|Home | About | Journals | Submit | Contact Us | Français|
The best unrelated donor (URD) for hematopoietic cell transplantation (HCT) is allele-matched at HLA-A,B,C and DRB1. Earlier studies mostly used incomplete or lower resolution HLA typing for analysis of transplant outcome. To understand the impact of incomplete HLA characterization, we analyzed 14,797 URD HCT (1995-2006) using multivariable regression modeling adjusting for factors affecting survival. Of 21 matching cohorts, we identified 3 groups with significantly different outcomes. Well-matched cases had either no identified HLA mismatch and informative data at 4 loci or allele matching at HLA-A,B & DRB1 (n=7,477, 50% of the population). Partially matched pairs had a defined, single locus mismatch and/or missing HLA data (n=4,962, 34%). Mismatched cases had ≥2 allele or antigen mismatches (n=2,358, 16%). Multivariate adjusted five-year survival estimates were: Well-matched: 54.1 (95% confidence interval), 52.9-55.4); Partially matched: 43.7 (42.3-45.2) and Mismatched: 33.4 (32.5-36.5), p<0.001. A better matched donor yielded 10-11% better 5 year survival. Importantly, intermediate resolution A,B and DRB1 allele matched “6/6 antigen matched” HCT had survival outcomes within the partially matched cohort. We suggest that these proposed HLA subgroupings be used when complete HLA typing is not available. This improved categorization of HLA matching status allows adjustment for donor-recipient HLA compatibility and can standardize interpretations of prior URD HCT experience.
Refinements in histocompatibility testing methodology have allowed clarification of critical genetic loci important in regulating alloimmune responses between donor and recipient surrounding unrelated donor (URD) hematopoietic stem cell transplantation (HCT). As histocompatibility testing has improved, expanded matching criteria have been applied to define the optimal donor and to identify critical loci to match in order to achieve superior patient outcome.1-4 The quality of donor-recipient matching can modify graft rejection and GVHD as well as protection against neoplastic relapse and the likelihood of long-term survival.5-10 Numerous important studies have defined the critical loci for matching to include HLA-A, B, C, and DRB1 in order to maximize chances for the patients’ best long-term outcome.11-18
The available, searchable donor file, however, is neither fully typed at all loci nor at allele level resolution to allow expeditious selection of the optimal donor. Currently, only approximately 9% of the US National Marrow Donor Program (NMDP) and Bone Marrow Donors Worldwide (BMDW) have intermediate resolution typing at HLA-A and B plus allele typing at DRB1 available for immediate online searching. Allele typing has become more readily available and more widely applied for donor search and identification over the 20 years of active URD HCT activity. Nonetheless, in 2006 only 83% of transplants (not searches) had high resolution typing at all 4 loci and 35% of URD transplants had 1 or more missing or mismatched loci. Importantly, since the founding of the NMDP and other international donor registries, many transplants have been performed using donors selected by incomplete typing with methodologies insufficient to adequately delineate the histocompatibility existing between donor and recipient. The NMDP and Center for International Blood and Marrow Transplant Research (CIBMTR) maintain an extensive database of previous transplants and prospectively collected long-term outcomes. Improved matching reclassification which incorporates the completeness and resolution of previous typings can better utilize this invaluable long-term data without attrition due to incomplete HLA typing. We analyzed the best available typing compiled from initial pretransplant histocompatibility testing, from transplant centers’ later retrospective high resolution typing and from the retrospective NMDP high resolution Donor-Recipient Pair project utilizing stored pretransplant blood samples from the NMDP Research Sample Repository10,14 in conjunction with clinical outcomes data in 14,797 URD transplants performed between 1995-2006.
In order to better classify the match grades of prior transplants and thereby guide retrospective analyses of their outcome where HLA typing may be incomplete or at lower resolution, we need to understand the potential impact of that unknown HLA data on outcome in any retrospective analyses performed. We classified the number of HLA loci tested (or unknown) and the typing resolution (antigen level equivalent to intermediate resolution DNA typing or allele level, high resolution DNA-based typing) based on a previously described standard definition20 for all transplants and examined both matching and clinical criteria relevant to predict their outcome. Multivariate modeling was then used to describe clusters of matching subsets with similar outcome and thereby define the outcome for HCT patients whose HLA-match grade was predicted by the regression models. We propose a new classification of HLA-matching into three groups to be used for all retrospective analyses of URD HCT. We urge the transplant community to replace the widely used, but inadequate classification of “MUD” (matched unrelated donor transplantation) in all future analyses where allele-level typing is incomplete.
All patients whose URD HCT was facilitated by the NMDP or reported to the CIBMTR between 1995-2006 were eligible to be included in the study (n = 15,867). Surviving patients who did not provide signed, informed consent to allow analysis of their clinical data were excluded. All surviving recipients included in this analysis were retrospectively contacted and provided informed consent for participation in the NMDP research program. Informed consent was waived by the NMDP Institutional Review Board for all deceased recipients. To adjust for the potential bias introduced by exclusion of non-consenting surviving patients, a corrective action plan (CAP)-modeling process randomly excluded the same percentage of deceased patients using a biased coin randomization with exclusion probabilities based on characteristics associated with not providing consent for use of the data in survivors. This CAP modeling left 14,887 (93.9%) cases to include a broad assessment of HCT recipients. Thirty-five cases were deleted because of rare diagnoses (breast cancer, other diseases or an inherited abnormality of platelets), 7 with missing follow-up data on their first transplant, 25 missing patient race and 23 in the HLA groups where there were fewer than 20 cases in the group. HLA data was compiled from the “best available typing” data table in the NMDP/CIBMTR database using either transplant center reported HLA-typing available prior to transplant, retrospectively submitted refined or higher resolution histocompatibility typing plus data derived from the NMDP-sponsored Donor-Recipient Pair Project which retrospectively performed multiple locus allele level typing for 5708 patients within the data file.
Clinical outcome data was extracted from the NMDP/CIBMTR outcomes database. Survival was modeled using a Cox multivariable regression with the following potentially relevant factors: patient age, donor age, disease and stage, Karnofsky performance score, patient CMV serology, cell source (bone marrow vs. filgrastim-primed peripheral blood (PBSC)), first or subsequent transplant, conditioning regimen intensity, patient and donor race, GVHD prophylaxis, and year of transplant. The model was stratified on the twenty-one defined HLA-matching groups derived from the patient population grouped by number of loci tested (from HLA-A, B, C, and DRB1), level of resolution (intermediate/low or high) at each locus and extent of mismatch (Table 1). HLA-DQ and –DP were not included as previous analyses had shown them of lesser importance.10,18 The final Cox model stratified on HLA matching group was used to estimate the baseline survival function and 95% confidence interval. One year survival estimates were based on Breslow’s estimator and a log-log transform confidence interval.19 Groups were clustered based upon similarity of these regression model defined 1 year survival estimates and the presence of zero, 1 or >= 2 identified missing or mismatched loci.
The patient demographics including patient, disease and transplant technique characteristics potentially affecting transplant outcomes are shown in Table 1B. Table 2 shows the estimated one year survival for an example, baseline patient (Age<18, AML Early disease, KPS 90-100, CMV negative, marrow cell source, 1st HCT, myeloablative conditioning, tacrolimus/methotrexate GVHD prophylaxis, performed in 1995-98) stratified on HLA matching group. Figure 1 depicts the baseline survival curves (adjusted as if for this example patient) for the HLA subgroups. Table 3 shows the relevant clinical factors modifying survival based on the final Cox model.
As shown, the one year survival estimates derived from the multivariable regression identified three clusters of HLA subgroups defined by available high resolution data, number of loci typed and absence of known mismatches (Table 2, Figure 1). The best group, identified as “well-matched,” had one year survival estimates from 62.0-68.3% and included 7477 patients, 50% of the whole population. The intermediate group, “partially matched” were missing either high resolution or HLA-C data or had a defined single locus mismatch. These included 4962, 34% of the population and had estimated one year survival from 55.0-61.3%. The “mismatched” group with the poorest outcome included those with 2 or more known mismatches or missing data. This group included 2358, 16% of the population and estimated one year survival between 34.4-56.7%. Although 1 year survival for group 12 overlaps in outcome with group 11, the 2 mismatched alleles in this group suggest it is more appropriately classified with the mismatched cohort. As shown in Figure 1, these survival estimates defined three HLA subgroup clusters with survival at one year for a patient with an average set of covariates estimated as 67.7 (95% CI 66.6-68.9) for the well matched, 56.1 (95% CI 54.6-57.5) for the partially matched and 33.4 (95% CI 32.5-36.5) for the mismatched URD HCT recipients.
Identifying these three HLA subgroup clusters we sought to estimate the survival for hypothetical patients with these three match grades after adjusting for the important and statistically significant clinical risk factors including patient age, race, Karnofsky performance status, year of transplant, conditioning intensity, and GVHD prophylaxis. In the figures below (Figures 2 and and3),3), we illustrate one year (Figure 2) and five year (Figure 3) survival for a first transplant, CMV seronegative patient with a 25-year-old marrow donor of match grade shown. The patient examples are assumed to be Caucasian with a Karnofsky performance status of 90-100%, transplanted in 2003-2006 using tacrolimus plus short course methotrexate as GVHD prophylaxis. The example transplant recipients all received myeloablative conditioning intensity except the 55-year-old MDS patients who are assumed to have received a reduced intensity conditioning regimen.
As shown, for all six example patients (Panels A-F, Figure 2 and and3),3), the defined matched grade groupings identify clearly distinct survival at one year and five years for all patient cohorts including those with either early or late disease stage, who were either older or younger. HLA-matching groups resulted in approximately 10-11% decrements in 5 year survival between the well-matched, partially matched and mismatched groups for each modeled patient cohort.
Greater differences were recognized within the first year post-transplant suggesting that the impact of HLA-matching has a more profound effect on early post-transplant mortality presumptively due to graft failure, GVHD and infection, than on later post-transplant (5 year outcomes) which may be dominated by both relapse and chronic GVHD and thus possibly less affected by HLA disparity.
Examining the HLA-matching groups in detail (Table 2) identifies best outcomes associated with no recognized mismatch and informative, though not necessarily high-resolution data at all four loci (HLA-A, B, C and DRB1) or with allele matching at HLA-A, B and DRB1. In group 4, high resolution typing at HLA-A, B and DRB1 overcomes the potentially unrecognized mismatches at the untested HLA-C locus because of strong linkage disequilibrium with HLA-C in high resolution HLA-B matched pairs. Though outcomes overlap with group 5, the lack of any defined mismatch suggests this subgroup should be in the well-matched group. Well-matched transplants therefore, are characterized by informative data at the HLA-A, B and DRB1 loci and no defined mismatch, even if high resolution typing is not available at Class I loci (groups 1-4).
In contrast, the partially matched group (Table 2, groups 5-11) includes defined, single locus mismatches at any of the four loci using either high resolution or low/intermediate resolution (allele or antigen) typing. Most surprisingly, the group traditionally called “matched” and referred to in the casual parlance as “MUD” (matched unrelated donor) is group 9. These are matched at low/intermediate resolution for HLA-A and B and high resolution for DRB1 with unknown data at HLA-C. This largest (in the partially matched) subgroup (1,742 patients, 12% of the whole population) has projected one year survival, 9% inferior to the well-matched group. This clearly indicates that the widely used term “6 of 6 matched URD” is both insufficiently precise and a definitive misnomer. In this group, many unrecognized mismatches exist which contradict that this MUD group is “matched” and explaining the poorer outcomes in this group. Similarly, 8 of 8 low resolution matching (group 10), while uncommon, also leads to worsened survival at one year. This emphasizes the importance of examining all four loci and the level of resolution to understand the impact of matching on outcome.
The third, mismatched cohort (Table 2, groups 12-21) has greater heterogeneity and includes either mismatch or missing information at two or more loci in all groups. Additionally, either single low resolution mismatch (group 14) or high resolution mismatch (groups 17 and 21) with unknown C typing yields poorer survival as does 6 of 6 low resolution matching with unknown data at HLA-C (group 13).
These data strongly emphasize that classification and analysis of previous URD HCT is best performed, and most informed when it considers data at the four most critical loci, HLA-A, B, C, and DRB1. In current donor selection, this has become widely adopted. During 2006, 83% of transplants performed had high-resolution typing at all 4 loci. Of these transplants, only 65% were Well-matched, while 26% were Partially matched and 6% Mismatched using this new classification schema (CIBMTR/NMDP data, unpublished). This new classification suggests that in retrospective analyses, pairs with no defined mismatch and informative data at all four loci can be analyzed together as the best donor:recipient pairs for transplantation among previous typed transplants. This new schema does not imply that the subgroups within the three cohorts are equivalent or should be used for current donor selection where informative, allele-level typing at all 4 loci is preferred.
Recognizing these three defined HLA-matching groups, the CIBMTR and NMDP propose that all subsequent retrospective analyses of URD HCT include the available informative data at all these four loci. We also propose that these three match grade classifications be used in all retrospective analyses of outcome. Reanalysis of data sets previously published which defined MUD by the older, and insufficiently precise term “6 of 6 matching” with low/intermediate resolution at Class I and missing data on HLA-C combines nearly 25% of patients in the intermediate, partially matched group with the other, truly well-matched cohort and thus confounds interpretation of all older URD transplant reports. A previous NMDP analysis identified allele level mismatches in 50% of “6/6 antigen matched” pairs; 31% were mismatched at HLA-C.14 Earlier clinical reports, therefore, may have obscured differences in outcomes which might have been attributable to incomplete HLA matching.
We urge the transplant community to adopt this schema for all retrospective and comparative analyses and we anticipate its validation in future studies of alternative donor transplantation including umbilical cord blood and reduced intensity conditioning cohorts. While allele level precision is of course preferred and should be utilized for current donor selection,10,18 this classification maximizes the use of available information to better define interpretations of prior experience and to inform physicians and their patients for future decision-making about the application and outcomes of URD HCT.
This project has been supported by funding from the NMDP and the Department of the Navy, Office of Naval Research Cooperative Agreement #N00014-99-2-0006 and grant #N00014-05-1-0859 to the NMDP. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the Office of Naval Research or the NMDP.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.