Hepatitis C virus (HCV)-induced cirrhosis is the leading indication for liver transplantation in most Western countries (1). The posttransplant natural history of HCV includes the persistence of a positive viremia and a recurrent viral infection in almost all patients (2). Although the posttransplant HCV course often remains relatively benign or can be controlled with antiviral drugs (3), approximately 20% of recipients redevelop cirrhosis within 5 years and up to 10% present a fulminant recurrence in the form of an acute cholestatic hepatitis (4, 5).
Retransplantation is the only potentially curative therapeutic option in case of advanced graft cirrhosis, but its use in HCV recipients is a matter of debate because several multicentric studies have shown significantly worse survivals in this group compared with non-HCV-positive patients (6–8). An ideal strategy would only allow a liver retransplantation in HCV candidates with the best expected outcomes. Although several scores have been developed to predict posttransplant survival after retransplantation and thereby provide help for patient selection, none specifically addressed the problem of retransplantation in HCV-positive patients (6). The scores applicable to HCV-positive patients often extrapolated the results of non-HCV-positive patients when survivals were similar between HCV and non-HCV-positive patients (9).
This registry-based study aimed at developing a score predicting survival after retransplantation for HCV and identifying a subgroup of patients with better (and acceptable) outcomes, to whom a second transplantation should preferentially be offered.
During the study period, 2289 HCV-positive patients received more than one liver and were listed in the Scientific Registry of Transplant Recipients (SRTR) database. Eight patients benefited from three liver transplants and were excluded. Seven hundred ninety patients underwent a retransplantation within 30 days and were excluded similar to 69 patients with incoherent data. One thousand four hundred twenty-two patients fulfilled the study inclusion criteria.
Eleven studied variables (11/17) significantly differed between the first and the second transplants (Table 1). Overall, recipients were sicker and donors of better quality at the time of the second transplantation.
In the univariate analysis, 14 variables were significantly associated with survival (P<0.15, Table 2), including three related to the first transplantation and 11 to the second. They reflected the expected greater impact of the variables at the time of the second transplantation on survival after the second transplantation. Of note, creatinine and bilirubin were also assessed after log transformation, and serum creatinine at the first transplant was significantly associated with survival (P=0.010), as was serum bilirubin at the second transplant (P=0.013).
All significant native or log-transformed variables were entered into the multivariate Cox regression analysis. After the Akaike procedure, six variables remained significant and increased the likelihood of survival after the second transplant, including donor age at the second transplant (DnAge), logarithmic transformation of creatinine at the second transplant (log Creat), logarithmic transformation of the interval time between transplants (log Int), International Normalized Ratio (INR) at the second transplant, serum albumin at the second transplant (Alb), and recipient age at the first transplant (RecAge) (Table 3). The proposed score predicting survival after the second transplantation in HCV-infected patients was as follows:
ΔThe maximal value for INR was 5.
Three groups were further designed: score I with scores <30, score II between 30 and 40, and score III >40. Three years after the second transplant, the Kaplan-Meier estimated survivals were 71% (95% confidence interval [CI]: 64–76%), 56% (95% CI: 53–60%), and 37% (95% CI: 32–42%) for scores I, II, and III, respectively (Fig. 1A, log-rank test: P<0.001). Predicted survivals according to the score intervals are presented in Figure 1A. The distribution of the score among the studied population is presented in Figure 1B.
The model showed a moderate prediction accuracy: Harrell C index was 0.59. The C index corrected for the optimism was 0.58, which suggested an absence of model overfitting.
At 3 years after the second transplant, the area under the time-dependant receiver operating characteristic (ROC) curve was 0.643 (95% CI: 0.629–0.657; Fig. 2). After correction for the optimism, the area remained similar (0.627).
The proposed score can accurately predict survival after retransplantation in HCV-infected patients. It identifies a subgroup of candidates with good expected outcomes, for whom listing should be recommended in priority.
Retransplantation of HCV-positive patients is a matter of intense controversy. In case of primary nonfunction or technical failure after a first transplantation, most transplant specialists would agree that a second transplantation should be performed (and this group of patients has been excluded from the present analysis). However, when the loss of function is related to a chronic impairment, such as chronic rejection or HCV recurrence, patients are often considered as nonsuitable candidates (in view of the limited graft availability). This position is supported by a number of studies demonstrating poor overall survivals after retransplantation for HCV (8, 10–14).
However, we believe that some HCV recipients who secondarily develop liver failure can have a good outcome and should be considered for a new transplantation. Their selection should ideally be based on objective criteria (potentially within a score). To date, only one study proposed a prediction model that could assess the risk after retransplantation in HCV-positive patients (9). As a limitation, this score was first developed for primary transplantation, independently of the cause, and was further extended to retransplantation adding a variable to the equation. Its applicability in HCV-positive patients was based on the observation of similar results between HCV and non-HCV-positive patients in an analysis including both primary and retransplantations. The present score focuses exclusively on HCV-positive patients and is based on a large registry population. Of note, although our score specifically focused on HCV-positive patients, the cause of retransplantation is not automatically due to HCV-related cirrhosis. The score is, therefore, directed to HCV-positive patients, whatever the cause of the need for retransplantation.
With the exception of donor age, only presecond transplant variables were selected for the score to be usable at the time of patient selection and listing. We elected to maintain donor age in the score because of its strong predictive value. When using the score at the time of listing, it is possible to calculate an expected survival taking hypothetical donor ages into consideration. To illustrate this point, a patient could present a good predicted survival with a 20-year-old donor but a poor predicted survival in case of a 50-year-old donor.
Retained variables included donor age, recipient age, creatinine, albumin, and INR at the second transplant and the interval between both transplants. Some of these variables have already been suggested to have some impact on posttransplant survival in HCV and non-HCV recipients of the first or second transplants (11, 13, 15–22). Overall, they seem logical as some reflect the degree of liver failure (albumin, INR, and, indirectly, creatinine). Donor age has a significant impact on the quality of the liver graft and on posttransplant survival, as demonstrated by the Donor Risk Index, and its effect is at least as important in HCV-positive patients (23–26). In addition, the time between both transplantations has been demonstrated to alter postsecond transplant outcome in HCV and non-HCV-positive patients: the longer, the better (8, 27, 28). Of note, a number of data were not available in the SRTR, including HCV genotype, viremia level, type of anti-HCV treatment, and biopsy result. In particular, the use of anti-HCV treatment (including newly released antiviral drugs) has the potential to alter the postretransplant course. The inclusion of some of these variables could further improve the score accuracy (which already shows a Harrell C index of 0.59).
To determine the model accuracy we used time-dependant ROC curves, which take into account censored and available informations, and, therefore, are more appropriate than classical ROC analyses. The optimism estimation allows determination of the impact of the model overfitting and gives the range of model accuracy if we were to repeat the analysis in similar populations. Although this procedure represents a powerful internal validation of the score, we acknowledge that the ultimate validation will need to test our model on an external population, which could be inspired by the present data.
One limitation of this multicentric registry-based study is the fact that all patients had been preselected by the transplant centers and that the studied sample, therefore, did not correspond to “naive” HCV-transplanted patients with recurrence and cirrhosis. To illustrate, it is possible that a number of centers were not considering patients with cholestatic hepatitis C syndrome for retransplantation and that they were underrepresented in the database. As stated in the Materials and Methods and the Discussion, the cause of retransplantation was not clearly reported in the SRTR. However, only the proposed strategy of investigation allows for the inclusion of enough patients to reach a meaningful statistical power.
Another limitation of the study was the accuracy and the availability of the SRTR data. Although a misclassification of HCV-positive patients is always possible, we postulate that frequency of such an event is low and does not significantly alter results. In addition, any misclassification is independent of survival and can only potentially decrease the predictive value of the score (which has already a good predictive value).
Of note, we elected not to propose a formal score cutoff to accept or reject an HCV candidate for a retransplantation, as this decision depends on patient's characteristics, graft availability, and local policies, and there must be room for individual score-unrelated decisions based on personal experiences. A center with more available donor may consider HCV candidates with lower potential survival after retransplantation.
Overall, the proposed score is objective and specifically designed for HCV-positive patients, accurately predicts outcome, and provides an important added value in the relisting decision and in the selection of the most appropriate candidates.
MATERIALS AND METHODS
Study Design and Population
This study analyzed data from the SRTR. The SRTR data system includes data on all donors, wait-listed candidates, and transplant recipients in the United States, submitted by the members of the Organ Procurement and Transplantation Network (OPTN), and has been described elsewhere (29). The Health Resources and Services Administration, U.S. Department of Health and Human Services provides oversight to the activities of the OPTN and SRTR contractors. This study has been reviewed and approved by the institutional Health Research Ethics Board.
All HCV-infected patients undergoing two liver transplants between January 1,1990, and January 31,2009, were included in the study. Patients undergoing the second transplant within 30 days of the first were excluded, because these early failures are usually related to primary graft nonfunction or technical failure. Of note, the cause of retransplantation was not clearly reported in the SRTR database, and some patients may have been retransplantated for non-HCV-related problems. As a consequence, the derived score may be applied to all HCV-positive patients before retransplantation, whether the cause of retransplantion is related to HCV.
Only patient survival from the second transplantation was used as an outcome variable. The date of death was obtained from data reported by the transplant centers and was completed by data from the U.S. Social Security Administration and the OPTN. Living patients were censored at the time of the last available follow-up.
Clinically relevant variables were assessed at the first and second transplantations and compared by using Student's t or chi-square test. Survival analyses were performed to determine factors associated to the risk of mortality after the second liver transplantation. A univariate analysis was conducted by log-rank test for categorical variables. For continuous variables, a Cox model with a likelihood ratio test was used to compare native variables with their log-transformed and categorical forms and to determine which form of the variable would be kept in the final model. In the particular case of INR, we attributed a maximal value of 5 to balance the weight of extreme values. Variables for which data were missing in more than 30% of the subjects were excluded from the multivariate analysis. However, when missing information was considered as random (INR and model for end-stage liver disease before 2002), missing values were replaced by the median of the variable. To select variables for the multivariate Cox regression model, a stepwise Akaike procedure was performed over the total population (30), which included any variable with a P value <0.15 in the univariate analysis. As shown by Stone (31), the Akaike procedure allows for the development of a model equivalent to a cross validation. The assumption of proportionality of the hazards was checked (Schoenfeld residuals).
To design a score available at the time of a potential relisting, postsecond transplant variables were excluded from the score design. The score was calculated based on the multivariate model coefficients, with each coefficient rounded to 2 digits after the comma. For some experiments, three groups of equivalent sizes were selected and the corresponding survival curves were assessed by the Kaplan-Meier method and compared by a log-rank test.
Validation of the Score
Because the assessment of the score was performed on the total population and not on a random fraction of the population, the validation was also performed on the total population. The predictive performance of the score was estimated by using Harrell C index (32). To assess the potential impact of overfitting, the optimism was assessed by bootstrap (1000 iterations): it represents the part of the Harrell C index lost that is caused by overfitting. The Harrell C index adjusted for the overfitting is the Harrell C index minus the optimism.
To assess the accuracy of the score in predicting survival for a given time horizon, time-dependent ROC curves were assessed, as proposed by Heagerty et al. (33). The area under the time-dependant ROC curve at 3 years after the second transplantation is given with the 95% CI obtained by bootstrap (1000 iterations).
Model building and survival analysis were performed with Stata 10.1 (StataCorp, College station, TX). Time-dependant ROC curve and Harrell C index were obtained by using the package SurvivalROC and the function validate.cph of the package Design for R 2.12.0.
The data reported here have been supplied by the Arbor Research Collaborative for Health (Arbor Research) as the contractor for the Scientific Registry of Transplant Recipients (SRTR). The interpretation and reporting of these data are the responsibility of the authors and in no way should be seen as an official policy of or interpretation by the SRTR or the U.S. Government.
2. Wright TL, Donegan E, Hsu HH, et al.. Recurrent and acquired hepatitis C
viral infection in liver transplant recipients. Gastroenterology 1992; 103: 317.
3. Aytaman A, Kaufman M, Terrault NA. Management of posttransplant hepatitis C
infection. Curr Opin Organ Transplant 2010; 15: 301.
4. Berenguer M, Ferrell L, Watson J, et al.. HCV-related fibrosis progression following liver transplantation
: Increase in recent years. J Hepatol 2000; 32: 673.
5. Gane EJ. The natural history of recurrent hepatitis C
and what influences this. Liver Transpl 2008; 14(suppl 2): S36.
6. Carrion JA, Navasa M, Forns X. Retransplantation
in patients with hepatitis C
recurrence after liver transplantation
. J Hepatol 2010; 53: 962.
7. Pelletier SJ, Schaubel DE, Punch JD, et al.. Hepatitis C
is a risk factor for death after liver retransplantation
. Liver Transpl 2005; 11: 434.
8. Yoo HY, Maheshwari A, Thuluvath PJ. Retransplantation
of liver: Primary graft nonfunction and hepatitis C
virus are associated with worse outcome. Liver Transpl 2003; 9: 897.
9. Ghobrial RM, Gornbein J, Steadman R, et al.. Pretransplant model to predict posttransplant survival
in liver transplant patients. Ann Surg 2002; 236: 315.
10. Berenguer M, Prieto M, Palau A, et al.. Severe recurrent hepatitis C
after liver retransplantation
for hepatitis C
virus-related graft cirrhosis. Liver Transpl 2003; 9: 228.
11. Yao FY, Saab S, Bass NM, et al.. Prediction of survival
after liver retransplantation
for late graft failure based on preoperative prognostic scores. Hepatology 2004; 39: 230.
12. Roayaie S, Schiano TD, Thung SN, et al.. Results of retransplantation
for recurrent hepatitis C
. Hepatology 2003; 38: 1428.
13. Neff GW, O'Brien CB, Nery J, et al.. Factors that identify survival
after liver retransplantation
for allograft failure caused by recurrent hepatitis C
infection. Liver Transpl 2004; 10: 1497.
14. Bellido CB, Martinez JM, Gomez LM, et al.. Indications for and survival
after liver retransplantation
. Transplant Proc 2010; 42: 637.
15. Doyle HR, Morelli F, McMichael J, et al.. Hepatic retransplantation
—An analysis of risk factors associated with outcome. Transplantation 1996; 61: 1499.
16. Azoulay D, Linhares MM, Huguet E, et al.. Decision for retransplantation
of the liver: An experience- and cost-based analysis. Ann Surg 2002; 236: 713.
17. Wong T, Devlin J, Rolando N, et al.. Clinical characteristics affecting the outcome of liver retransplantation
. Transplantation 1997; 64: 878.
18. Markmann JF, Gornbein J, Markowitz JS, et al.. A simple model to estimate survival
of the liver. Transplantation 1999; 67: 422.
19. Rosen HR, Madden JP, Martin P. A model to predict survival
following liver retransplantation
. Hepatology 1999; 29: 365.
20. Linhares MM, Azoulay D, Matos D, et al.. Liver retransplantation
: A model for determining long-term survival
. Transplantation 2006; 81: 1016.
21. Rosen HR, Prieto M, Casanovas-Taltavull T, et al.. Validation and refinement of survival
models for liver retransplantation
. Hepatology 2003; 38: 460.
22. Facciuto M, Heidt D, Guarrera J, et al.. Retransplantation
for late liver graft failure: Predictors of mortality. Liver Transpl 2000; 6: 174.
23. Maluf DG, Edwards EB, Stravitz RT, et al.. Impact of the donor risk index on the outcome of hepatitis C
virus-positive liver transplant recipients. Liver Transpl 2009; 15: 592.
24. Gastaca M. Extended criteria donors in liver transplantation
: Adapting donor quality and recipient. Transplant Proc 2009; 41: 975.
25. Boin IF, Ataide EC, Leonardi MI, et al.. Elderly donors for HCV(+) versus non-HCV recipients: Patient survival
following liver transplantation
. Transplant Proc 2008; 40: 792.
26. Feng S, Goodrich NP, Bragg-Gresham JL, et al.. Characteristics associated with liver graft failure: The concept of a donor risk index. Am J Transplant 2006; 6: 783.
27. Markmann JF, Markowitz JS, Yersiz H, et al.. Long-term survival
of the liver. Ann Surg 1997; 226: 408.
28. Ghabril M, Dickson R, Wiesner R. Improving outcomes of liver retransplantation
: An analysis of trends and the impact of Hepatitis C
infection. Am J Transplant 2008; 8: 404.
29. Merion RM. 2006 SRTR Report on the State of Transplantation. American Journal of Transplantation 2007; 7 (Part 2); pp. 1317–1318.
30. Akaike H. New look at the statistical model identification. IEEE Trans Autom Control 1974; 19: 716.
31. Stone M. An asymptotic equivalence of choice of model by cross-validation and Akaike's criterion. J R Stat Soc 1977; 39: 44.
32. Harrell FE Jr, Lee KL, Mark DB. Multivariable prognostic models: Issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996; 15: 361.
33. Heagerty PJ, Lumley T, Pepe MS. Time-dependent ROC curves for censored survival
data and a diagnostic marker. Biometrics 2000; 56: 337.