Inferior outcome of allogeneic stem cell transplantation for secondary acute myeloid leukemia in first complete remission as compared to de novo acute myeloid leukemia

Following chemotherapy, secondary acute myeloid leukemia (sAML), occurring after antecedent hematologic diseases, previous chemotherapy or radiation, has an inferior prognosis compared with de novo AML. To define the outcome of sAML in the context of allogeneic stem cell transplantation (alloSCT), a retrospective, registry-based comparison was performed, including 11,439 patients with de novo and 1325 with sAML. Among transplants in first complete remission (CR1) (n = 8,600), the 3-year cumulative incidence of relapse (RI) and non-relapse mortality (NRM) was 28.5% and 16.4% for de novo, and 35% and 23.4% for sAML. Three-year overall survival (OS), leukemia-free survival (LFS) and Graft-versus-Host Disease/relapse-free survival (GRFS) was 60.8%, 55.1%, and 38.6% for de novo, and 46.7%, 41.6%, and 28.4% for sAML, respectively. In multivariate analysis, sAML was associated with a lower OS (HR = 1.33 [95% CI = 1.21–1.48]; p < 10−5), LFS (HR = 1.32 [95% CI = 1.19–1.45]; p < 10−5) and GRFS (HR = 1.2 [95% CI = 1.1–1.31]; p < 10−4) and higher NRM (HR = 1.37 [95% CI = 1.17–1.59]; p < 10−4) and RI (HR = 1.27 [95% CI = 1.12–1.44]; p < 10−3). Results of the Cox model were confirmed in a matched-pair analysis. In contrast, results did not differ between de novo and sAML after alloSCT in induction failure or relapse. Hence, this analysis identified sAML as an independent risk factor for outcome after alloSCT in CR1.


Introduction
Secondary acute myeloid leukemia (sAML) arises from underlying hematological disorders such as myelodysplastic syndrome (MDS), myeloproliferative neoplasm (MPN), MPN/MDS overlap or bone marrow failure syndromes. Furthermore, therapy related AML (tAML), which can develop after a prior treatment by chemotherapy or radiation, can be assigned to the sAML classification 1 .
Treating sAML with conventional chemotherapy or autologous stem cell transplantation is associated with a poor prognosis [2][3][4] . Therefore, in comparison to de novo AML, sAML is thought to have an inferior prognosis in general [5][6][7] . Allogeneic stem cell transplantation (alloSCT) is regarded as the treatment option for sAML with the best chance of achieving a long-term remission. Considering the transplant setting 8,9 , patients with sAML tend to have less available HLA identical siblings, and the majority receive reduced intensity conditioning (RIC) 10,11 . However, it is not clear whether having sAML per se is a risk factor for outcome after alloSCT, when all wellknown risk factors are controlled for 8 .
So far, only smaller studies have addressed this question 12 . Therefore, the Acute Leukemia Working Party (ALWP) of the European Society of Blood and Marrow Transplantation (EBMT) performed a retrospective registry-based analysis, comparing the outcome of adult patients diagnosed with sAML and de novo AML, who had received alloSCT.

Study design
Data were provided by the EBMT registry, which comprises more than 600 transplant centers providing reports and annual follow-up on all consecutive stem cell transplantations. Audits are routinely performed to determine the accuracy of the data. Since 1990, patients have provided informed consent, authorizing the use of their personal information for research purposes. The study was approved by the general assembly and review board of the ALWP.
Inclusion criteria were age ≥ 18 years, alloSCT between January 2000 and December 2016 from a matched related, 9/10 or 10/10 antigen matched unrelated, or a T-cell replete haploidentical donor for either de novo or sAML, and available information on cytogenetics. The number of patients fitting inclusion criteria was 11,439 patients with de novo AML and 1325 with sAML.

Definitions
Secondary AML was defined as AML with an antecedent MDS, MPN or other malignant hematologic disorder, bone marrow failure syndrome, or solid tumor with prior chemotherapy or radiation 8 . For a subgroup analysis, tAML was defined by previous treatment of solid tumors, Hodgkin or non-Hodgkin Lymphoma (NHL), acute lymphoblastic leukemia (ALL) or chronic lymphocytic leukemia (CLL) using chemotherapy or radiation 14 .
As recommended complete remission (CR) was defined by <5% bone marrow (BM) blasts, absence of circulating blasts and extramedullary disease. Failure to achieve CR after two courses of standard induction chemotherapy was defined as primary induction failure (PIF). Relapse was defined by more than 5% BM blasts or reappearance of circulating blasts after a documented CR 15 . OS was defined as the interval between day of alloSCT and day of death or last follow-up, LFS as interval between alloSCT and date of leukemia persistence, relapse or progression. NRM was defined as death from any cause without relapse or progression. GRFS was defined as absence of acute GVHD III-IV, chronic GVHD requiring systemic treatment, relapse, or death 16,17 . Reduced intensity conditioning (RIC) was defined using EBMT criteria 18 . Cytogenetic subgroups were defined according to SWOG criteria 19 .

Statistical analysis
Descriptive statistics used median, inter-quartile range (IQR), minimum (min) and maximum (max) for continuous data, counts and percentages for categorical variables. Patient, disease, and transplant-related characteristics for the two cohorts (de novo or secondary AML) were compared by using χ 2 statistics for categorical variables and the Mann-Whitney test for continuous variables. The date of transplantation was the starting point for time-to-event analysis. Survivors were censored at last contact. Cumulative incidence was used to estimate the endpoints of NRM, RI, acute and chronic GVHD to accommodate for competing risks. Relapse and death were considered competing events for acute and chronic GVHD. Probabilities of OS, LFS, and GRFS were calculated using the Kaplan-Meier method. Univariate analyses were done using the Gray's test for cumulative incidence functions and the log rank test for OS, GRFS, and LFS.
A multivariate Cox proportional-hazards model was performed to account for imbalances of risk factors between the two cohorts. Due to a significant interaction between disease stage at transplant and diagnosis (de novo or sAML), the analysis was stratified by stage at alloSCT. All variables associated with a significant outcome in univariate analysis (p < 0.05), factors known from the literature to have a potential influence on outcome, and variables not equally balanced between cohorts were included in the Cox model.
To confirm the results from the Cox model, a matchedpair analysis of secondary versus de novo AML was performed, using the following criteria for matching: Age (±3 years), cytogenetics, stage at alloSCT, KPS, conditioning, in and ex vivo TCD, donor type, donor/recipient gender and CMV serostatus combinations, and graft source.
Results were expressed as hazard ratio (HR) and 95% confidence intervals (CI). Proportional hazards assumptions were checked for all models using the Grambsch-Therneau residual-based test. All tests were 2-sided. The Type I error rate was fixed at 0.05 for the determination of factors associated with time-to-event outcomes. Statistical analyses were performed with SPSS 24.0 (SPSS Inc, Chicago, IL) and R 3.4.1 (https://www.R-project.org/).

Patients' disease and transplant characteristics
Using the criteria defined above, 11,439 patients with de novo AML and 1325 with sAML were included. Median follow-up after alloSCT was 36.1 and 33.1 months for de novo and sAML, respectively. Patients' disease-and transplant-related characteristics are outlined in Table 1 Comparison between de novo and secondary AML As mentioned above, a significant interaction between disease status at transplant and diagnosis (de novo or sAML) was observed. Therefore, the analysis of the role of sAML had to be stratified according to disease status at
In a multivariate Cox model, sAML was associated with a lower OS (HR = 1. The inferior outcome of patients with sAML was also observed in an exploratory subgroup analysis of patients below and above the age of 60 years (data not shown).

Separate comparison for AML evolving from myeloid malignancies/bone marrow failure syndromes and treatment-related AML
To evaluate a potential difference among patients with sAML evolving from myeloid malignancies or bone marrow failure syndromes versus treatment-related AML, a second Cox model was fitted, including the two subtypes of sAML as covariates. Both sAML after myeloid malignancies/BM failure and tAML were associated with increased OS, LFS, and GRFS following alloSCT in CR1 when compared to de novo AML (supplemental Table S8).

Transplantation in primary induction failure or active relapse
Among patients with sAML, 199 (15.02%) and 124 (9.36%) had been transplanted in PIF and relapsed disease, respectively; in de novo AML, 607 (5.3%) had PIF, and 1009 (8.82%) had relapsed disease at time of alloSCT. In contrast to the cohort transplanted in CR1, sAML was not identified as a relevant factor for OS (PIF: HR = 0.92; p = 0.465; relapse: HR = 0.95; p = 0.64; cf. Tables 2 and 3), nor for RI, NRM, LFS, and GRFS. Risk factors that affected OS, RI, LFS, and GRFS in the cohort transplanted for active disease included cytogenetics and KPS > 80% (cf. Table 3 and supplemental Tables 4-7 for details).

Role of sAML for GVHD
Independently of disease status before transplantation, sAML was not a risk factor for the occurrence of acute or chronic GVHD. Risk factors for acute GVHD grade II-IV across all subgroups were donor type, female donor for male patient and in vivo TCD. In the CR1 cohort, among others, age, cytogenetics, KPS, and conditioning regime were additional factors. Risk of chronic GVHD was mainly determined by graft source and in vivo TCD. In CR1, age, female donor to male recipient and donor type were additional factors (supplemental Tables 9-10 or details).

Matched-pair analysis
To confirm the results from the Cox model, a matchedpair comparison was performed between de novo and sAML. Using the criteria mentioned above, 877 wellmatched pairs were identified (cf. supplemental Table  S11). The matched-pair analysis confirmed the findings of the Cox model: Overall, sAML was associated with a lower 3-year cumulative incidence of OS of 46.9% (95% CI, 43.3-50.5) compared to 53.9% (95% CI, 50.1-57.7) in de novo AML (p = 0.01). Furthermore, patients with sAML had a lower LFS (p = 0.03), GRFS (p = 0.04), and a higher NRM (p = 0.01). However, when the analysis was stratified by disease status at alloSCT, again differences were only seen among patients transplanted in CR1 (719 pairs, 82%). The 3-year OS in the CR1 sAML and in the de

Discussion
In this large-scale, registry-based analysis, a multivariate model identified sAML as an independent risk factor for OS, LFS, GRFS, RI, and NRM after alloSCT in CR1, as compared to de novo disease. A matched-pair analysis confirmed these findings, which were observed both in treatment-related AML and AML evolving from other myeloid disorders or BM failure syndromes. In contrast, in patients with active disease at alloSCT, with relapse being by far the most frequent cause of treatment failure, the classical risk factors such as cytogenetics, age and KPS determined outcome. Secondary AML had no influence on outcome in the context of the aggressive nature of advanced disease.
Patients diagnosed with sAML tend to be of older age, carry more unfavourable cytogenetics and show poor response to chemotherapy 20,21 . Regardless of cytogenetics and preceding therapy of antecedent haematological diseases, sAML is known to have a poor long-term remission even when using aggressive induction therapy 22,23 . Separate studies on patients with treatment-related AML following intensive chemotherapy, identified tAML as an independent adverse prognostic factor 2,7,14,24 . However, data on the role of sAML as a risk factor for outcome after alloSCT are scarce. Michelis et al. reported on a total of 264 patients with sAML and de novo AML, transplanted in CR1 12 . Age, hematopoietic cell transplant comorbidity index, and karyotype were considered for matching. In contrast to our results, no clear influence of having sAML could be demonstrated regarding OS, NRM, LFS, and RI after alloSCT, using multivariate analysis and a propensity score model. It can be speculated, that the lower patient numbers precluded Michelis at al. from detecting a statistically significant difference (p = 0.18), given that the 3-year OS in their cohort was 55% for de novo and 46% for sAML.
Besides the classical risk factors, i.e., cytogenetics and age, as well as donor type, our data identify sAML as an independent intrinsic factor for inferior outcome after alloSCT in CR1. Both patient and disease specific factors might contribute to this finding. First, characteristics associated with NRM (i.e., age, HLA mismatch, female donor to male patient, CMV seropositivity) had a high impact on OS in patients with sAML. This observation could suggest a lower ability to generally tolerate side effects and complications in the context of alloSCT, such as toxicity of the conditioning and immunosuppression, infections and GVHD. It might be a consequence of prior chemotherapy given for the antecedent hematologic or solid organ tumors, and might not be reflected by the performance score at time of alloSCT among patients with sAML 21,25 . Second, given the higher number of patients with sAML who have adverse cytogenetics, indicating lower sensitivity to standard chemotherapy and hypomethylating agents, the quality of CR might generally be inferior in sAML as compared to de novo disease. A better remission quality, however, is decisive for the outcome after alloSCT in sAML 26 , as well as de novo AML 27,28 . Third, the more aggressive nature of a secondary malignancy as such might play a role in the inferior outcome of patients with sAML, when transplanted in CR1.
Previous reports have suggested, that AML following antineoplastic radiation or chemotherapy (tAML) be considered an independent entity 25 . Although there are no pathognomonic genetic aberrations known for tAML, this subgroup had a worse outcome than MDS-related AML in several studies 2,7,14,24 . In a previous study from our group, slight differences in NRM and survival had been observed between patients with AML evolving from MDS/MPN and other hematologic malignancies, but not solid tumors 8 . In the present analysis, we therefore repeated the Cox model including tAML and AML evolving from another hematologic disease, as covariates. Both subgroups achieved comparable outcomes and were associated with increased OS, LFS and GRFS following alloSCT in CR1. Hence, as suggested earlier, the described two subtypes might be grouped as sAML in the setting of alloSCT 12 , although our numbers were to small to compare outcome of sAML following MDS versus sAML versus MPN. Due to the retrospective nature of our study, some limitations, including missing information on MRD status and pre-and post-transplant therapies in most patients, must be acknowledged. In addition, there is a risk of selection bias, as patients without information on cytogenetics were deliberately excluded from this analysis, considering the high relevance of cytogenetic subgroups for outcome. Nevertheless, overall outcome data in our cohort are comparable to data from other studies 8 , suggesting that the cohort was representative. It must also be acknowledged that information on molecular aberrations was available only in a minority of patients from both cohorts and could therefore not be considered. However, in contrast to cytogenetic aberrations, which without doubt have similarly strong prognostic value in both de novo and sAML, less is known on the comparable value of molecular markers in both subgroups, which is why inclusion of these markers in this comparison would be difficult. Further, in the matched-pair analysis, patients with sAML and especially those in CR2 are most probably underrepresented, and fewer patients than expected could finally be included in the model. Therefore, the matchedpair analysis can only be regarded as confirmatory for the overall results obtained from the Cox model, which included all patients, and failed to reproduce the increased risk of relapse in sAML, most likely due to limited numbers.
This large-scale retrospective study identified sAML as an independent risk factor for alloSCT in CR1. Both the inferior quality of CR and higher sensitivity to toxicity might have contributed to the inferior outcome of sAML. These data help to improve risk stratification and prognostic estimates after alloSCT for sAML. Furthermore, the results may contribute to the design of optimized transplant protocols. Previous studies have suggested comparable outcomes after RIC and MAC in sAML, however, with a trend for better outcome after MAC 11,29 . Hence, a myeloablative, but reduced toxicity conditioning regimen such as the combination of Fludarabine and Treosulfan 30 might be of particular value in patients with sAML. In addition, strategies to improve the quality of remission prior to alloSCT in those patients with frequently low sensitivity to chemotherapy might be of particular importance 31 .