Pediatric liver retransplantation: Outcomes and a prognostic scoring tool

Authors

  • Adam Davis,

    Corresponding author
    1. Department of Pediatrics, University of California at San Francisco, San Francisco, CA
    • Department of Pediatrics, University of California at San Francisco, 500 Parnassus Avenue, MU4E, San Francisco, CA 94143
    Search for more papers by this author
    • Telephone: 415-476-5892; FAX: 415-476-1343

  • Philip Rosenthal,

    1. Department of Pediatrics, University of California at San Francisco, San Francisco, CA
    2. Department of Surgery, University of California at San Francisco, San Francisco, CA
    Search for more papers by this author
  • David Glidden

    1. Department of Epidemiology and Biostatistics, University of California at San Francisco, San Francisco, CA
    Search for more papers by this author

Abstract

Nine to twenty-nine percent of pediatric liver transplant recipients require retransplantation. No previous pediatric study has proposed a prognostic scoring system. We have used the United Network for Organ Sharing transplantation database to conduct a retrospective cohort study of patients who were less than 18 years of age when they received their retransplant (n = 1130). Using a random two-thirds of the subjects, we developed a prognostic scoring system by performing a multivariate Cox analysis with non-laboratory clinical characteristics. The scoring system was verified in the remaining one-third of the subjects. Stratifying the verification group into risk groups by prognostic score demonstrated its predictive value. Those in the low-risk category had survival similar to that of primary liver transplant recipients. Those in the high-risk category had 2.4 (95% confidence interval: 1.6-3.7) times the risk of death or retransplantation as those in the low-risk category. Risk factors in the scoring system included being on life support at the time of retransplant, receiving a split liver graft, and having an original diagnosis of neonatal cholestasis, familial cholestasis, paucity of bile ducts, or congenital abnormalities. Protective factors in the scoring system included older age at the time of transplantation and having acute rejection contribute to graft failure. In conclusion, with simple clinical characteristics, this scoring tool can modestly discriminate between those children at high risk and those children at low risk of poor outcome after liver retransplantation. Liver Transpl 15:199–207, 2009. © 2009 AASLD.

Hepatic retransplantation is the only option for survival when a transplanted liver fails. Retransplantation accounts for 10% to 22% of liver transplants worldwide across all ages,1 and in previous single-center series, 9% to 29% of pediatric liver transplant recipients were eventually retransplanted.2–9 Given the scarcity of donor organs, the ethics of retransplantation have been controversial.10, 11 In pediatrics, the more frequent use of split organs and living related donors somewhat attenuates the zero-sum game dynamics of organ donation. Nonetheless, understanding the outcomes of retransplantation in children is paramount for caring for these complicated patients and for deciding how to ethically distribute the scarce resource of donated organs.

The outcomes of retransplantation have been examined in greater depth in adults than in children. Prognostic models have been created for adults undergoing retransplantation,1, 10, 12–14 but none has been created for children. In at least 1 previous series, it has been shown that pediatric retransplant patients have better outcomes than adult patients.15 Many series have evaluated the outcomes of children undergoing retransplantation at single centers, and recently a single multicenter study involving the Studies of Pediatric Liver Transplantation (SPLIT) registry was published.3, 8, 9, 16–22 The single-center reports lacked a sufficient number of patients to perform complex predictive modeling. The largest contained only 92 children who were retransplanted.9 The SPLIT study included 246 retransplants. It evaluated predictive factors but did not create a prognostic model. We have used the United Network for Organ Sharing (UNOS) transplant database in order to maximize the number of pediatric retransplant patients in our analysis and enable us to perform prognostic modeling.

In this article, we evaluate the outcome of pediatric liver retransplantation and propose a prognostic scoring system to risk-stratify patients by clinical characteristics.

Abbreviations

CI, confidence interval; HR, hazard ratio; ROC, receiver operating characteristic; SD, standard deviation; SPLIT, Studies of Pediatric Liver Transplantation; TPN, total parenteral nutrition; UNOS, United Network for Organ Sharing.

PATIENTS AND METHODS

Subjects

The UNOS liver transplant database was examined to retrospectively identify all patients who were less than 18 years old at the time of primary liver transplantation between September 30, 1987 and September 1, 2006. Patients who had multivisceral transplants were excluded from analysis. The remaining patients were included in a 2-part analysis.

Comparative Analysis of Successive Transplantations

In this analysis, re-transplant-free survival of primary liver transplants was compared to successive retransplantations. Retransplantations were included if they occurred prior to the subject's 28th birthday. A univariate robust Cox proportional hazards regression model23 clustered by the individual subject was used, and corresponding Kaplan-Meier curves were calculated.

Retransplantation Prognostic Scoring

A prognostic scoring system for liver retransplantation in children was developed. Only first retransplants that occurred prior to the subject's 18th birthday were included in the analysis. The subjects were randomly divided into 2 groups. Two-thirds were placed in the modeling group, which was used to develop the scoring system, and one-third was placed in the validation group, which was used to validate the scoring system. The characteristics of the 2 groups were compared with t tests for continuous variables and chi-square testing for categorical variables.

The scoring system that we developed was created by using a technique similar to one used by Dickson et al.24 in primary biliary cirrhosis. Clinical variables that would have been known at the time of the retransplantation decision and that were complete in the UNOS database were used. Univariate Cox proportional hazard analysis25 of transplant-free survival was performed for each of these variables. We then created a multivariate Cox proportional hazards model. Variables were added one at a time in a forward selection process. We chose the variables to include by examining their P values in the univariate analyses. We began with the characteristic that had the smallest P value and added subsequent characteristics to the model in order of increasing P value. Variables stayed in the model if their P value or the P value of 1 of their categories was less than 0.3. Characteristics that did not meet this criterion were not included in the final analysis.

In a Cox model, each patient is given a risk score. This risk score is modeled as R = X1β1 + X2β2 + … + Xkβk, where X1 to Xk are the values of the variables for k characteristics and β1 to βk are the regression coefficients. For example, if X2 were “being male,” then a male would have a β2 higher risk score than a female patient with otherwise identical characteristics.

This property allowed us to use the coefficients to create a prognostic scoring system. We included the coefficients from our model that had a P value less than 0.05. We transformed those coefficients for simplicity of use by multiplying each of them by 2 and rounding them to the closest integer. Those transformed integers became our prognostic scoring system. The era of transplant was not included in the prognostic scoring system as this would not help to discriminate the prognosis of future patients undergoing retransplantation.

We applied the prognostic scoring system to each patient by adding the values for each of their specific characteristics. This gave an overall prognostic score for each individual.

In order to validate this score, the validation group was analyzed. The validation group was stratified into 3 categories by prognostic score: high risk, medium risk, and low risk. A univariate Cox proportional hazards model and a Kaplan-Meier survival analysis were performed by prognostic score risk groups. The area under the receiver operating characteristic (ROC) curve for the ability of the prognostic score to predict outcome was calculated in the modeling and validation groups. Statistical significance was defined as a P value < 0.05. All statistical analysis was performed in STATA 9.0.

RESULTS

Subjects

According to the UNOS database, 8649 pediatric primary liver transplants were performed between January 1, 1989 and September 1, 2006. Nearly 15% (14.7%) of the patients underwent at least 1 retransplantation. A similar percentage of twice transplanted patients received a third transplant (21.1%), and a similar percentage of thrice transplanted patients received a fourth (18.7%). Table 1 summarizes the subjects' demographics, retransplantation history, original diagnoses, type of transplant, and status at the time of database analysis.

Table 1. Patient Characteristics of Pediatric Liver Transplant Recipients from the United Network of Organ Sharing Database (January 1, 1989–September 1, 2006)
Patient Characteristic (n = 8649)n or MeanSD or (%)
  1. Abbreviations: SD, standard deviation; TPN, total parentenal nutrition.

Age (years)4.7±5.5
Male4175(48)
Year of initial transplant  
 1989–19942923(34)
 1995–19992354(27)
 2000–20063372(39)
Ethnicity  
 White/Caucasian5186(61)
 Black/African-American1502(18)
 Hispanic/Latino1460(17)
 Asian291(3)
 Other/unknown210(2)
Total number of liver transplants  
 17375(85)
 21052(12)
 3187(2)
 432(0.4)
 53(0.03)
Indication for primary transplantation  
 Biliary atresia3356(39)
 Metabolic disease1068(12)
 Fulminant failure of unknown etiology868(10)
 Neonatal or familial cholestasis/paucity of ducts/congenital abnormalities820(10)
 TPN cholestasis582(7)
 Autoimmune460(5)
 Malignancy328(4)
 Cryptogenic cirrhosis285(3)
 Viral hepatitides218(3)
 Cystic fibrosis158(2)
 Drugs and toxins107(1)
 Vascular abnormalities37(0.4)
 Other349(4)
Recipient liver organ type  
 Whole6173(71)
 Reduced1724(20)
 Split752(9)
Last known patient status  
 Follow-up time (years)5.5±4.9
 Alive6583(76)

Comparative Analysis of Successive Transplantations

Transplant-free survival was significantly better after the initial liver transplant than with retransplantation. Patients undergoing their first retransplantations were 1.9 [95% confidence interval (CI): 1.7-2.0] times more likely to die or be retransplanted than those undergoing primary liver transplantation. Transplant-free survival for primary transplants at 10 years was similar to transplant-free survival for first retransplants at 6 months (65%). The hazard ratio (HR) continued to increase with each successive retransplant, although the CIs overlapped. Table 2 summarizes the comparative analysis, and Fig. 1 shows the transplant-free survival curves.

Table 2. Hazard Ratios and Accumulative Event-Free Survival at Various Time Intervals by Liver Transplant Number
Transplant NumbernHazard Ratio (95% CI)Event-free Survival [% (95% CI)]
6 Months1 Year5 Years10 Years
  1. Abbreviation: CI, confidence interval.

Primary transplant8649Reference80 (79–81)77 (77–78)70 (69–71)65 (64–67)
Retransplantation      
 1st12741.9 (1.7–2.0)65 (62–67)60 (58–63)50 (48–53)46 (43–49)
 2nd2222.3 (2.0–2.7)58 (51–64)56 (49–62)45 (38–51)35 (28–43)
 3rd353.1 (2.1–4.7)48 (31–64)45 (29–61)27 (12–43)27 (12–43)
Figure 1.

Pediatric liver retransplantation: transplant-free survival by successive transplantations.

Retransplantation Prognostic Scoring

No significant differences between the characteristics of the modeling group (n = 740) and those of the validation group (n = 390) existed. Table 3 summarizes the characteristics of each group.

Table 3. Comparison of Pediatric Liver Retransplant Patients Randomly Assigned to the Modeling Group and Validation Group
Patient CharacteristicsMean or (%)P Value
Modeling GroupValidation Group
(n = 740)(n = 390)
  1. Abbreviation: TPN, total parenteral nutrition.

Age (years)5.25.20.89
Male(47)(45)0.46
Ethnicity  0.51
 White/Caucasian(60)(57) 
 Black/African American(18)(22) 
 Hispanic/Latino(16)(15) 
 Asian(4)(3) 
 Other/unknown(2)(2) 
Era of transplant   
 1987–1994(42)(44)0.56
 1995–1999(27)(24) 
 2000–2006(31)(32) 
Indication for primary transplantation  0.38
 Biliary atresia(38)(40) 
 Fulminant failue of unknown etiology(9)(9) 
 Metabolic disease(8)(9) 
 Autoimmune(5)(4) 
 Neonatal or familial cholestasis/paucity of ducts/congenital abnormalities(7)(7) 
 Vascular abnormalities(6)(4) 
 Viral hepatitides(4)(3) 
 Cryptogenic cirrhosis(3)(5) 
 TPN cholestasis(1)(1) 
 Malignancy(3)(1) 
 Cystic fibrosis(1)(1) 
 Drugs and toxins(1)(1) 
 Other(14)(14) 
Contributing factors to graft failure   
 Vascular thrombosis(37)(33)0.13
 Primary graft failure(22)(24)0.47
 Chronic rejection(11)(11)0.98
 Biliary tract complication(7)(9)0.20
 Infection(7)(7)0.75
 Acute rejection(5)(7)0.12
 Recurrent disease, nonhepatitis(1)(1)0.49
 Recurrent hepatitis(1)(1)0.53
 De novo hepatitis(1)(0)0.15
Wait list time (days)87950.66
Time since primary liver transplant  0.67
 <1 week(25)(23) 
 1 week to <30 days(26)(27) 
 30 days to <1 year(23)(22) 
 1 year to <5 years(18)(21) 
 5 years or more(7)(6) 
Patient on life support at time of retransplantation(49)(50)0.80
Recipient liver organ type  0.50
 Whole(74)(76) 
 Reduced(16)(15) 
 Split(10)(8) 
Last known patient status   
 Follow-up time (years)4.24.60.26
 Alive(57)(56)0.78

In a univariate analysis, older age, more recent era of transplantation, chronic rejection as a contributor to graft failure, prolonged wait-list time, and having one's retransplant more than 5 years since the initial transplant were all significantly protective for the survival of the retransplanted liver. An original diagnosis of neonatal cholestasis, familial cholestasis, paucity of ducts, congenital abnormalities, or total parenteral nutrition cholestasis was a significant risk factor, as was being on life support at the time of retransplantation or receiving a split liver graft. Many other characteristics showed a strong trend of effect but did not reach statistical significance.

In multivariate analysis, fewer characteristics retained statistical significance (Table 4). Being originally diagnosed with neonatal cholestasis, familial cholestasis, paucity of ducts, or congenital abnormalities led to an HR of 1.7 (95% CI: 1.2-2.4). Having a retransplant during the most recent era of transplant was protective [HR: 0.52 (95% CI: 0.38-0.70)], as was being at least 5 years old at the time of retransplantation [HR: 5 years to <12 years, 0.58 (95% CI: 0.34-0.98); >12 years, 0.55 (95% CI: 0.31-0.98)]. Being on life support at the time of retransplantation remained a significant risk factor with an HR of 1.7 (95% CI: 1.4-2.2), as did receiving a split liver graft [HR: 1.8 (95% CI: 1.2-2.5). None of the other characteristics reached statistical significance.

Table 4. Multivariate Cox Proportional Hazard Analysis of Transplant-Free Survival After Initial Retransplantation
Patient CharacteristicHazard Ratio (95% CI)CoefficientP Value
  1. Abbreviations: CI, confidence interval; TPN, total parenteral nutrition.

Age   
 <6 monthsReference
 6 months to <1 year0.73 (0.43–1.23)−0.320.23
 1 year to <5 years0.71 (0.43–1.16)−0.350.17
 5 years to <12 years0.58 (0.34–0.98)−0.550.04
 12 years to <18 years0.55 (0.31–0.98)−0.590.04
Ethnicity   
 White/CaucasianReference
 Black/African American1.30 (0.99–1.72)0.270.06
 Hispanic/Latino0.98 (0.73–1.31)−0.020.88
 Asian1.45 (0.86–2.46)0.370.16
 Other/unknown1.22 (0.60–2.45)0.200.58
Era of transplant   
 1989–1994Reference
 1995–19990.91 (0.70–1.18)−0.100.48
 2000–20060.52 (0.38–0.70)−0.65<0.01
Indication for primary transplantation   
 Biliary atresiaReference
 Fulminant failue of unknown etiology1.32 (0.90–1.92)0.270.15
 Metabolic disease0.98 (0.67–1.45)−0.020.93
 Autoimmune0.85 (0.46–1.56)−0.160.60
 Neonatal or familial cholestasis/paucity of ducts/congenital abnormalities1.67 (1.18–2.36)0.51<0.01
 Vascular abnormalities1.46 (0.35–6.12)0.380.60
 Viral hepatitides1.13 (0.64–1.97)0.120.68
 Cryptogenic cirrhosis1.58 (0.94–2.64)0.460.08
 TPN cholestasis1.82 (0.89–3.70)0.600.10
 Malignancy1.50 (0.86–2.61)0.410.15
 Cystic fibrosis0.89 (0.32–2.49)−0.120.83
 Drugs and toxins1.57 (0.55–4.42)0.450.40
 Other1.13 (0.59–2.18)0.120.71
Contributing factors to graft failure   
 Vascular thrombosis0.79 (0.62–1.00)−0.240.05
 Chronic rejection0.74 (0.49–1.12)−0.300.16
 Acute rejection0.56 (0.32–0.98)−0.580.04
 Recurrent disease, nonhepatitis1.72 (0.71–4.20)0.540.23
Time since primary liver transplant   
 <1 weekReference
 1 week to <30 days1.05 (0.79–1.40)0.050.71
 30 days to <1 year1.26 (0.92–1.73)0.230.14
 1 year to <5 years1.36 (0.92–2.01)0.310.13
 5 years or more0.89 (0.44–1.80)−0.110.75
Patient on life support at time of retransplantation1.74 (1.36–2.22)0.55<0.01
Recipient liver organ type   
 WholeReference
 Reduced1.25 (0.93–1.67)0.220.14
 Split1.75 (1.22–2.50)0.56<0.01

Using the procedures laid out in the Patients and Methods section, the prognostic scoring system, summarized in Table 5, was created. A recipient's prognostic score was determined by the summation of the values for each characteristic listed in Table 5 that the individual had. The area under the ROC curve for the ability of the prognostic score to predict outcome in the modeling group was 0.61. The area under the ROC curve in the validation group was 0.62. The prognostic scores for individuals ranged from −2 to 2. When the validation group was divided, scores of less than 0 were considered low-risk, scores of 0 were considered medium-risk, and scores of 1 or greater were considered high-risk. Compared to the low-risk group, the medium-risk group had an HR of 1.5 (95% CI: 1.0-2.2), and the high-risk group had an HR of 2.4 (95% CI: 1.6-3.7). Subjects in the low-risk group had 82% (95% CI: 72%-89%) transplant-free survival at 1 year, whereas those in the high-risk group had transplant-free survival of only 49% (95% CI: 40%-57%) over the same period. These results are summarized in Table 6. Kaplan-Meier curves by risk group are shown in Fig. 2.

Table 5. Prognostic Scoring System for Pediatric Patients Undergoing First Liver Retransplantation
Patient CharacteristicScore
Age 
 5 years to 18 years−1
Indication for primary liver transplant 
 Neonatal or familial cholestasis/paucity of bile ducts/congenital abnormalities1
Contributing factors to graft failure 
 Acute rejection−1
Patient on life support at time of retransplantation1
Recipient liver organ type 
 Split1
Table 6. Hazard Ratios and Accumulative Survival at Various Time Intervals by Prognositic Scores of Pediatric Patients Receiving First Liver Retransplanation
Prognostic ScorenRangeHazard Ratio (95% CI)Event-Free Survival [% (95% CI)]
6 Months1 Year5 Years10 Years
  1. Abbreviation: CI, confidence interval.

Low risk (<0)83−2 to −1Reference84 (75–91)82 (72–89)71 (60–80)59 (45–70)
Medium risk (0)16801.5 (1.0–2.2)66 (58–73)62 (54–69)55 (47–62)50 (42–58)
High risk (> 0)1391 to 22.4 (1.6–3.7)52 (43–60)49 (40–57)39 (31–47)33 (25–42)
Figure 2.

Pediatric liver retransplantation: event-free survival of initial retransplantation by prognostic score.

DISCUSSION

Retransplantation remains a controversial undertaking, given the scarcity of donated organs and the fact that previous reports have demonstrated a poorer outcome with retransplanted livers than with primary transplants.9, 14, 21, 22, 26 The rate of retransplantation in pediatric liver transplant recipients reported in our study is similar to the figures reported in previous studies.2–9 This rate is similar for successive retransplantations through the fourth grafted liver. This demonstrates that multiple retransplants are currently an accepted course of action in pediatric transplant centers. Our study has confirmed previous work demonstrating the diminishing returns of increasing numbers of transplants in an individual. However, the largest drop-off in efficacy of transplant was between the primary transplant and the initial retransplantation. The transplant-free survival after retransplantation at 1 year in our study was 60%, which was consistent with previously reported values, which have ranged from 54% to 73%.9, 15, 17, 18, 20, 22, 26–28

This simple survival analysis does not tell the whole story. In this study, we have demonstrated that with pretransplant clinical variables, patients can be stratified into various risk categories. Patients in the low-risk category have transplant-free survival after their initial retransplant similar to that of all patients after their primary transplant. This ability to stratify patients by prognostic score will assist providers in providing anticipatory guidance to families regarding their children's prognosis after retransplantation. In the future, one could also imagine an organ allocation system that weighted predicted outcome as part of its criteria for allotment.

The Kaplan-Meier survival curve (Fig. 2) for the 3 risk categories demonstrates that the majority of the difference in outcome between the risk groups is accrued during the first 6 to 12 months. This suggests that our prognostic scoring system may predict short-term perioperative risk.

Looking at the characteristics that made it into the multivariate model discloses some interesting findings. Older patients do significantly better than younger patients with retransplantation. This finding holds true even when we control in a multivariable model for time since previous transplant, recipient liver organ type, and other likely confounders. This is consistent with the findings found in the SPLIT study.16 Speculative reasons for why this finding occurs include older recipients having superior graft size matching, greater metabolic reserve to handle the critical postsurgery state, a more mature immune system to handle the posttransplant immunocompromised state, and greater resilience in the face of opportunistic infections. This study was not designed to delineate the reasons for the divergent outcomes.

Black/African American ethnicity nearly reached statistical significance in multivariate analysis (P = 0.074). The estimated HR of 1.29 (95% CI: 0.98-1.70) suggests that black children receiving liver retransplantation may have increased risk of death or graft failure versus their white counterparts. Previous studies in children receiving liver transplants have found both no effect of race29 and significant effects on survival.30 This study did not control for socioeconomic, educational, or immunologic characteristics that may have been confounders in this finding. Racial disparities in health outcomes have been identified in many disciplines31–34; examining the validity and etiology of these disparities will be crucial to providing optimal care to all of our patients.

A more recent era of retransplantation (2000-2006) was statistically protective in univariate analysis and multivariate analysis. This suggests that our surgical and medical management of retransplantation has improved over the last 18 years, and this parallels the improved survival of pediatric liver transplant patients over the same time period.35 The era of transplant was not included in our final prognostic scoring system as it is not a characteristic that is useful for stratifying patient risks.

The only original diagnosis category that was statistically significant in our multivariate model was the heterogeneous grouping of neonatal cholestasis, familial cholestasis, paucity of bile ducts, and congenital abnormalities. Other diagnoses had trends worth considering in future studies, including cryptogenic cirrhosis, total parenteral nutrition cholestasis, malignancy, and fulminant failure of unknown etiology. It is unclear from this study how these diagnoses worsened outcome, but it can be hypothesized that many of these diagnoses can be associated with extrahepatic abnormalities that may be acting as comorbidities post-retransplantation.

Interestingly, neither acute rejection nor chronic rejection portended a poor outcome with retransplantation. In fact, both had a trend toward a protective effect, with acute rejection reaching statistical significance and therefore being included in our scoring system. This disputes the hypothesis that once a patient rejects one graft, he is at increased likelihood to have graft failure with subsequent grafts. Primary graft failure demonstrated a high risk in univariate analysis. However, when we controlled for life support status at the time of retransplantation, this risk effect disappeared. We do acknowledge that most patients with this diagnosis are on life support at the time of retransplantation, but our study suggests that it is the fact that the patient is critically ill at the time of the operation rather than the diagnosis itself that has the greatest effect on postsurgical transplant-free survival.

Previous studies have reported a dramatic difference in transplant-free survival between early and late retransplants.16, 17, 22 They have generally defined “early” as retransplantation within 30 days of primary transplantation. In univariate analysis, we saw a trend of better outcomes with longer intervals between transplants with a significant protective effect of very late retransplants, which occur 5 years after the primary transplant. However, this effect was muted substantially when we controlled for other characteristics. Given the moderate trend, we can hypothesize that very late retransplants do have a better outcome than those within the first 30 days. Interestingly, we saw a trend toward a small deleterious effect of having a retransplant after 30 days and at less than 1 year. Our findings strongly suggest that whether the patient is in critical condition requiring life support is more predictive than the timing of the retransplant in determining outcome and that the marked survival difference previously described for late retransplants in uncontrolled analyses was likely a result of this population being less critically ill.

Receiving a split liver has been found to be a risk for poorer outcomes in a previous series, and our study confirmed this finding.9, 16

When our scoring system was applied to the patients of our validation group, their scores ranged from −2 to 2. By stratifying them into risk groups, we were able to show reasonable discrimination. Our scoring system is unfortunately not discriminating enough to decide who is a retransplant candidate but certainly could affect how a practitioner could provide prognostic guidance to the family and the patient. Letting a family know that their child is in a low-risk group and is likely to do as well with his retransplant as most children do with their initial one may be very reassuring to them. Conversely, letting a family know that their child will be retransplanted but is in a high-risk group may help prepare them for the possibility of a poor outcome.

Our study was limited by the limitations of the UNOS database and the accuracy and consistency of the information within it. There were 270 patients who did not have any factors contributing to graft failure listed. We did not have missing data for our other variables. There are no standard definitions of diagnoses and factors contributing to graft failure. Inconsistent use of these terms would have likely moved our findings toward the null. The SPLIT study had the advantage of more stringent terminology and found 3 independent risk factors: elevated international normalized ratio, age under 1 year, and technical variant grafts.16 We did not have access to laboratory data in our database but, likely because of our increased number of subjects, were able to identify 2 additional independent risk factors (life support at the time of transplantation and an original diagnosis of neonatal cholestasis, familial cholestasis, paucity of bile ducts, or congenital abnormalities) and 1 additional protective factor (acute rejection contributing to graft failure). It is important to remember that this study included the patients studied in the SPLIT study, and so it would be useful to verify this scoring system in a consortium outside of the United States with a different population.

Our study was also limited by the type of data to which we had access. We did not have access to patient laboratory data or donor data and therefore were not able to include them in our analysis.

If we are able to develop increasingly discriminating prognostic tools, we may eventually be able to better allocate precious livers to those patients likely to have maximal benefit from them.

Acknowledgements

The authors thank UNOS for graciously allowing access to its database. This work was supported in part by NIH training grant number T32 DK 007762.

Ancillary