Increased age-associated mortality risk in HLA-mismatched hematopoietic stem cell transplantation.

We investigated a possible interaction between age-associated risk and HLA-mismatch associated risk on prognosis in different age categories of recipients of unrelated hematopoietic stem cell transplants (HSCT) (n=3019). Patients over 55 years of age transplanted with 8/10 donors showed a mortality risk of 2.27 (CI 1.70-3.03, P<0.001) and 3.48 (CI 2.49-4.86, P<0.001) when compared to 10/10 matched patients in the same age group and to 10/10 matched patients aged 18-35 years, respectively. Compared to 10/10 matched transplantations within each age category, the Hazards Ratio for 8/10 matched transplantation was 1.14, 1.40 and 2.27 in patients aged 18-35 years, 36-55 and above 55 years. Modeling age as continuous variable showed different levels of risk attributed to age at the time of transplantation [OS: 10/10: Hazards Ratio 1.015 (per life year); 9/10: Hazards Ratio: 1.019; 8/10: Hazards Ratio 1.026]. The interaction term was significant for 8/10 transplantations (P=0.009). Findings for disease-free survival and transplant-related mortality were similar. Statistical models were stratified for diagnosis and included clinically relevant predictors except cytomegalovirus status and Karnofsky performance status. The risk conferred by age at the time of transplantation varies according to the number of HLA-mismatches and leads to a disproportional increase in risk for elderly patients, particularly with double mismatched donors. Our findings highlight the importance of HLA-matching, especially in patients over 55 years of age, as HLA-mismatches are less well tolerated in these patients. The interaction between age-associated risk and HLA-mismatches should be considered in donor selection and in the risk assessment of elderly HSCT recipients. Copyright© Ferrata Storti Foundation.

Introduction

Unrelated hematopoietic stem cell transplantation is a rapidly evolving field offering a curative therapy for various hematologic diseases. In particular, the proportions of older patients and patients transplanted with unrelated donors have increased over the last decade.[1,2] One prerequisite was the introduction of reduced intensity conditioning regimens (RIC) as an alternative to myeloablative conditioning (MAC) in elderly patients as well as in patients with co-morbidities.[3,4] There is already a wealth of data showing that RIC is a safe and effective treatment form for patients previously not eligible for hematopoietic stem cell transplantation (HSCT).[5,6] As a consequence, therapeutic schemes for elderly patients have been established which now include HSCT as treatment option in some clinical instances.[7] Nevertheless, classical risk factors still apply, and while increasing age did not influence the incidence of acute or chronic graft-versus-host disease (GvHD),[8] transplant-associated morbidity and mortality as well as disease relapse still pose challenges in elderly patients.[9,10] One study investigating a significant number of transplanted ALL patients aged over 45 years showed a substantially higher rate for transplant-related mortality (TRM) in MAC-treated patients with HLA-mismatches when compared to the RIC-treated cohort, prompting the authors to discourage MAC conditioning in this patient group altogether.[11] This observation suggests an interaction between transplantation-associated mortality caused by age-associated risk and HLA-mismatching. Age and HLA-matching status are important clinical predictors for the outcome of HSCT and are used among others for risk assessment in HSCT.[12] We analyzed the relationship between age-risk and HLA-risk in a large cohort of patients transplanted with unrelated donors and tested the hypothesis that age-risk varies according to HLA-matching status. Such a differentiation might have an impact on donor search and selection recommendations.

Methods

Patients

A total of 3019 adult patients transplanted for malignant hematologic disorders were included in this analysis. Transplantations were performed at German transplant centers between 1997 and 2011.

All patients received a first allogeneic unrelated transplant from bone marrow (BM) or peripheral blood stem cells (PBSC) with no more than 2 HLA-mismatches on 5-loci (HLA-A, -B, -C, -DRB1 and -DQB1). Disease stage definitions were adopted from a previous study defining the European Group for Blood and Marrow Transplantation (EBMT) risk score.[12] MAC was defined according to the recommendations of the EBMT Central Registry Office (MedAB manual forms).[13] Treatments with busulfan 16 mg/kg + cyclophosphamide 120–200 mg/kg, cyclophosphamide 120 mg/kg fractionated total body irradiation (TBI) 12Gy, etoposide VP-16 30–60 mg/kg + TBI 12Gy fractionated/10Gy single dose, BEAM polychemotherapy, CBV polychemotherapy or TBI 10–14Gy; busulfan 16 mg/kg are considered as myeloablative. Less intense regimens were considered as RIC. Patient and donor consent for HLA typing and for the analysis of clinical data were obtained. The study was approved by the ethical review board of the University of Ulm (project number 263/09).

HLA-typing

All patients and donors were high resolution typed for HLA-A, -B, -C, -DRB1 and -DQB1. Ambiguities within exons 2+3 for HLA-class I and exon 2 for HLA-class II alleles were resolved. Ambiguities involving non-expressed (null) alleles were resolved according to NMDP confirmatory typing requirements. Differences in exon 2 and 3 for HLA-class I alleles and exon 2 for HLA-class II alleles were considered as HLA-mismatch irrespective of the vector of mismatches.[14] Patient HLA-C KIR ligand status was inferred from high resolution HLA-C typing (C1=Asn80; C2=Lys80). Resulting phenotypes were C1C1, C1C2 and C2C2.

Statistical analysis

For univariate analysis of overall survival (OS), the Kaplan-Meier method and logrank testing was applied. Multivariate analysis for OS and disease-free survival (DFS) was performed using extended Cox-proportional hazards models.[15]

For TRM and RI, univariate competing risks analysis and multivariate competing risks regression for stratified data was used.[16] Backward stepwise exclusion was used for multivariate model selection. Evaluated covariates were: patient age, HLA-matching status, disease stage, conditioning regimen intensity, treatment with antithymocyte globulin (ATG), year of transplantation, time to transplantation, graft source, donor-recipient sex combination, KIR ligand status, and donor origin (national vs. international). For antithymocyte globulin (ATG) treatment, some data were missing (Table 1). Models were validated by inclusion of missing values as a separate group and by omission of cases with missing values, and no bias was found.[17]

Table 1.

Patients’ characteristics.

Stratification was used to account for heterogeneity of diagnosis. Violations of the proportional hazards assumption (PHA) by disease stage, conditioning regimen intensity and transplantation before 2004 were adjusted using time-dependent modeling of these covariates.[15] A significant center effect was adjusted using a frailty term with gamma distribution.[18] To assess the relationship between age and HLA-compatibility, subgroups were formed and analyzed as factors: age group 18–35 years (HLA-match: 10/10, 9/10 and 8/10), 36–55 years (HLA-match: 10/10, 9/10 and 8/10), and over 55 years (HLA-match: 10/10, 9/10 and 8/10). The cut-off value of 55 years for elderly patients has been used in previous studies and the cut-off value of 35 years is close to the arithmetic mean between the age boundaries in the remaining patients.[19]

In addition, an interaction model between age and number of HLA-mismatches was investigated. The relative risk conferred by age was visualized as age-dependent risk in different HLA-match categories relative to an 18-year old patient transplanted with a 10/10 matched donor as baseline. In this model, the covariate age was included as a continuous variable and no violation of the PHA was found. P=0.05 was considered statistically significant.

Results

Patients’ characteristics are given in Table 1. Patients over 55 years of age formed the second largest age group (n=1195, 39.6%). The distribution of diagnoses reflects the current spectrum of indications, with acute myeloid leukemia (AML) being the most frequent diagnosis (n=924, 30.6%). Single HLA-mismatches were present in 30.2% (n=911) and double mismatches occurred in 8.7% (n=261) of all patients. Although the proportion of HLA-DQ mismatches among double mismatched transplantations was slightly higher in older patients, there was no statistically significant difference in the distribution of 8/10 mismatches. Ethnicity was almost exclusively Caucasian. MAC was used in 62.1% (n=1875) of the patients, with peripheral blood stem cells (PBSC) being the leading graft source (n=2694, 89.2%). More than half of the transplantations were performed in the years between 2008 and 2011 (n=1671, 55.4%). Median follow-up time was 29 months. Table 2 and Figure 1 show the results of the univariate OS analysis in patients according to their HLA-matching status and age group. Logrank-testing showed no significant difference between 10/10, 9/10 and 8/10 matched transplantations in the youngest age group (aged 18–35 years). In the intermediate age group (36–55 years) a highly significant difference (P<0.001) was found with higher mortality for patients transplanted with single or double mismatches. In patients over 55 years of age, the differences were even more pronounced, showing high mortality, especially in the 8/10 matching group (P<0.001).

Table 2.

Univariate analysis in different age categories.

Figure 1.

Kaplan-Meier estimates for overall survival. Kaplan-Meier estimates for overall survival according to HLA-matching status (10/10 black lines, 9/10 blue lines, 8/10 red lines) in different age categories. (A) Age 18–35 years, P=not significant. (B) age 36–55 years, P<0.001. (C) age >55 years, P<0.001.

In multivariate modeling, these results could be confirmed for OS showing no significant differences between single and double mismatched transplantations in the younger age group (Table 3). Risk sharply increased with age in the respective mismatch groups, reaching the highest relative risk in the age group over 55 years (HR: 3.48, CI 2.49–4.86, P<0.001). Similar patterns were seen for DFS and TRM with hazard ratios spreading with increasing numbers of HLA-mismatches and increasing age, thus conferring highest risk for patients aged over 55 years with double HLA-mismatches [DFS: Hazard Ratio (HR) 2.74, CI 2.00–3.76, P<0.001 and TRM: HR 3.79, CI 2.29–6.30, P<0.001]. No significant differences were observed for relapse incidences.

Table 3.

Risk estimates for HLA mismatches according to age categories.

Modeling an interaction term between age and number of HLA-mismatches allowed estimation of age risk within matched, single-mismatched and double mismatched patient groups. Age risk showed increasing risk estimates with increasing number of HLA-mismatches. In 10/10 matched transplantations, this additional risk per life year at time of transplantation was lowest (HR: 1.015, CI 1.010–1.020; P<0.001). It increased, however, with the decreasing degree of HLA-compatibility between donor and patient (9/10, HR: 1.019, CI 1.014–1.024, P<0.001 and 8/10 HR: 1.026, CI 1.020–1.031, P<0.001). The interaction term for age and 2 HLA-mismatches was significant (P=0.009). The Cox regression model is a multiplicative hazard model. In order to visualize the component of age-risk within the respective HLA-match groups, the change of risk contributed to the prognosis by age at the time of transplantation was plotted relatively to an 18-year-old ‘baseline’ patient with a 10/10 matched donor. This visualization is based on the different age-associated risk estimates within each HLA-match category as observed in the multivariate model for OS, and it illustrates the change in risk with increasing age (Figure 2).

Figure 2.

Age risk by HLA-matching status. Relative risk contributed by the continuous covariate age at the time of transplantation according to different levels of HLA-mismatches (completely matched 10/10: black, single mismatched 9/10: blue, double mismatched 8/10: red).

Discussion

We found a statistically significant interaction between HLA-matching status and age-associated risk. This interaction can be interpreted as different levels of age-associated risk according to the number of HLA-mismatches. Our findings substantiate that transplantation for patients aged over 55 years with two HLA-mismatches are particularly risky with a highly significant hazard ratio of 3.48 (CI 2.49–4.86; P<0.001) when compared to 10/10 matched patients younger than 35 years. If compared to 10/10 transplantations within each age category, double mismatches increased mortality risk for OS by a factor of 1.14 in the lowest age group, by a factor of 1.40 in the middle age group, and 2.27 in patients aged over 55 years. This disproportional increase and the poor one-year survival rate of only 19% in double mismatched transplantations for elderly patients highlights the importance of HLA-matching especially in this group of patients.

Luckily, donors with 2 HLA-mismatches had to be accepted only in a small fraction of patients aged over 55 years (6.3%). The age cohorts showed expected structural differences in composition with regard to diagnosis and conditioning regimen, as well as graft source. Multivariate analysis adjusted for differences in conditioning treatment, while graft source showed no differential impact on survival end points.

It is known that older patients tolerate conditioning related toxicity less well than younger patients, which is the reason for the development and the use of conditioning regimes with reduced intensity.[6,20,21] Treatment-associated toxicity correlates strongly with transplant-related mortality and therefore it greatly influences OS. HLA-mismatches also associate strongly with treatment-related morbidity and -mortality. This relationship explains our findings from the perspective of transplant biology, suggesting that older patients tolerate HLA-mismatches less well than younger patients as it is also the case for treatment-related toxicity.

On the other hand, it cannot be deduced from this data whether younger patients benefit less from better-matched donors, as life expectancy is higher and HLA-associated risk cumulates over time.

This finding was only made possible because of the relatively high proportion of older patients in our dataset. As most of the transplantations were performed in the years between 2008 and 2011, our dataset reflects the substantial increase in elderly patients transplanted in Germany in recent years.

Other large studies investigating the impact of risk factors in HSCT contained significantly fewer older patients, which is why this interaction may have remained unnoticed in these studies.[22-24]

Interestingly, in the youngest age group, no significant difference was found between completely 10/10 matched transplantations and single or double mismatched transplantations. However, this age category was the smallest, consisting of only 17.5% of the cases, which limits interpretation of this particular result. Testing for proportional hazards assumption in our models showed no significant violation for the covariate age, which was treated as a continuous variable in the interaction model and in the prediction plot (Figure 2). Thus, the way we chose to visualize the disproportional increase in hazard ratios for age-risk at the time of transplantation is justified.

Our results were obtained from a cohort transplanted with allogeneic unrelated PBSC or bone marrow as a graft source. In our analysis, graft source did not differentially impact outcome, which is why no separate analysis for each graft source was made. Similar findings were reported in other studies.[25,26] Data on the impact of haploidentical transplantation or cord blood transplantations on the outcome of HSCT in elderly patients are very limited, so that a sensible risk-benefit comparison of our data with alternative graft or transplant sources is difficult. However, cord blood transplantation has been reported to result in similar outcomes in a small cohort of single mismatched transplantations in elderly patients treated with RIC.[27]

In multivariate analysis (Table 4), some predictors showed violation of the proportional hazards assumption (PHA). These violations can be explained by a higher early mortality for patients transplanted in advanced disease stage, transplanted before 2004 and treated with MAC. To reflect this relationship, an extended Cox regression model was fitted to obtain regression estimates for the respective predictors according to time periods where PHA is satisfied, as we have shown before.[28] In analysis of OS, advanced disease stage showed a substantially higher mortality risk until day 314 but not thereafter. Patients treated with RIC showed a significantly lower early mortality until day 96 and a non-significantly different risk afterwards. In addition, patients transplanted before 2004 showed a higher mortality risk until day 198 after transplantation but not thereafter. Similar findings were present in an analysis of DFS. In our models, also a patient C2C2 KIR-ligand status as well as an international donor status was associated with adverse outcome, which we have reported before.[29] ATG treatment was not included in the final models because it did not reach statistical significance.

Table 4.

Multivariate analysis.

Our analysis encompassed some simplifications, namely that any HLA-mismatch was considered equally. HLA-DPB1 mismatches were not included and the vector of mismatches was also not regarded.

We included HLA-DQB1 mismatches in this study, because a previous analysis on the same dataset has shown that these mismatches are associated with higher mortality risk.[29]

HLA-DPB1 mismatches have been shown to influence outcome of HSCT, but due to lower linkage disequilibrium, HLA-DPB1-mismatches in HLA-A, -B, -C, -DRB1 and -DQB1 matched and mismatched transplantations are almost equally distributed.[30] Therefore, we may assume that our results are not biased by not including HLA-DPB1. The vector of mismatches was not considered, because no significant differences in survival outcome have been seen for unidirectional mismatches when compared to bidirectional mismatches for the end points analyzed in our study.[14]

We refrained from including Karnofsky performance status and donor-recipient cytomegalovirus status due to the high proportion of missing data for these variables, which is a limitation of our analysis.

When selecting donors for elderly patients, the additional risk associated with HLA-mismatches in this age group should be considered. Especially when only donors with double HLA-mismatches are available for such a patient, the substantial risk conferred in this situation must be carefully weighed against the benefit of transplantation. Cord blood transplantation might be an alternative in such cases, although data regarding the impact of alternative graft sources for transplantation of elderly patients are still limited.