The effect of HLA class I (A and B) and class II (DR) compatibility on liver transplantation outcomes: An analysis of the OPTN database
The purpose of this study was to explore the relationship between human leukocyte antigen (HLA) compatibility and liver transplantation outcomes by analyzing the effect of HLA compatibility on 5-year graft survival. We analyzed first liver transplants between 1987 and 2002 in the Organ Procurement and Transplantation Network (OPTN) database, where A, B, or DR loci data were available. Graft failure was defined as retransplantation or death from transplant-related cause. We evaluated associations between total and locus-specific match levels and 5-year graft survival. Multivariable Cox proportional-hazard models were used to evaluate statistical interactions and adjust for the effect of potential confounders. Among 29,675 first-time transplants, the overall degree of HLA match had no effect on 5-year graft survival, even after controlling for potential confounders. Univariate and multivariable analyses showed that the 0 HLA antigen mismatch cohort of patients had higher 5-year graft failure rates than the other 6 antigen mismatch groups. However, this occurred in a small group with a disproportionately large number of live donors and split-liver recipients. When these recipients were excluded from the analysis, the effect was no longer seen. Finally, multivariable, locus-specific analyses showed no association between 5-year graft survival and degree of match/mismatch and the A, B, or DR loci. In conclusion, this careful examination of the OPTN database, with respect to HLA match or mismatch and liver graft survival, reaffirms that HLA matching has no clinically significant impact on this outcome. Liver Transpl 12:652–658, 2006. © 2006 AASLD.
The benefit of human leukocyte antigen (HLA) compatibility in kidney, heart, and bone marrow transplantation is indisputable. Closer compatibility leads to a lower incidence of acute and chronic rejection and improved graft survival.1, 2 As a result, HLA compatibility between donor and recipient is viewed as an important determinant of transplantation outcomes. Using HLA phenotypes to match donor with recipient when feasible has become standard practice.
In liver transplantation, HLA compatibility between the donor and recipient has been largely of academic interest. That is to say, donor and recipient HLA compatibility is not considered to have a clinically significant effect on liver transplant outcomes. Histocompatibility, therefore, does not factor into donor and recipient matching. Nonetheless, the notion that the degree of HLA matching may have an adverse impact on graft or patient outcomes has been explored by several investigators. Most importantly, Markus et al. observed, in a large cohort of liver transplantation recipients, that greater HLA compatibility at the A and DR loci was associated with a higher incidence of nonrejection-related graft failure.3 This group also observed a higher incidence of graft failure due to rejection with a lower degree of HLA compatibility. These seminal observations introduced the possibility of a “dualistic” role of HLA in liver transplantation—that is, protection from rejection with greater donor and recipient compatibility, but a higher risk of nonrejection-related graft failure under the same conditions.3 Subsequently, Donaldson et al. confirmed the observation that class I compatibility exerted a dual effect on graft survival.4 Finally, Neumann et al. found lower rejection rates in closer HLA-matched transplants; however, graft survival in transplants for primary sclerosing cholangitis was significantly impaired in the presence of increasing HLA-DR compatibility. Similar results were noted for autoimmune hepatitis, although the findings were not statistically significant.5
The question of whether donor and recipient HLA compatibility has any clinically significant impact on liver transplant outcomes, namely graft survival, has not been asked using the vast body of data available through the United Network for Organ Sharing (UNOS) Organ Procurement and Transplantation Network (OPTN) database. This resource, by virtue of its size, affords an opportunity to search for an effect of donor and recipient HLA compatibilities on transplant outcomes that smaller cohorts may fail to detect. In addition, confounding variables and other factors that may affect outcomes can also be examined in the context of the degree of HLA matching. Thus, the primary aim of this study was to evaluate liver transplant graft survival as a function of HLA match and mismatch using the UNOS database of donor and recipient pairs.
MATERIALS AND METHODS
This was a retrospective cohort study of patients represented in the OPTN database as of October 14, 2003. To be included in the analysis, the transplantation must have been the first for that recipient, it must have been completed between 1987 and June, 2002, and complete information needed to be available for the relevant outcome and confounding variables used in the analysis. UNOS randomly changes the transplant dates by up to 180 days (in either direction) to preserve patient anonymity. Therefore, these transplantations could have occurred as late as December, 2002.
The OPTN database contained 61,148 liver transplants performed in 55,193 recipients. From these, a total of 29,695 first-time transplants met our selection criteria for 1 or more of the analyses; 28,608, 29,575, 29,571 and 28,735 met the selection criteria for the combined match/mismatch, A, B, and DR locus-specific analyses, respectively. Of the first-time recipients excluded from the analysis, 21,965 (86.1%) were missing data for all 3 HLA loci, another 2,980 (11.7%) had transplantations outside of the included date range, another 98 (0.4%) were missing data on graft failure status or timing, and another 455 (1.8%) were missing 1 or more candidate-confounding variables.
Our primary outcome was graft failure during the first 5 years after transplantation, with graft failure defined as retransplantation or death from a transplant-related cause. The latter included death attributable to 1 of 9 UNOS categories of graft failure: biliary tract complications, graft vs. host disease, hepatitis, other infections, primary graft failure, other recurrent disease of hepatic origin, rejection, vascular thromboses, and other graft failure. Graft failure did not include any deaths occurring more than 5 years after transplantation. We chose to limit our analyses to the first 5 years after transplantation due to our assumption that any clinically significant sequelae of HLA compatibility or incompatibility would have occurred within this time period. Individual causes of graft failure could not be evaluated with the complete data set, as data on these causes were often missing. Of the 3,858 graft failures meeting selection criteria, for example, 1,622 (42.3%) and 1,683 (43.2%) had missing or unknown data for the variables indicating biliary disease or de novo hepatitis, respectively, as a contributing cause of failure. Similarly, we could not evaluate rejection as a specific outcome as 87.0% of recipients included in 1 or more analyses (25,840 of 29,695) were missing data on both acute and chronic rejection.
Survival time was defined as the time from first transplant to date of graft failure. Patients with intact grafts 5 years from transplantation were censored at that time point. Patients lost to follow-up during the first 5 years after transplantation were censored at the last recorded medical visit. Recipients who died during the first 5 years after transplantation were included in the analysis but censored at the time of death if the cause of death was trauma or suicide, cardiovascular or cerebrovascular disease, gastrointestinal bleeding, nonhepatic malignancy, generalized sepsis, or pulmonary and renal failure. Although several of these causes of death may have been related to the transplant process or complications of immunosuppression, they were censored because we considered it unlikely that HLA compatibility would play a role in a substantial number of patients with these disease processes. Therefore, classification of these non-immune-mediated causes of death as being directly graft related would potentially have diluted or masked the effect of HLA matching on graft survival.
We conducted 5 sets of analyses. In the first 2, we assessed the association between graft survival and combined HLA match or mismatch levels at 3 loci. For these 2 sets, total HLA matches or mismatches were counted, each with a possible range from 0 to 6 (2 possible alleles at each of 3 loci: A, B, or DR). Because of the small number of recipients having 6 matches with their donors, those with 5 or 6 HLA matches were collapsed into a single category during analysis of HLA matches. In contrast, all 7 categories of mismatches (0 to 6 mismatches) were kept in the mismatch analysis. In the other 3 sets of analyses, we evaluated the association between graft survival and degree of match/mismatch at each locus (A, B, DR) individually. For each locus, recipients were placed into 1 of 5 categories based on their match/mismatch level with the donor: 2 matches and 0 mismatches, 1 match and 0 mismatches (homozygous donor), 1 match and 1 mismatch, 0 matches and 1 mismatch (homozygous donor), 0 matches and 2 mismatches. Matches and mismatches were always counted from the recipient's perspective.
For all 5 sets of analyses, graft survival was estimated using the Kaplan-Meier method and univariate comparisons were made using the log-rank test.6 For each of the 5 sets, a global test of significance for all match/mismatch categories was conducted (e.g., testing the overall effect of match level for A locus) and then individual match categories (e.g., 2 A locus matches and 0 mismatches) were evaluated. Multivariable Cox proportional hazard (PH) models were developed to evaluate statistical interactions and adjust for the effect of potential confounding variables.7 In the PH models, we assessed statistical interactions between HLA match/mismatch level and 8 other variables: diagnosis of autoimmune disease (autoimmune hepatitis, primary sclerosing cholangitis, primary biliary cirrhosis, graft vs. host disease secondary to non–liver transplant), diagnosis of viral hepatitis, donor and recipient sex, donor and recipient race, age group, and treatment year. We then evaluated the effect of candidate confounding variables: recipient diagnosis, year of transplant, concurrent multiorgan transplantation, donor type (live vs. cadaveric), donor-recipient ABO blood type compatibility, cytomegalovirus infection, time on waiting list before transplantation, medical condition prior to transplantation (in intensive care unit, other hospitalization, or not hospitalized), as well as all 8 variables listed in the interaction evaluation. We began with a main-effects model and evaluated the statistical significance of the interaction terms one at a time. Then, confounding was assessed using a backward elimination algorithm. Candidate confounding variables were retained in final models when removing them changed one or more coefficient(s) for the HLA match/mismatch level by more than 10%. Statistical significance of the HLA match/mismatch level was assessed again in the final model. Separate models were developed to assess confounding by hepatitis C virus (HCV) and cytomegalovirus status, using the records from 1994 through 2002 for which HCV and cytomegalovirus test results were available. In the PH regression, all statistical significance was determined using the likelihood ratio test.8 A P of ≤0.05 was considered statistically significant. For each of the 5 sets of analyses, PH models were developed for the analysis of overall effect (e.g., overall effect of total HLA mismatches). Individual match/mismatch levels (e.g., having 2 mismatches) were evaluated with PH models only if the log-rank test for that level was significant. All analyses were conducted with SAS version 8.2 (SAS Institute Inc., Cary, NC).
Characteristics of and Graft Survival in the Final Study Cohort
Among the 29,695 recipients included in 1 or more of the analyses, the majority was white and male and the most common diagnosis was viral hepatitis (Table 1). Among the 3,858 (13%) recipients with graft failure in the first 5 years, 1,001 (3.4%) died of a liver-related cause without retransplantation, and 2,857 (9.6%) received a second transplant. The median graft survival time was 271 days (range, 0 to 1,821 days) for those who died and 31 days (range, 0 to 1,818 days) for those who received a second transplant.
Table 1. Characteristics of Transplant Recipients and Donors (N = 29,695 Procedures)
|Recipient Gender|| || |
|Donor Gender|| || |
|Recipient race|| || |
| White, non-Hispanic||22,700||76.44|
| Black, non-Hispanic||2,386||8.04|
| Multiracial, non-Hispanic||2||0.01|
|Donor race|| || |
| White, non-Hispanic||23,165||78.01|
| Black, non-Hispanic||3,232||10.88|
| Multiracial, non-Hispanic||9||0.03|
|Treatment Year*|| || |
|Pretransplantation diagnosis|| || |
| AHN, cryptogenic||828||2.79|
| Alcohol induced||5,576||18.78|
| Cirrhosis, cryptogenic||2,923||9.84|
| Hepatic malignancy||1,101||3.71|
| Metabolic, genetic||3,243||10.92|
| Secondary biliary cirrhosis||166||0.56|
Combined HLA Match and Mismatch Analyses
Among the 28,608 recipient-donor pairs analyzed to assess the effect of combined HLA compatibility on graft survival, the median number of HLA matches was 1 (range, 0 to 6), and the median number of mismatches was 5 (range, 0 to 6). In the univariate analyses, the tests of global effect of HLA match or HLA mismatch on graft survival were not significant (P = 0.24 and 0.12, respectively). When testing individual categories, 1 level of HLA mismatch was statistically different from the others; the category of recipients with 0 HLA mismatches had significantly higher 5-year graft failure (19.3%) when compared to all other levels of HLA mismatch (12.9% failure, P = 0.02) (Table 2). This category, however, had only 114 recipients, representing 0.4% of all recipients. These recipients were clearly different in other ways from the general recipient population, with a significantly higher proportion of live donors (32% vs. 2% in other mismatch categories) and higher proportion receiving partial or split-organ transplants (33% vs. 4% in other mismatch categories). In the multivariable analysis, the effect of 0 HLA mismatches remained statistically significant (adjusted P = 0.03; hazard ratio = 1.59; 95% confidence interval, 1.04, 2.43). Considering the potential effect of partial organ transplantation, we repeated this analysis after excluding these recipients and found the effect of 0 HLA mismatches to no longer be statistically significant (log-rank test; P = 0.08).
Table 2. 5-Year Graft Failure by Donor-Recipient HLA Match Level at A, B, and DR Loci
|Number of combined HLA mismatches (N = 28,608)|| || || |
| 0||114 (0.4)||22 (19.3)||8.28*|
| 1||211 (0.7)||23 (10.9)||3.87|
| 2||1,095 (3.8)||129 (11.8)||3.86|
| 3||3,773 (13.2)||482 (12.8)||4.24|
| 4||7,603 (26.6)||957 (12.6)||4.11|
| 5||10,209 (35.7)||1,344 (13.2)||4.41|
| 6||5,603 (19.6)||737 (13.2)||4.48|
|Number of combined HLA matches (N = 28,608)|| || || |
| 0||8,821 (30.8)||1,168 (13.2)||4.50|
| 1||11,146 (39.0)||1,422 (12.8)||4.23|
| 2||5,791 (20.2)||748 (12.9)||4.22|
| 3||2,327 (8.1)||296 (12.7)||4.20|
| 4||414 (1.4)||42 (10.1)||3.40|
| 5 or 6||109 (0.4)||18 (16.5)||7.44|
|A locus matches, mismatches (N = 29,575)|| || || |
| 2 matches, 0 mismatches||765 (2.6)||94 (12.3)||4.13|
| 1 match, 0 mismatches||1,521 (5.1)||196 (12.9)||4.27|
| 1 match, 1 mismatch||9,893 (33.4)||1,249 (12.6)||4.15|
| 0 matches, 1 mismatch||3,026 (10.2)||380 (12.6)||4.23|
| 0 matches, 2 mismatches||14,370 (48.6)||1,927 (13.4)||4.49†|
|B locus matches, mismatches (N = 29,571)|| || || |
| 2 matches, 0 mismatches||338 (1.1)||47 (13.9)||4.91|
| 1 match, 0 mismatches||458 (1.5)||64 (14.0)||4.89|
| 1 match, 1 mismatch||6,458 (21.8)||830 (12.9)||4.25|
| 0 matches, 1 mismatch||2017 (6.8)||268 (13.3)||4.52|
| 0 matches, 2 mismatches||20,300 (68.7)||2,638 (13.0)||4.32|
|DR locus matches, mismatches (N = 28,735)|| || || |
| 2 matches, 0 mismatches||818 (2.9)||110 (13.4)||4.52|
| 1 match, 0 mismatches||1,038 (3.6)||140 (13.5)||4.43|
| 1 match, 1 mismatch||9,436 (32.8)||1,198 (12.7)||4.20|
| 0 matches, 1 mismatch||3,087 (10.7)||391 (12.7)||4.12|
| 0 matches, 2 mismatches||14,356 (50.0)||1,868 (13.0)||4.38|
In the other multivariable analyses, we found no significant interactions between HLA match or mismatch and ethnicity, gender, transplantation during childhood, or recipient diagnoses of viral hepatitis or autoimmune disease. There was a statistically significant interaction, however, between HLA match level and the age group of recipients ≥50 years old (P = 0.01). The interaction between combined HLA match level and age group was driven mostly by a differential graft failure rate for those with 2 matches. In recipients <50 years of age, those with 2 matches had a higher failure rate (16.1%) than those with 0, 1, 3, 4, 5, or 6 matches in the same age group (14.7% failure rate). In contrast, recipients ≥50 years of age had a lower failure rate (9.2%) with 2 matches when compared to those in this age group with 0, 1, 3, 4, 5, or 6 matches (10.7% failure). As the locus-specific analysis will show, no single locus accounted for this effect. Given the illogical pattern of this interaction and the lack of a biological mechanism that would explain it, we considered it likely that its statistical significance was due to a Type I error. Therefore, these interaction terms were not carried further in multivariable models that evaluated combined HLA match level.
After adjusting for potential confounding, we found no significant association between combined HLA match or mismatch level and 5-year graft survival (adjusted P = 0.37 and 0.29, respectively, in final Cox PH model).
In univariate analyses of locus-specific match level, only 1 comparison was statistically significant; for HLA-A, the global test of effect was not significant, but recipients with 2 A-locus mismatches were found to have slightly higher 5-year graft failure rates (13.4%, Table 2) when compared to those with 0 or 1 mismatches (12.6% failure, log-rank test, P = 0.03).
In multivariable analyses, we found no significant interactions for any of the 3 locus-specific analyses, A locus included. Similarly, we found no association between the match/mismatch level at the B and DR loci and 5-year graft survival, even after adjusting for confounding (adjusted P = 0.86 and 0.85, respectively, in final Cox PH model). In multivariable analysis, the global effect of HLA-A locus match/mismatch was not significant (adjusted P = 0.33) and the effect of having 2 A-locus mismatches was no longer significant (adjusted P = 0.12; hazard ratio = 1.05; 95% confidence interval, 0.99, 1.12).
In this study, we hoped to determine decisively, using the OPTN liver transplant database, whether HLA matching in liver transplantation has any impact on graft survival and whether this impact differed in recipients with autoimmune and viral diagnoses. As will be discussed, several limitations to the database must be considered when comparing our findings to those of others. Nevertheless, the large number of transplant recipients captured in the database afforded increased statistical power for many comparisons that had not been possible in previous studies including evaluation of statistical interactions. Similarly, the large number of variables in the database allowed us to control for potential confounding by many important determinants of graft survival. In our analysis, we found no association between the degree of combined HLA matching and graft survival. When evaluating HLA mismatches in multivariable analysis, the only statistically different subgroup was the one with 0 HLA mismatches. However, these recipients represented only 0.4% of all recipients and they were clearly different from the general recipient population in that there was a large proportion of live donor and partial graft recipients. When we excluded these recipients from the analyses, the 0 mismatch group was no longer different. In the univariate locus-specific analysis, we identified a higher rate of graft failure with 2 A locus mismatches; however, this relationship did not persist in multivariable analysis.
The effect of HLA compatibility on liver transplantation outcomes has been the focus of several studies spanning more than a decade. Earlier studies identified various effects of HLA compatibility on liver transplant outcomes.3, 4, 5 Markus et al. in 1988 demonstrated that increased HLA compatibility was associated with diminished graft survival, although transplant rejection was reduced.3 Neumann et al., in their study of 924 patients, found that the total number of HLA matches had no influence on overall graft survival, and acute cellular rejection was reduced with increased matching. When analyzing survival based on etiology of liver disease, survival was significantly reduced in transplants for primary sclerosing cholangitis in the presence of 1 or 2 HLA-DR compatibilities. Similar trends were noted for autoimmune hepatitis, although not statistically significant.5 Knechtle et al. identified that a mismatch at the HLA A locus correlated with graft loss, and that better matching improved patient and graft survival.9 These studies, point out that HLA matching and mismatching play a dual role in liver transplantation, on the one hand enhancing graft survival but, on the other hand, increasing the risk of rejection with increasing levels of HLA match, as first suggested by Markus.3 However, these studies also highlight the complex interplay of HLA compatibility and success of the graft.
Investigators have also found that HLA compatibility may affect the recurrence of hepatitis B and C viruses following liver transplantation. Gaur et al. found that compatibility at the HLA DQB locus correlated with recurrent HCV.10 In another study, compatibility of class I antigens correlated with recurrent HBV.11 Finally, Manez et al. found that matching at the HLA B locus increased the risk of recurrent viral hepatitis.12 In our study, we tested the hypothesis that HLA matching might have a differential effect for recipients with prior viral hepatitis and also introduced a series of controls for potential confounding by HCV or cytomegalovirus. We could not directly evaluate recurrent viral hepatitis, and we had no data on HLA DQB. However, our results suggest that if an association between HLA matching at A, B, and DR loci and recurrent viral hepatitis exists, this association did not result in a significant difference in graft survival.
The apparent discrepancies between our results and previously published studies may, in part, be the result of limitations of the OPTN database. In particular, given the anonymity of the data, we could not control for factors that may have impacted the outcome, such as immunosuppression handling and surgical technique. Also, single-center data may have an advantage over OPTN data, in that patients may have received similar care in a given center and investigators had the opportunity to verify diagnoses. Moreover, the OPTN data allowed us to consider only a part of the HLA system—that is, HLA A, B (Class I) and DR (Class II). Others have shown a potential effect of other HLA antigens or antibody reactivity in graft survival.13–15 Also, some have shown an effect of HLA G in rejection.13, 16 These limitations notwithstanding, our study has the advantage of greatly increased statistical power with controls to examine various potential confounding and interacting variables.
This paper reviews the data of patients experiencing liver graft loss and asks, Does HLA matching at the A, B, and DR loci, either combined or at the locus-specific level, have an effect on graft survival? To answer this question fully, one must consider the possibility that any immunological effects of HLA mismatching could be masked by other nonimmunological processes that may affect graft survival, such as recurrent liver disease, surgical technique, technical complication, and immunosuppression handling. Any of these factors may have an effect on graft outcome. It is also possible that HLA compatibility could have a differential effect for different types of graft failure in live donor or partial transplantation, as our data suggest. The data necessary to answer some of these questions are either not contained within or incomplete in the existing UNOS database. Thus, we can conclude that HLA matching or mismatching had no clinically significant impact on liver graft survival in cadaveric liver transplantation. However, while graft loss statistics may show no impact in cadaveric liver transplantation, the question of the effect of HLA matching in live donor transplantation and on the physiological success of the organ remains unanswered. This study of UNOS data will be extended to include the effects of the HLA compatibility on graft outcome in individual disease entities within our own center, for which data are available.
The authors thank Maricruz Velez, MPH for her assistance in preparation of the manuscript.