Before the advent of liver transplantation, there was little incentive for clinicians to develop predictors of the inevitable progression to death for children with end-stage liver disease. Over the last 20 years, the dramatic success of liver transplantation has given both children and adults better than an 80% chance of long-term survival (1). As a consequence, cadaveric organs are increasingly in short supply (2), and the need to rank the order in which patients receive a transplant has become an urgent mandate for transplant physicians.
Until now, the two basic tenants of prioritizing patients for liver transplantation, regardless of age, were first to establish criteria as to when patients should be listed for transplantation (3,4) and second to develop an allocation algorithm based on urgency, waiting time accrued, and the geographic locations of the patient and donor. The principle of prioritizing patients based on urgency has been important in the algorithm since its inception. Currently, four broad urgency statuses for active candidates are defined (2). Relatively few patients are in the most urgent categories (Status 1 and 2A), but large numbers of patients are in the less urgent categories (2B and 3). For example, a “snapshot” of the United Network of Organ Sharing (UNOS) waiting list on February 28, 2001 showed 20 patients listed at status 1, 88 at status 2A, 3,159 at status 2B and 11,594 at status 3. As a result, waiting time has become increasingly more influential in determining liver allocation, particularly at Status 2B. Given that the number of patients added to the liver transplant waiting list continues to far outstrip the relatively constant supply of cadaveric organs, and that 8.2% of patients died on the waiting list in 2000 (2), liver allocation practices have come under intense scrutiny. In particular, the notion that waiting time should be such an influential factor in allocation has been seriously questioned. Freeman et al. (5) have now shown definitively that waiting time has no relationship to death on the waiting list for patients at less urgent statuses.
In 1999, the Institute of Medicine, at the request of the U.S. Department of Health and Human Services (DHHS), reviewed the issue of liver allocation. It was concluded that waiting time was not only a poor predictor of pretransplantation survival but should be replaced by a continuous scale measuring the severity of end-stage liver disease, rather than by broad arbitrary categorizations of urgency (6). DHHS charged the liver transplant community to develop a scoring system that predicted pretransplantation death and could be used to establish a new algorithm for liver allocation.
Currently, the most widely used score for assessing end-stage liver disease is the Child-Turcotte-Pugh score based on the indices ascites, encephalopathy, bilirubin, albumin, and protime (7). Although the Child-Turcotte-Pugh score has been extensively used, it has several drawbacks. First, it was devised to predict the risk of portal hypertensive bleeding in adults; second, both ascites and encephalopathy are subjective assessments; third, the scoring system does not assess a continuum of increasing protime or decreasing albumin; and finally, it is not applicable to children.
Early in 2000, members of the UNOS liver-intestinal and pediatric committees met to develop a new scoring system that would predict pretransplantation mortality based on the severity of end-stage liver disease. At the outset it was agreed that a pediatric severity-of-illness scale would be developed separately to address the special characteristics of children. In particular, the problem of the high annual mortality rate of children <1 year of age on the waiting list (twofold higher than adults 18–50 years of age) had to be considered (2). As well, because adults outnumber children on the list by about 15 to 1, children with a similar severity of illness, but often shorter waiting times, may suffer a competitive disadvantage for access to a rationed resource. The unique problem of the detrimental impact on growth and development for children with chronic liver disease also had to be addressed. Moreover, it was agreed that both the adult and pediatric severity-of-illness scores would hold to common principles: elements of the score would be objective, measurable, and few in number, and the statistical methods for developing and testing the models would be the same. The adult scoring system, MELD (model of end-stage liver disease score), has been described elsewhere (8).
The pediatric end-stage liver disease (PELD) score described herein was developed and tested using the database of the Studies of Pediatric Liver Transplantation (SPLIT). This database, established in 1995, enrolls children at the time of listing for liver transplantation and collects extensive data at listing and every 6 months until the time of transplantation (9). Comprehensive posttransplantation follow-up data are also collected. Data are collected in real time and submitted on line to the coordinating center (The EMMES Corporation), which is responsible for data verification and analysis. As of June 2000, SPLIT had registered 1144 children from 29 centers within both the U.S. and Canada, of whom 706 had already received a liver transplant. Using this data set, we have developed a model using five parameters that predicts poor pretransplantation outcome of children with end-stage liver disease.
MATERIALS AND METHODS
All analyses were for children with chronic liver disease undergoing their first transplantation. The first endpoint used for model development was death before transplantation. For this model, 884 children were evaluated, of whom 41 died. Because of these relatively few events in the first endpoint, and also because of the clinical importance of children becoming critically ill and requiring intensive care while awaiting transplantation, a second composite endpoint was examined. This endpoint was defined as death or admission to the intensive care unit (ICU) for children not in the ICU at listing. For this model, 779 children were not in the ICU at the time of listing, and of these, 74 children either died or were moved to the ICU before receiving a transplant. Data regarding age range, primary diagnosis, gender, race, blood type, patient status at time of listing, number undergoing transplantation, and growth failure are shown in Tables 1 and 2.
Seventeen factors that might be associated with poor outcome before transplantation were initially considered for inclusion in the model. We then selected six variables that met three criteria: objective, verifiable, and not event based (e.g., occurrence of ascites, variceal bleeding) for further analysis. The six factors were age, growth failure, renal function, international normalized ratio (INR), albumin, and total bilirubin.
Both age and growth failure seemed to be important variables affecting outcome. Using Kaplan-Meier estimates of time to event, we graphically examined various age ranges and standardized Z scores for growth failure to determine where the effect for poor outcome was strongest. Age <1 year and height or weight 2 standard deviations below the age-adjusted and gender-adjusted means had the strongest effect on poor pretransplantation outcome (Figs. 1 and 2).
Tables 3 and 4 show the distribution of the mean and median values for glomerular filtration rate (GFR), serum creatinine, albumin, and total bilirubin for the endpoints death or death–moved to ICU, respectively. Renal function was evaluated using a calculated GFR (Schwartz formula) (10) as the variable tested given that serum creatinine itself does not give an accurate measure of renal function in such a diverse population ranging in age and body size from infancy to late teen years.
Cox proportional hazards methodology was used to develop models predicting the two time-to-event endpoints. We first performed univariate Cox regression analysis of all factors, and those found to be significant at the 0.05 alpha level were selected for the multivariate analysis. The final multivariate model was obtained by a backward elimination stepwise selection method with the criteria for factor removal being P ≥0.01. The area under the receiver operating characteristic curve (AUC ROC) was the tool used to evaluate the ability of the severity scores to rank patients according to their risk of death or death–moved to ICU. Although a perfect scoring system would always assign a higher score to a patient who is expected to die by 3 months as compared with another patient who is expected to be alive, in practice the opposite is possible. The AUC ROC measures the probability of correct ranking of such a pair of patients by the scoring system being evaluated. Therefore, AUC ROC ranges from 0 to 1, with 1 corresponding to a perfect scoring system. Severity scores with AUC ROC of ≤0.7 are considered to have very little predictive ability.
At the time of this study, because some centers were still submitting prothrombin times rather than the INR, INR data were missing for 267 patients. The INR is now recognized as the more accurate measure because it removes the factor of interlaboratory variability. INR values were imputed from the prothrombin data for 153 of these 267 patients. Using a cohort of patients for which both INR and prothrombin time data were available, a linear regression model was developed regressing INR on prothrombin time to obtain the missing INR data.
To be consistent with the adult model, the laboratory values were transformed to their natural logarithms. This transformation reduces the influence of extreme observations.
Tables 5 and 6 show the results of the univariate analyses of the six variables: age less than 1 year, GFR, albumin, total bilirubin, INR, and growth failure for the endpoint death before transplantation (Table 5) and death–moved to ICU before transplantation (Table 6). GFR was not a significant predictor for either endpoint. Although growth failure was a strong predictor of death–moved to ICU before transplantation, it lost significance for the endpoint death before transplantation. Although a higher percent of subjects with growth failure experienced the composite outcome, the pretransplantation death rates were similar in patients who had growth failure (5.7%) and those who did not (4.2%).
The results from the multivariate Cox regression analyses of factors found to be significant at the univariate level are shown in Table 7. In the model for predicting death before transplantation, age, total bilirubin, and INR were statistically significant predictors; for the model predicting death–moved to ICU before transplantation, age was not found to be predictive, whereas albumin, growth failure, total bilirubin, and INR were significant predictors. Because age was so strongly associated with predicting death before transplantation, a third model was evaluated to include all of the factors significant for the endpoint death–moved to the ICU before transplantation, as well as age <1 year. The regression coefficients of the predictors in each of the three models (shown in Table 7) were then used as the multiplicative factors to compute the severity scores. Thus, each of the three models leads to a separate severity score. The formulas for the three pediatric severity scores as well as the MELD formula (11) are given in Table 8.
To compare the ability of the four severity scores to predict the two outcomes of interest at 3 months, we computed the AUC ROC. Patients alive at the last contact but with less than 3 months of follow-up were excluded from the analysis, as were patients receiving a transplant less than 3 months after listing. Patients moving to the ICU less than 3 months before transplantation were included in the death–moved to ICU analysis. Since the factors used in each score are not the same, the number of patients and events differ for each score. Table 9 shows the AUC ROC of the three pediatric models as well as MELD. The PELD 2 and PELD 3 models had the best predictive outcomes for both endpoints. The AUC ROC for both PELD 2 (bilirubin, INR, albumin, growth failure) and PELD 3 (bilirubin, INR, albumin, growth failure, and age <1 year) were nearly 10 points higher than the MELD score (bilirubin, creatinine, INR) and 4 to 7 points higher than PELD 1 for all comparisons. All four scores were somewhat better at predicting the endpoint death before transplantation than the composite endpoint death–moved to ICU before transplantation.
Table 10 compares the ability of the same four severity scores to predict the two outcomes of interest at 6 months. The AUC ROC for the PELD models exceeded the MELD model by 10 to 18 points. The performances of the PELD models were similar.
We have also evaluated whether adding events such as variceal bleeding, encephalopathy, or ascites into the model improves the predictive ability of PELD. The result of these analyses showed that the addition of such events to the model has very little effect on the prediction of the two outcomes at the 3-month time point for all four severity-of-illness scores. At best, the AUC ROC increased by 0.03 (AUC ROC for death–moved to ICU before transplantation increased from 0.821 to 0.851 by adding variceal bleeding to the model using the PELD 3 severity score).
We have begun a preliminary analysis examining how well the three PELD models and MELD predict death after transplantation. The four severity scores, calculated at the time of listing, were used as predictors in a Cox regression model. The relative risk of death after transplantation for the four models, shown in Table 11, can be interpreted as the increase in risk of death corresponding to a unit increase in the severity score. The PELD 3 model was the best predictor of death after transplantation, whereas MELD was the least predictive. Other analyses are in progress including whether the prediction of death after transplantation differs depending on whether PELD is calculated at the time of listing or at the time of transplantation.
The severity-of-illness score we have developed for children with end-stage liver disease is the first to be derived and statistically tested in a large multicenter database. The importance of understanding predictors of poor outcome before transplantation was first recognized in 1987 by Malatack et al. (12), who identified four risk factors associated with progressive liver disease before transplantation: history of ascites, indirect bilirubin >6 mg/dL, cholesterol <100 mg/dL, and partial thromboplastin time prolonged >20 sec. Interestingly, although Malatack’s study included a smaller number of patients, we have also confirmed that increasing bilirubin and coagulation defects are important prognostic factors for poor pretransplantation outcome in children.
For a severity-of-illness scoring system to be used as a valid predictor of pretransplantation outcome, it should be developed and tested in a large, representative database of patients awaiting liver transplantation. The database used to develop PELD has several strengths. Data entry begins at the time of listing and is entered in real time rather than retrospectively. Because data entry into SPLIT began in 1995, the database reflects relatively current listing practices for children with end-stage liver disease in the U.S. and Canada. SPLIT is a multicenter registry and represents a broad cross-section of the pediatric liver transplant waiting list nationally. At the time of the PELD analysis, SPLIT was registering approximately 50% of pediatric transplant patients on the UNOS waiting list. Although this falls short of the ideal 100%, and the risk of bias therefore still exists, this level of representation is certainly broader than can be achieved with a single-center study. The pediatric database included children of all diagnoses and all degrees of urgency, covering the full spectrum of illness severity of children with end-stage liver disease. In contrast, the original MELD model was developed to assess survival in a population of patients undergoing the transjugular intrahepatic portasystemic shunt procedure (11). Although MELD has subsequently been convincingly validated in several other data sets, all data sets have had either single-center or diagnostic restrictions of some kind (13). In the most inclusive data set MELD has been tested in, 311 patients on the UNOS waiting list, Status 3 adult patients could not be included for lack of data (14). As a result, direct comparisons between MELD and PELD must be made with caution because the expected distribution of pretransplantation patient survival rates of the more inclusive pediatric database may be different from the adult populations used in developing and testing MELD.
Apart from the six variables selected for the univariate analyses, several other factors are worthy of comment. Children with biliary atresia or cholestatic liver disease represented 46% and 15%, respectively, of the SPLIT database, so that any effect of other diagnoses was lost because of the much smaller numbers in the other diagnostic categories. Consistent with the adult scoring system, we also concluded that diagnosis was not a useful parameter and that its inclusion in the model could be perceived as discriminatory of certain subgroups of patients on the waiting list. However, in contrast to adults awaiting orthotopic liver transplantation, for whom renal insufficiency is an important predictor of pretransplantation mortality (serum creatinine is a component of MELD), in this pediatric population a calculated GFR did not have a statistically significant effect on pretransplantation death or moving to the ICU. Renal failure has previously been shown to be associated with increased mortality after pediatric liver transplantation (15).
The events portal hypertension with bleeding, spontaneous bacterial peritonitis, and hepatic encephalopathy did seem to have lower Kaplan-Meier probabilities of pretransplantation survival. However, on closer examination of the data, we found that the number of patients with the event who reached either endpoint was quite small. For example, 49 children had portal hypertension with bleeding before transplantation, but only 11 reached the most inclusive endpoint, death–moved to the ICU before transplantation. Therefore, the small numbers of patients experiencing the events in question led us to question the validity of including these events in the final model. This concept was also in agreement with the development of MELD, in which the decision was made to exclude such events from the model. We have subsequently shown, just as was demonstrated in a further analysis of MELD, that including events did not improve the predictability of PELD. This finding reaffirms clinical experience that poor outcome after an event such as variceal bleeding is more contingent upon the degree of end-stage liver disease at the time of the event than the occurrence of the event itself. Miga et al. (16) recently reported that in children with biliary atresia, the serum bilirubin concentration at the time of a first esophageal variceal bleed had a significant impact on outcome.
In contrast to the single endpoint, probability of death before transplantation, chosen to evaluate MELD, we felt that a second composite endpoint, death/moving to the ICU before transplantation, was clinically important and should be simultaneously analyzed. Although the requirement for ICU care has not been well studied for predicting pretransplantation death, it is well established that ICU care before transplantation is one of the strongest predictors of posttransplantation survival both in children and adults undergoing liver transplantation (17). As a consequence, one of the most pressing clinical goals for transplant physicians is to try to optimize the chances that patients with end-stage liver disease will undergo transplantation before they become critically ill and require intensive care. The beneficial effects of achieving this objective were recently reported by Freeman et al. (18). In their experience, a regional variance of the allocation system, which prioritized potential recipients by severity of illness rather than waiting time, resulted in more patients receiving transplants before requiring ICU care, with a decrease in mortality on the transplant waiting list.
Growth failure and very young age are unique characteristics of children with end-stage liver disease. In the multivariate analysis evaluating the two endpoints, death before transplantation and death–moved to ICU, age was a highly statistically significant predictor of pretransplantation death. This was not surprising because children <1 year still have the highest annual death rate on the UNOS waiting list, despite the use of living donors and partial liver transplants. In the UNOS database for 1999, the annual death rate (per 1000 patient years at risk) for children <1 year old was 234.4 as compared with 113.4 to 119.4 for adults 18 to 50 years old (2). Similarly, growth failure, although more predictive in the composite event model (death–moved to ICU) than age <1 year, was not found to be predictive when death was the only outcome. Correlation between the factors contributes to difficulties in interpretation. Children <1 year of age had the highest growth deficits and also the highest number of deaths before transplantation. It is also possible that the relative importance of these two factors may change as the database matures. Our decision to evaluate a model including all parameters significant for either endpoint used the totality of information currently available and was supported by the high statistical significance each variable had in its respective endpoint. As well, there is convincing clinical evidence that young age (19) and malnutrition (reflected by growth failure and to some extent serum albumin) effect liver transplant outcome in children (20,21). When we demonstrated that the predictability of the five-parameter model for either endpoint was better than the model including only age, bilirubin, and INR and comparable to the model excluding age, we felt justified in recommending that the most clinically relevant model, which maintains statistical validity for a severity-of-illness score for pediatric end-stage liver disease, should include all five parameters.
There are several caveats to be considered when using PELD to predict pretransplantation death. First, the number of events for fulfilling either endpoint in our analysis were relatively small. Second, although the statistical validation of the model using AUC ROC showed a very high predictive ability, this result may be positively influenced because the model was validated in the same database from which it was derived. Validating PELD in a separate database is essential and currently underway using the Pittsburgh Pediatric Liver Transplant database. Preliminary data suggest that the AUC ROC for PELD 3 is >0.80. As well, the model must now be evaluated prospectively. This can be done in the SPLIT database and potentially in the UNOS database if the appropriate data are collected. However, the results will be influenced by the ongoing transplantation of patients under current UNOS rules, which continues to use waiting time as an important parameter in allocating organs.
Several observations are relevant to the potential use of PELD to determine the rank order in which children with chronic liver disease are allocated cadaveric organs. In the current model, PELD is calculated at the time of listing. How PELD changes with time may be an important additional predictor of pretransplantation and posttransplantation survival. Regional differences can be expected to have an effect on the use of PELD in an allocation algorithm. The same PELD score at listing may result in a different probability of pretransplantation death between the UNOS geographic regions. Characteristics that may differ by region are the number of children on the waiting list and their overall severity of end-stage liver disease, both of which may reflect individual centers’ listing practices and experience, and organ procurement rates. Finally, PELD has been developed in a model considering only a first liver transplantation. It remains to be validated for patients awaiting retransplantation.
Other patients with unique characteristics will also require individualized treatment. This will be particularly important for pediatric patients with metabolic diseases. Children with metabolic diseases comprise about 12% of the SPLIT database and were included in the development of PELD because more that half presented with end-stage liver disease secondary to alph-1-antitrypsin deficiency, Wilson’s disease, or cystic fibrosis. Other children without evidence of chronic liver disease who are at risk for life-threatening extrahepatic organ involvement—particularly children susceptible to hyperammonemic crises causing profound central nervous system damage, patients with hepatopulmonary syndrome, and children with unresectable liver tumors—will require special consideration. For these exceptional cases the regional review boards will assign an appropriate score to facilitate timely transplantation. It must further be recognized that the score assigned that corresponds to the estimated probability of death before transplantation will be a judgment decision and will vary by region.
In November 2000, the UNOS Board approved both MELD and PELD as the severity-of-illness scores to be used for cadaveric donor liver allocation. However, because MELD and PELD are computed from different equations, it is clear that the actual numeric score for a comparable severity of illness will not be the same for a child as an adult. For example, a MELD score of 40 and a PELD score of 20 both might both predict the same probability of death at 3 months. Because the allocation sequence will first rank by a score, and not differentiate between a pediatric and adult recipient, if the desired goal is to match the probability of death before transplantation between adults and children, then a method of making the two scores equivalent would be required. Although it is mathematically valid to develop an equation that would convert a PELD score to a MELD score or vice versa, such a conversion makes the important assumption that both scoring systems were developed in a similar and representative population of patients, allowing us to compute the respective baseline survival function. As noted above, this is not the case for MELD, which was developed in patients undergoing a transjugular intrahepatic portasystemic shunt procedure. Currently underway is a prospective study of MELD and PELD scores from patients on the UNOS waiting list. This information should be more representative of the actual population of patients awaiting liver transplantation nationally.
A pediatric severity-of-illness score has been developed and validated in a large multicenter data set of children awaiting liver transplantation. The five-parameter score is based on objective verifiable elements. Once the equation is programmed into a handheld calculator or computer, the score can be easily calculated. It is proposed that PELD, along with its adult counter part, MELD, will be used to devise a new liver allocation system based on severity of chronic liver disease rather than waiting time. It is essential to understand that after implementation of PELD and MELD, rigorous ongoing evaluation will be required, particularly in view of the many exceptions to PELD and MELD that can be anticipated. The effect on posttransplantation survival of a national policy that de-emphasizes waiting time and directs organs to the sickest patients first is unknown and will require careful analysis. Public perceptions and ethical considerations (22) of a policy emphasizing “sickest first” have not been well evaluated. In a recent study of public attitudes it was found that the patient prognosis is an important consideration in allocating a resource in short supply (23). Whether the overall cost of liver transplantation will increase if sicker patients are prioritized will need to be carefully evaluated (24). However, both MELD and PELD have been carefully developed, statistically evaluated, and accepted by a consensus of transplant professionals, and it is appropriate to move forward cautiously. It is not until these severity-of-illness scores are implemented as the new liver allocation system, that a more accurate evaluation of the true predictive value will be possible.
1. Goss JA, Shackleton CR, McDiarmid SV, et al. Long-term results of pediatric liver transplantation: an analysis of 569 transplants. Ann Surg 1998; 228: 411.
2. Annual Report the US Scientific Registry of Transplant Recipients and the Organ Procurement and Transplantation Network: Transplant Data 1990-1999. US Department of Health and Human Services, 2000.
3. McDiarmid SV, Millis MJ, Olthoff KM, So SK. Indications for pediatric liver transplantation. Pediatr Transplant 1998; 2: 106.
4. Lucey MR, Brown KA, Everson GT, et al. Minimal criteria for placement of adults on the liver transplant waiting list: a report of a national conference organized by the American Society of Transplant Physicians and the American Association for the Study of Liver Diseases [see comments]. Liver Transpl Surg 1997; 3: 628.
5. Freeman RB, Edwards EB. Liver transplant waiting time does not correlate with waiting list mortality: implications for liver allocation policy. Liver Transpl 2000; 6: 543.
6. Organ Procurement and Transplantation. Assessing current policies and potential impact of the DHHS final rule. Committee on Organ Procurement and Transplantation Policy, Division of Health Sciences Policy, Institute of Medicine. Washington, DC: National Academy Press, 1999: 75.
7. Pugh RNH, Murray-Lyon IM, Dawson JL, Pietrony MC, Williams R. Transection of the esophagus for bleeding esophageal varices. Br J Surg 1973; 60: 649.
8. Kamath PS, Wiesner RH, Malinchoc M, et al. A model to predict survival in patients with end-stage liver disease. Hepatology 2001; 33: 464.
9. SPLIT Research Group. Studies of Pediatric Liver Transplantation (SPLIT): year 2000 outcomes. Transplantation 2001; 72: 463.
10. Schwartz GJ, Haycock GB, Edelman CM Jr, Spitzer A. A simple estimate of glomerular filtration rate in children derived from body length and plasma creatinine. Pediatrics 1976; 58: 259.
11. Malinchoc M, Kamath PS, Gordon FD, Peine CJ, Rank J, Ter Borg PC. A model to predict poor survival in patients undergoing transjugular intrahepatic portosystemic shunts. Hepatology 2000; 31: 864.
12. Malatack JJ, Schaid DJ, Urbach AH, et al. Choosing a pediatric recipient for orthotopic liver transplantation. J Pediatr 1987; 111: 479.
13. Kim WR, Kamath PS, Benson JT, Kremers WK, Dickson ER, Wiesner RH. The model for end stage liver disease (MELD) in the prediction of survival in patients with liver cirrhosis [Abstract]. Am Soc Transplant Phys 2001; 1: 283.
14. Wiesner RH, Edwards EB, Kamath PS, et al. Mayo end stage liver disease model (MELD) score predicts liver transplant waiting list mortality: implications for liver allocation policy [Abstract]. Am Soc Transplant Phys 2001; 1: 461.
15. Ellis D, Avner ED, Starzl TE. Renal failure in children with hepatic failure undergoing liver transplantation. J Pediatr 1986; 108: 393.
16. Miga D, Sokol RJ, MacKenzie T, Narkewicz MR, Smith D, Karrer FM. Survival after first esophageal variceal hemorrhage in patients with biliary atresia. J Pediatr 2002 (in press).
17. Belle SH, Beringer KC, Detre KM, eds. An update on liver transplantation in the United States: recipient characteristics and outcome. UNOS liver registry. Clinical Transplants 1995: Pitsburgh 19.
18. Freeman RS, Rohrer RJ, Katz E, et al. Preliminary results of a liver allocation plan using a continuous medical severity score that de-emphasizes waiting time. Liver Transpl 2001; 7: 173.
19. Colombani PM, Cigarroa FG, Schwarz K, Wise B, Maley WE, Klein AS. Liver transplantation in infants younger than 1 year of age. Ann Surg 1996; 223: 658.
20. Shepherd RW, Chin SE, Cleghorn GJ, et al. Malnutrition in children with chronic liver disease accepted for liver transplantation: clinical profile and effect on outcome. J Paediatr Child Health 1991; 27: 295.
21. Moukarzel AA, Najm I, Vargas J, McDiarmid SV, Busuttil RW, Ament ME. Effect of nutritional status on outcome of orthotopic liver transplantation in pediatric patients. Transplant Proc 1990; 22: 1560.
22. Yoshida EM. Selecting candidates for liver transplantation: a medical ethics perspective on the micro allocation of a scarce and rationed resource. Can J Gastroenterol 1998; 12: 209.
23. Ubel PA, Loewenstein G. Public perceptions of the importance of prognosis in allocating transplantable livers to children. Med Decis Making 1996; 16: 234.
24. Showstack J, Katz PP, Lake JR, et al. Resource utilization in liver transplantation: effects of patient characteristics and clinical practice. JAMA 1999; 281: 65.