Dear Editor,

Allogeneic hematopoietic stem-cell transplantation (allo-HCT) is a curative treatment for patients with acute myeloid leukemia (AML), typically offered to patients in complete remission (CR) but at high risk for relapse [1, 2]. However, up to 40% of patients present with primary refractory disease or relapsed disease failing to achieve CR [3]. Their prognosis remains poor, with a 5-year overall survival (OS) as low as 10% [4, 5]. Prior studies of allo-HCT in this context primarily focused on myeloablative regimens and did not consider genetic mutations of leukemic cells [6]. With advances in allo-HCT practices, these studies may now be outdated and warrant reassessment.

In this study, we aimed to evaluate the role of allo-HCT for relapsed/refractory AML at two large academic centers in the United States in recent years. We conducted a retrospective analysis of allo-HCT recipients with AML who were not in CR at the time of transplantation between 2009 and 2020. Our cohort consisted of 151 adult patients with AML who underwent allo-HCT while not in CR at the time of transplant (Table 1). The median blast count in the bone marrow was 10% (range 0–94), with 33% of patients having circulating blasts, and 7.9% having extramedullary disease. We categorized the patients into three groups, as described in the Supplementary Methods section: not in complete remission (NCR; n = 67, 44.4%), primary induction failure (PIF; n = 61, 40.4%), or untreated relapse/relapse refractory (UR/RR; n = 23, 15.2%). Most (n = 117, 77%) transplants utilized matched donors, and grafts were mainly unmanipulated peripheral blood stem cells (n = 127, 89%). The disease risk at diagnosis was classified as intermediate in 36% and poor in 61%, according to the National Comprehensive Cancer Network (NCCN) criteria (Version 3.2021) (Table 1) [1]. Molecular testing data were available only for the MSK cohort, with 30% of patients having at least one of the following mutations at diagnosis: TP53, RUNX1, ASXL1, or FLT3-ITD (Supplementary Table 1). Myeloablative conditioning (MAC) was administered to 64% of patients, while 36% received reduced intensity (RIC) or non-myeloablative (NMA) conditioning. The most common GVHD prophylaxis across all graft types was a combination of a calcineurin inhibitor and methotrexate (61%).

Table 1 Patient characteristics.
Full size table

The 3-year probability of OS from transplant among all patients was 24% (95% CI: 18–32) (Fig. 1A). OS did not differ by disease category: NCR, PIF, or UR/RR (Fig. 1B) or between patients transplanted pre- vs. post-2015. The most common causes of death were disease progression or relapse (64%) and GVHD (with or without a concurrent infection, 15%) (Supplementary Table 2). In a univariable analysis, donor type, donor–recipient CMV serostatus, recipient CMV serostatus alone, and the presence of circulating blasts were associated with OS. After adjusting for these covariates in a multivariable model, donor type, positive recipient CMV serostatus, and the presence of circulating blasts remained significantly associated with worse OS (Supplementary Table 3). Donor–recipient CMV serostatus was excluded from the model due to its overlap with recipient CMV serostatus. Kaplan–Meier plots illustrating OS according to the significant univariable covariates, circulating blasts, and recipient CMV serostatus are shown in Fig. 1C, D, respectively.

Fig. 1: Clinical outcomes following allo-HCT in patients not in complete remission (CR).
figure 1

Overall survival (OS) analysis: A OS of the entire cohort; B OS stratified by disease category; C OS stratified by the presence of circulating blasts; D OS stratified by recipient CMV serostatus; E cumulative incidence of relapse, stratified by disease categories; F non-relapse mortality (NRM); G incidence of acute graft-versus-host disease (GVHD) grades II–IV; H incidence of acute graft-versus-host disease (GVHD) grades III–IV.

Full size image

The cumulative incidence of relapse/POD among all patients at 3 years post-transplant was 58% (95% CI: 50–65), and relapse/POD did not differ significantly across disease categories (NCR vs. PIF vs. UR/RR, Fig. 1E). The median time from transplant to relapse, calculated among patients who experienced relapse, was 3.8 months (IQR 2.4–7.4). In a univariable analysis, donor–recipient CMV serostatus, recipient CMV serostatus, and the presence of circulating blasts were significantly associated with relapse/POD, whereas the presence of a complex karyotype showed a borderline association with relapse (p = 0.051) (Supplementary Table 4). In addition, the presence of at least one alteration in TP53, RUNX1, ASXL1, or FLT3-ITD was borderline associated with an increased hazard of relapse/POD in the univariable analysis. After adjusting for the presence of circulating blasts and recipient CMV serostatus in a multivariable model, both covariates remained significantly associated with an increased hazard of relapse. Since molecular data were fully available only for the MSK cohort, a separate multivariable analysis was performed in this subgroup. In this analysis, the presence of high-risk mutations showed a borderline independent association with an increased risk of relapse (Supplementary Table 4).

At 3 years post-transplant, the cumulative incidence of NRM was 22% (95% CI: 16–29, Fig. 1F). NRM did not differ by disease category (data not shown). In a univariable analysis, donor type, conditioning intensity, and high baseline ferritin levels were associated with increased NRM (Supplementary Table 4). In multivariable analysis, conditioning intensity and ferritin remained significant, with donor type showing borderline significance.

Lastly, we assessed the incidence of aGVHD in our patient cohort. The cumulative incidence of grades 2–4 aGVHD at 3 months was 40% (95% CI: 33–48), while for grades 3–4 aGVHD, it was 13% (95% CI: 8.4–19) (Fig. 1G, H). In univariable analyses, conditioning intensity was associated with an increased incidence of grades II–IV aGVHD, whereas donor type and disease risk at diagnosis showed a borderline association (Supplementary Table 5).

In recent years, letermovir prophylaxis has been implemented to prevent CMV reactivation [7, 8]. Given the relevance of CMV serostatus to OS in our analysis, we conducted a subanalysis to assess the impact of letermovir use. Specifically, OS was compared between CMV-seropositive patients who received letermovir prophylaxis (n = 16) and those who did not (n = 82). No significant difference in OS was observed between the groups (p = 0.2). However, the small number of patients in the letermovir group warrants further evaluation in larger studies.

Several studies have defined ultra-high-risk AML using different criteria [9, 10]. In our study, we defined ultra-high-risk AML as patients with either a complex karyotype, monosomal karyotype, or TP53 mutation (analysis conducted only on MSK data, n = 19). The median OS was 8.2 months (range, 4.4–19) in the ultra-high-risk group compared with 14 months (range, 9.9–32) in the remaining patients (p = 0.094). There were no significant differences in the risk of relapse (p = 0.3) or NRM (p > 0.9).

Most previous studies did not find an advantage for incorporating TBI into MAC regimens for patients with AML undergoing allo-HCT [11,12,13]. However, it remains unclear whether high-risk AML patients might benefit from the inclusion of TBI [14,15,16]. We therefore compared outcomes of MAC conditioning with and without TBI in patients not in CR and observed no significant differences in OS (medians: TBI 6.9 vs 12 months, p > 0.9) or relapse (p = 0.6).

Several studies have examined the role of allo-HCT for patients not in CR [17]. Most of these studies focused on MAC and reported an OS of 20–30% at 2 years. Additional research has aimed to identify prognostic factors for these high-risk patients undergoing allo-HSCT. The Duval score, developed in 2010 from a study of patients with acute leukemia who were not in CR and received MAC, aimed to identify patients who might still benefit from allo-HCT [6]. According to the Duval score, the factors associated with a worse prognosis for AML include a duration of first CR of less than 6 months, poor cytogenetics, an unmatched donor, the presence of circulating blasts, and a Karnofsky Performance Score (KPS) < 90. Other adverse prognostic indicators reported in other studies include BM blasts ≥25%, adverse cytogenetics, and patient age above 60 [18]. Patient CMV seropositivity has also been associated with worse outcomes, including reduced survival [19], as observed in our study, and increased NRM [20].

Since the introduction of the Duval score and the publication of most prior studies, the treatment of patients undergoing allo-HCT has improved significantly [21, 22]. RIC and NMA regimens are now widely used. Previous scoring systems did not incorporate molecular data, the HCT-CI, or RIC/NMA regimens. Our study includes a contemporary cohort of patients with AML undergoing allo-HCT while not in CR. Despite this, outcomes were comparable to those reported in earlier studies. Our data reinforce the detrimental impact of circulating blasts on outcomes in this patient population; therefore, this factor should be considered in the routine assessment of patients undergoing allo-HCT who are not in CR at the time of transplant. In our cohort, RIC regimens were associated with increased NRM, likely reflecting a selection bias in patients receiving RIC. The presence of poor molecular markers was associated with higher relapse rates among AML patients. Our analysis was limited by the small cohort size and the relatively low frequency of certain mutations, which warrants further investigation. Notably, while FLT3 mutation is now an intermediate-risk marker [1], it was considered poor-risk in the 2021 NCCN guidelines, which we used for this study. These guidelines better align with our patient cohort, in whom FLT3 inhibitors were less commonly used.

Study limitations include a narrow non-CR definition and exclusion of post-transplant interventions. Our data included limited use of contemporary GVHD prophylaxis based on post-transplant cyclophosphamide, which has now been expanded to different donor types, especially mismatched unrelated donors (MMUDs) and matched sibling donors [23]. Consequently, our findings may not fully represent current practices, particularly for patients receiving MMUD grafts. In addition, we were unable to incorporate certain variables that are recognized as relevant prognostic factors, such as donor–recipient sex mismatch [24], or the EASIX score, which has been shown to predict transplant-associated mortality [25], because these data were not consistently available across cohorts.

Despite these limitations, our study provides valuable insights into this high-risk population, highlighting that a subset of patients can achieve durable remission. Moreover, circulating blasts and CMV seropositivity emerged as significant adverse prognostic factors. Further studies are warranted to define the prognostic role of CMV seropositivity in the letermovir era.