver replacement is a crude but effective treatment for some patients. The logistic problems in finding a suitable donor before the onset of irreversible complications that would make the procedure pointless have highlighted the need for robust models that use readily available markers. The King's criteria remain the most effective to date; modification by inclusion of other factors such as lactate and AFP may allow refinement and so lead to a more precise identification of those who would benefit the procedure, but it must be stressed that these models will aid, but not replace, clinical judgment.