Haploidentical versus unrelated allogeneic stem cell transplantation for relapsed/refractory acute myeloid leukemia: a report on 1578 patients from the Acute Leukemia Working Party of the EBMT.

Primary refractory or relapsed acute myeloid leukemia is associated with a dismal prognosis. Allogeneic stem cell transplantation is the only therapeutic option that offers prolonged survival and cure in this setting. In the absence of a matched sibling donor, transplantation from unrelated 10/10 HLA allele-matched or 9/10 HLA allele-mismatched donors and haploidentical donors are potential alternatives. The current study aimed to compare the outcomes of acute myeloid leukemia patients with active disease who received allogeneic stem cell transplantation from a haploidentical donor with post-transplant cyclophosphamide (n=199) versus an unrelated 10/10-matched donor (n=1111) and versus an unrelated 9/10-mismatched donor (n=383) between 2007 and 2014 and who were reported to the European Society for Blood and Marrow Transplantation registry. Propensity score weighted analysis was conducted in order to control for disease risk imbalances between the groups. The leukemia-free survival rates at 2 years of recipients of grafts from a haploidentical donor, an unrelated 10/10-matched donor and an unrelated 9/10-mismatched donor were 22.8%, 28% and 22.2%, respectively (P=NS). In multivariate analysis, there were no significant differences in leukemia-free survival, overall survival, relapse incidence, non-relapse mortality, or graft-versus-host-disease-free relapse-free survival between the three groups. Two predictive factors were associated with a higher relapse incidence: transplantation during first or second relapse compared to primary refractory acute myeloid leukemia and poor cytogenetics. Allogeneic stem cell transplantation may rescue about 25% of acute myeloid leukemia patients with active disease. Importantly, the outcomes of transplants from haploidentical donors were comparable to those from 10/10-matched and 9/10-mismatched unrelated donors. Therefore, a haploidentical donor is a valid option for acute myeloid leukemia patients with active disease. Copyright

Introduction

After initiation of intensive chemotherapy for acute myelogenous leukemia (AML), failure to respond is a major unfavorable prognostic factor.[1,2] Obtaining a morphological complete remission (CR) after induction has been defined as a prognostic factor and even, until recently, considered as a prerequisite for allogeneic stem cell transplantation (HSCT). However, up to 30% of adults with newly diagnosed AML fail to achieve CR after two courses of intensive chemotherapy.[1] Moreover, once a first or second CR has been obtained, approximately half of younger patients and 80% of older patients relapse.[1,3,4] In both clinical situations, refractory and/or relapsed AML, active disease remains a major therapeutic challenge. Consequently, the accumulating evidence that HSCT can deliver long-term disease-free survival in a proportion of patients with AML with active disease represents an importance advance in the treatment in this very high-risk patient population.[5-9] Defining the impact of donor selection is still a major issue. It has been demonstrated that HSCT from matched sibling donors is a valid option, leading to a disease-free survival rate in the range of 20-30% for this very high-risk patient population.[7,10-13] More recently, HSCT from unrelated donors (UD) was used for patients with primary refractory AML, with an overall survival (OS) rate of about 22%.[13-15] Since 2010, the use of haploidentical HSCT has surged by about 300% among European Society of Blood and Marrow Transplantation (EBMT) centers.[16,17] Indeed, over recent years, haploidentical donors have been increasingly adopted as a valid source of donor cells for HSCT in AML in the absence of HLA-compatible matched sibling donors or matched UD. Based on several non-randomized comparative studies evaluating HSCT from haploidentical donors (Haplo HSCT),[18-21] the combined data suggest similar outcomes for Haplo and UD HSCT.[22,23] However, only small series are available for patients with resistant and/or relapsed AML undergoing HSCT from alternative donors. Craddock et al. reported that the OS rate of 36 patients who received an UD transplant with a reduced-intensity conditioning regimen (RIC) was 36% at 5 years, which was similar to that of 18 patients given myeloablative conditioning (MAC).[14] For patients with active leukemia, HSCT from alternative or mismatched donors may, theoretically, be of advantage, as HLA disparities may augment donor/recipient alloreactivity.

However, relatively few data are available for the very high-risk population of patients with refractory or relapsing AML transplanted from alternative donors while in active disease. In view of the fact that the development of Haplo HSCT is significantly influenced by the use of post-transplantation cyclophosphamide (PTCy), and because advances in supportive care influence outcomes, a safety and efficacy update comparison between Haplo and UD HSCT in a large cohort of patients with active disease is highly warranted to further support decision-making. With this aim, the present study, based on the EBMT - Acute Leukemia Working Party (ALWP) database, was conducted in order to compare outcomes of AML patients with active disease after Haplo HSCT versus 10/10 or 9/10 HLA-matched UD HSCT.

Methods

Study design and data retrieval

This is a retrospective, multicenter, registry-based analysis. Data for this study were provided and approved by the ALWP of the EBMT group registry. The EBMT registry is a voluntary working group of more than 600 transplant centers, mostly located in Europe, which are required to report all consecutive stem-cell transplantations and follow-up data once a year. Data are entered, managed, and maintained in a central database with internet access; each EBMT center is represented in this database. There are no restrictions on centers for reporting data, except for those required by law on patients’ consent, data confidentiality and accuracy. Quality control measures include several independent systems: confirmation of the validity of the entered data by the reporting team, selective comparison of the survey data with MED-A data sets in the EBMT registry database, cross-checking with national registries, and regular in-house and external data audits. All patients provided informed consent to the use of their data in retrospective studies. The Review Board of the ALWP as well as the ethics committee of the EBMT approved this study.

Eligibility criteria for this analysis included adult patients (aged >18 years) with active AML including primary refractory AML, AML in first relapse and second relapse who had received a first HSCT from a 10/10 HLA allele-matched UD (UD 10/10), or a 9/10 HLA allele-mismatched UD (UD 9/10) or a haploidentical donor (≥2 antigen mismatches or more out of 8) with post-transplant cyclophosphamide (Haplo PTCy) as graft-versus-host disease (GvHD) prophylaxis. Active AML was defined by the failure to achieve CR (bone marrow blasts >5%) despite induction chemotherapy. Cytogenetic groups were defined according to Grimwade et al.[24] The source of stem cells could be either bone marrow or granulocyte colony-stimulating factor-mobilized peripheral blood stem cells. All UD were HLA-matched (10/10) or mismatched at one loci (9/10) (-A, -B, -C, -DRB1, -DQB1). We excluded patients who had undergone umbilical cord blood HSCT, so that the analysis was restricted to a more homogeneous study population. MAC was defined, according to the EBMT, as a regimen containing total body irradiation with a dose >6 Gy, a total dose of oral busulfan >8 mg/kg or a total dose of intravenous busulfan >6.4 mg/kg.[25] The FLAMSA sequential conditioning regimen consisted of a combination of a short course of intensive chemotherapy using fludarabine 30 mg/m2/day, intermediate-dose cytosine arabinoside 2 g/m2/day, and amsacrine 100 mg/m2/day from day –12 to –9, followed, after a 3-day rest, by RIC using 4 Gy total body irradiation on day –5, cyclophosphamide 40 to 60 mg/kg/day on days –4 and –3, and antithymocyte globulin from days –4 to –2; the 4 Gy total body irradiation could have been replaced by a total dose of intravenous busulfan of 6.4 mg/kg (or an equivalent oral dose).[26-28]

Endpoints

OS was calculated from the date of transplantation until death or last observation alive. Leukemia-free survival (LFS) was calculated from the date of transplantation until relapse or last disease-free follow-up. Relapse and death from any cause were considered events. Non-relapse mortality (NRM) was defined as death without prior relapse. Neutrophil recovery was defined as achieving absolute neutrophil counts greater than or equal to 0.5×109/L for three consecutive days. The diagnosis and grading of acute[29] and chronic GvHD[30] were performed by transplant centers using the standard criteria. Cytogenetic abnormalities were classified according to Medical Research Centre criteria, and graft-versus-host-free, relapse-free survival (GRFS) as previously published.[31]

Statistical analysis

Patient-, disease-, and transplant-related variables were compared between the three groups (Haplo PTCy, UD 10/10, UD 9/10) using the chi-square test for categorical variables and the Mann-Whitney test for continuous variables. The median follow-up was estimated using the reverse Kaplan-Meier method. Variables considered were patient’s age at transplantation, donor/recipient gender, interval from diagnosis to transplantation, cytogenetic group, type of conditioning (RIC/MAC/FLAMSA), source of stem cells (peripheral blood stem cells versus bone marrow), patient/donor cytomegalovirus serology, Karnofsky Performance Status (KPS) at the time of transplantation, in vivo T-cell depletion, and year of transplantation. Factors that significantly differed between the three groups with P-values of <0.05, and all known as potential prognostic factors were included in the final models. Cumulative incidence functions were used to estimate relapse incidence (RI) and NRM in a competing risk setting, because death and relapse compete with each other. To study chronic GvHD, we considered relapse and death to be competing events. Probabilities of LFS and OS were calculated using Kaplan-Meier estimates. Univariate analyses were performed using the Gray test for cumulative incidence functions and the log-rank test for LFS and OS. Associations of patient and graft characteristics with outcomes were evaluated in multivariate analysis, using a Cox proportional hazards model. All tests were two-sided.

We used propensity score weighing to control for pretreatment imbalances in observed variables. The following factors were included in the propensity score model: patient’s age, time from diagnosis to transplantation, year of transplant, status at transplant, cytogenetic group, donor/patient cytomegalovirus serology, conditioning (RIC versus MAC), and gender matching (female donor to male recipient versus other). Propensity scores were estimated using generalized boosted models.

As the study question was whether Haplo PTCy could replace UD 10/10 or UD 9/10, we weighted the groups receiving either UD 10/10 or UD 9/10 HSCT to match the characteristics of patients receiving Haplo HSCT, by estimating the average treatment effect among the treated group (Haplo HSCT being the treated group). We then used pairwise average treatments to fit the weighted Kaplan-Meier and Cox models separately for Haplo PTCy versus UD 10/10 HSCT and Haplo PTCy versus UD 9/10 HSCT.

The type I error rate was fixed at 0.05 for determination of factors associated with time to events. Analyses were performed using the R statistical software version 3.2.3 (R Development Core Team, Vienna, Austria). Propensity score analysis was performed using the mnps function of the Twang package and weighted analyses using the survey package.

Results

Patients, disease and transplant characteristics

Data were obtained from 218 reporting centers (Online Supplementary Data). The patients’ and disease characteristics are summarized in Table 1. Of the total 1693 HSCT, 1111 were UD 10/10, 383 were UD 9/10 and 199 were Haplo PTCy. The three cohorts of patients differed for several variables (Table 1). The median follow-up was longer for the UD 10/10 and the UD 9/10 groups than for the Haplo-PTCy group. The follow-up completeness index at 2 years, which is the ratio of total observed person-time and the potential person-time of follow-up at 2 years[32] was 73% for Haplo PTCy, 76% for UD 10/10 and 80% for UD 9/10. Significantly more patients received MAC regimens in the Haplo PTCy group in comparison to both the UD 10/10 and the UD 9/10 groups. There were more cytomegalovirus-positive recipient-donor pairs in the Haplo-PTCy group. Peripheral blood stem cells were, as expected, the main source of stem cells in the UD 10/10 and UD 9/10 groups, while peripheral blood stem cells represented 52.8% of the stem cell source in the Haplo-PTCy group.

Table 1.

Baseline characteristics of patients.

Engraftment and graft-versus-host disease

The cumulative incidence of engraftment at day 30 was 85.5% [95% confidence interval (95% CI): 79-90.2], 92.3% (95% CI: 90.5-93.7) and 92.2% (95% CI: 88.9-94.6) in the Haplo PTCy, UD 10/10 and 9/10 groups, respectively (P<10−3 for both comparisons).

Lower incidences of all acute GvHD grades were observed after Haplo PTCy than after UD 9/10. The cumulative incidences of grade II–IV acute GvHD were 28.2% and 36.3%, respectively (P=0.03) and those of severe grade III–IV acute GvHD were 8.9% and 16.1%, respectively (P=0.02) (Table 3). No difference was observed in the incidence of grade II-IV acute GvHD between the Haplo PTCy and UD 10/10 groups (P=NS).

Table 3.

Univariate analysis for acute graft-versus-host disease and chronic graft-versus-host disease.

At 2 years, the cumulative incidence of chronic GvHD was lower in the Haplo PTCy group than in the UD 9/10 group (19.3% and 27.4%, respectively, P=0.04), while no difference was found in the incidence of chronic GvHD between the Haplo PTCy and UD 10/10 groups (P=NS) (Table 3). The cumulative incidence of extensive chronic GvHD was similar in the three groups of patients (Haplo PTCy - 11%, UD 10/10 -11.6% and UD 9/10 -11.6%).

The percentages of patients who achieved CR within 100 days were 79.7%, 77% and 78.3% in the Haplo PTCy, UD 10/10 and UD 9/10 groups, respectively (P=NS). Performing a landmark analysis at day 100 for comparing outcomes between patients who achieved CR before day 100 to those who did not indicated that CR seems to be a surrogate marker for subsequent outcome. The probability of being alive and free of disease 1 year after HSCT was only 13.7% for patients who did not achieve CR before day 100 (data not shown).

We also analyzed chronic GvHD as a time-dependent variable demonstrating that the association of chronic GvHD with a lower RI [hazard ratio (HR)=0.77, 95% CI: 0.60-0.99, P=0.04) was counterbalanced by a higher NRM (HR=1.98, 95% CI: 1.38-2.84, P<10−3), and thus did not translate into better LFS (P=NS).

Leukemia-free survival, overall survival, relapse incidence and non-relapse mortality

In univariate analysis, the LFS rate at 2 years was 22.8% in the Haplo PTCy group versus 28% in the UD 10/10 group and 22.2% in the UD 9/10 group (P=NS) (Figure 1A, Online Supplementary Table S1). Multivariate analysis showed lower LFS rates in patients with poor cytogenetics (HR=1.35, 95% CI: 1.11-1.62, P=0.002), those transplanted in second relapse in comparison to those transplanted in primary refractory AML (HR=1.31, 95% CI, 1.03-1.65, P=0.03), and in patients transplanted from cytomegalovirus seropositive donors (HR=1.22, 95% CI: 1.05-1.41, P=0.01). In contrast, better LFS was associated with KPS ≥90 at transplantation (HR=0.66, 95% CI: 0.58-0.76, P<10−3), with a shorter time from diagnosis to transplantation (HR=0.99, 95% CI: 0.98-0.99, P=0.02), and with RIC in comparison to MAC (HR=0.84, 95% CI: 0.71- 0.99, P=0.03). Of note, no effect was observed for donor type (Table 4).

Figure 1.

Leukemia-free survival, overall survival, relapse incidence and non-relapse mortality in patients with acute myeloid leukemia allografted during active disease. (A) The 2-year probability of leukemia-free survival (LFS) was 22.8% (95% CI: 16.3-29.2) in the group transplanted from a haploindentical donor with post-transplant cyclophosphamide (Haplo) versus 28% (95% CI: 25-30.9) in the 10/10 HLA-matched unrelated donor group (UD 10/10), and 22.2% (95% CI: 17.6-26.7) in the 9/10 HLA-mismatched unrelated donor group (UD 9/10) (P=NS). (B) The 2-year probability of overall survival (OS) was 29.3% (95% CI: 22.1-36.6) in the Haplo group versus 34.7% (95% CI: 31.5-37.8) in the UD 10/10 and 27.6% (95% CI: 22.7-32.5) in the UD 9/10 groups (P=NS). (C) The 2-year cumulative incidence of relapse (RI) was 52% (95% CI: 44.3-59.1) in the Haplo group versus 46.3% (95% CI: 43.1-49.4) in the UD 10/10 and 51.1% (95% CI: 45.7-56.3) in the UD 9/10 groups (P=NS). (D) The 2-year cumulative incidence of non-relapse mortality (NRM) was 25.3% (95% CI: 19.2-31.8) in the Haplo group versus 25.7% (95% CI: 23.1-28.5) in the UD 10/10 and 26.7% (95% CI: 22.2-31.4) in the UD 9/10 groups (P=NS).

Table 4.

Multivariate analysis for leukemia-free survival, overall survival, relapse incidence, non-relapse mortality and graft-versus-host disease-free, relapse-free survival.

The OS rate at 2 years did not differ between the three groups of patients (29.3% in the Haplo PTCy group versus 34.7% in the UD 10/10 group and 27.6% in the UD 9/10 group, P=NS) (Figure 1B). These results were confirmed by multivariate analysis. In the latter, three predictive fac tors were associated with lower OS: disease status (second relapse versus primary refractory AML), poor cytogenetics and the patient being positive for cytomegalovirus, whereas KPS ≥90 at transplant, RIC versus MAC and shorter time from diagnosis to transplantation were associated with a better OS (Table 4).

We did not find any differences in terms of RI between the three groups (Table 2, Figure 1C). This result was also confirmed in multivariate analysis showing that poor cytogenetics and disease status (first and second relapse versus primary refractory AML) were the only risk factors associated with increased RI, whereas KPS, age at transplantation and time from diagnosis to transplantation were protective factors (HR=0.76, 95% CI: 0.64-0.90, P=0.001; HR=0.92, 95% CI: 0.87-0.98, P=0.007; and HR=0.98, 95% CI: 0.97-0.99, P<10−3, respectively) (Table 4).

Table 2.

Transplantation outcomes.

No differences in NRM were noted between the three groups of patients in univariate analysis (Table 2, Figure 1D). Multivariate analysis demonstrated that patients’ age (per 10 years) and cytomegalovirus positivity were associated with higher NRM (HR=1.18, 95% CI: 1.08-1.28, P<10−3; and HR=1.37, 95% CI: 1.07-1.77, P=0.01, respectively), while RIC compared to MAC and KPS ≥90 were associated with lower NRM (HR=0.64, 95% CI: 0.49-0.84, P=0.001; and HR=0.52, 95% CI: 0.41-0.64, P<10−3; respectively) (Table 4). Notably, no effect was observed for the type of donor.

In addition, no significant differences were found in GRFS according to donor type in the multivariate analysis. Three factors were associated with a better GRFS: longer time from diagnosis to transplantation, RIC versus MAC, and a KPS ≥90. Patients with poor cytogenetics had a lower GRFS (Table 4).

As shown in Online Supplementary Table S2, most events happened within the first year after HSCT.

Finally, in order to reduce the effects of confounding factors, we performed a weighted analysis on propensity scores (weighted average treatment). The results of the weighted Kaplan-Meier and Cox analyses confirmed the study results as described in Table 5. In the weighted analysis on propensity scores, the Haplo PTCy group had a significantly lower incidence of grade III-IV acute GvHD compared to that of patients in the UD 10/10 group (8.9% versus 14.5%, respectively, P=0.04), as confirmed by Cox analysis (P=0.049).

Table 5.

Weighted Cox model for leukemia-free survival, overall survival, relapse incidence, non-relapse mortality, graft-versus-host disease-free, relapse-free survival and graft-versus-host disease.

Causes of death

Leukemia was the most common cause of death (accounting for 50% of the deaths in the Haplo PTCy group, 54% in the UD 10/10 group, and 54.5% in the UD 9/10 group). GvHD was the second most common, being the cause of death in 11.5% of the patients in the Haplo PTCy group, 12.1% in the UD 10/10 group, and 15.2% in the UD 9/10 group. Infection was the cause of death in 27%, 20.8%, and 20.6% of the patients in the Haplo PTCy, UD10/10 and UD 9/10 groups, respectively.

Discussion

In the present study, we compared the transplantation outcomes after Haplo HSCT with PTCy versus transplantation from matched (10/10) or mismatched (9/10) UD in AML patients with active disease. The LFS rate was about 25% and the OS rate about 30% after HSCT for this high-risk population with advanced disease with no difference between Haplo PTCy and UD 10/10 or 9/10. Rates of acute GvHD grades II-IV and of chronic GVHD were similar between Haplo PTCy and UD 10/10, and the same held true for the 2-year NRM. It is important to note that higher rates of grades II-IV acute GvHD and chronic GvHD were observed in the UD 9/10 group without there being an impact on RI. Although we could hypothesize a stronger graft-versus-leukemia effect after Haplo PTCy than after UD HSCT, we did not observe differences in terms of RI or LFS. This finding is in accordance with those of our previous study, in which we compared relapse rates between patients with primary refractory AML undergoing allogeneic transplantation from unrelated versus sibling donors and found no difference.[13] One hypothesis is that, besides the very strong tolerance induction mediated by the PTCy in the Haplo setting in the case of active leukemia, the very aggressive biology of the disease and its refractoriness to several lines of chemotherapy lead to fast disease progression and relapse early after transplantaion. Thus, the graft-versus-leukemia effect, even across broad HLA disparities, is too weak or too slow to control the leukemia.

Of note, about 37% of patients received a RIC regimen in our study. As expected, the NRM rate was significantly lower in the RIC group than in the MAC group, with no difference in RI between the two groups. In all, LFS, OS, and GRFS were significantly better after RIC than after MAC. We could hypothesize that this is because even a high intensity conditioning regimen does not have a strong impact on this chemo-refractory leukemia. In our current study, 384 of the patients received a sequential approach with aplasia-inducing chemotherapy followed by the conditioning regimen. However, the question of which treatment should be used in a given patient cannot yet be answered. In a recent meta-analysis of leukemic patients with induction failure, Wattad et al. concluded that HSCT without prior salvage chemotherapy and HSCT in CR after salvage therapy resulted in comparable survival outcomes, and both strategies were significantly superior to HSCT performed after failure of salvage therapy.[33] In the present study, no differences in outcome were found between patients who received MAC or sequential regimens. One hypothesis to explain this is that the refractoriness of the malignant leukemic clone to chemotherapy makes the conditioning regimen unable to induce remission, or even a transient response allowing sufficient time for the alloreactive cells to mediate the graft-versus-leukemia effect.[34]

Importantly, an interval from diagnosis to transplant longer than the median was a negative prognostic factor for LFS, OS, RI and GRFS in multivariate analysis. These data, which are consistent with those of a study by Craddock et al.[14] and our previous results,[13] highlight the urgent need to search for a donor for AML patients with active disease who lack a matched sibling donor. Haploidentical donors are available for the majority of patients, providing access to further stem cell donations or donor lymphocyte infusions as needed.[35] Furthermore, it is a factor on which the physicians can have an influence, unlike many other factors. In accordance, Wattad et al. showed that patients transplanted during refractory disease after salvage therapy had a significantly poorer outcome compared to that of patients who proceeded directly to transplantation, and those transplanted in CR after salvage therapy.[12] Thus, we could recommend not to try additional lines of chemotherapy to achieve CR in patients with active disease, but to take advantage of the first available donor in order to proceed to transplantation.

The cytogenetic characterization of the leukemia represents a major prognostic factor for LFS, RI, OS, and GRFS.[36,37] We previously reported that poor-risk cytogenetics was an adverse pre-HSCT variable in patients with primary refractory AML who underwent HSCT with a graft from a matched sibling or matched UD.[13] Accordingly, in the present study, primary refractory AML with poor cytogenetic characteristics was associated, at 2 years, with a significant decrease in LFS and OS, and increase in RI. These data pave the road for investigating additional approaches relying on sequential conditioning regimens and/or post-transplant treatments.[38-42]

Being retrospective and registry-based, this study has some limitations: several of the patients’ characteristics differed between the groups. We addressed this limitation, at least in part, by using the propensity score technique. In addition, there was a relative inherent selection process for HSCT in our study and a relative lack of infor mation on the reasons for an EBMT center allocating patients to HSCT from a haploidentical donor versus UD, so that distinguishing the choice of the donor from the role of a potential center effect is difficult. Finally, the counts of circulating and bone marrow blasts at the time of HSCT were missing for a substantial number of patients. However, the aim of this analysis was to compare the two types of donors using EBMT registry data. The design of the study and inclusion criteria were intended to answer this clinical question and were not, therefore, adapted for developing a prognostic score. There are ongoing trials aiming to compare outcomes after Haplo PTCy versus UD in hematologic malignancies, but they do not focus specifically on the setting of active AML disease (NCT02623309). Therefore, in the absence of any prospect of such comparative studies, our data suggest that haploidentical donors are equally effective as 10/10 matched and 9/10 mismatched UD for allogeneic transplantation in patients with active AML.

In conclusion, our results indicate that, when an HLA-identical sibling donor is not available for an AML patient with active disease who is, otherwise, a candidate for HSCT, a haploidentical donor may be used with the expectation of similar rates of NRM, LFS, OS, and GRFS at 2 years, compared with those achieved with 10/10 matched and 9/10 mismatched UD.