Measurable residual disease (MRD) has emerged as a robust surrogate of remission depth and long-term outcome predictor in acute myeloid leukemia (AML), and is now recognized as a key biomarker for post-remission risk stratification [1, 2]. Its clinical relevance has been demonstrated across multiple techniques, including multiparameter flow cytometry and molecular approaches, and has been incorporated into international recommendations [1]. More recently, ultra-high-sensitivity next-generation sequencing (UHS-NGS) has enabled the simultaneous tracking of multiple leukemia-associated mutations at very low variant allele frequencies, further refining MRD assessment [3,4,5].

Most of the evidence supporting the prognostic value of UHS-NGS has however been generated in cohorts of younger AML patients treated with intensive chemotherapy [3, 4]. In contrast, older individuals, who account for a substantial proportion of AML cases, have frequently been excluded from these studies, or remain under-represented, largely because comorbidities preclude the use of intensive treatment approaches. AMLs arising in older patients are characterized by distinct molecular and clonal features, which may influence both MRD detectability and prognostic interpretation. Consequently, dedicated clinical data supporting the use and interpretation of UHS-NGS in older patients remain limited, and its application in this population has been guided by expert consensus rather than robust prospective evidence [1].

To explore the clinical relevance of UHS-NGS MRD in this population, we analyzed AML patients 60 years or older treated intensively and enrolled in the ALFA-1200 cohort (NCT01966497) [6, 7]. Patients received standard intensive induction chemotherapy followed by 2 intermediate-dose cytarabine (IDAC) courses in those achieving complete remission (CR) or CR with incomplete platelet recovery (CRp). Patients with intermediate or adverse-risk AML according to the ELN-2010 classification could proceed to allogeneic hematopoietic cell transplantation (allo-HCT) [8]. There was no maintenance therapy in patients not referred to allo-HCT. Among the 506 patients included, 369 achieved CR/CRp. Post-IDAC bone marrow samples collected after the second IDAC consolidation course were available for 113 of these patients, with no significant differences in baseline characteristics compared with patients without available post-IDAC samples (Supplementary Table 1). Six patients were excluded due to the absence of NGS data at diagnosis, and an additional 14 were excluded because no trackable molecular marker could be identified. UHS-NGS MRD was performed in 93 patients with available post-IDAC bone marrow samples and at least one trackable marker at diagnosis (Supplementary Fig. 1).

UHS-NGS MRD was assessed at post-IDAC using a previously described error-corrected sequencing strategy [5]. The median VAF was 0.054% (range: [0.0046–51.5%]), with 6% (6/93) of patients presenting low-level MRD below 0.01% above this threshold. At a sensitivity threshold of 10⁻⁴, MRD was detectable in 49 of 93 patients (52.7%) after exclusion of DNMT3A, TET2, ASXL1, and IDH1/2 (DTAI) mutations, in line with previously published methodological considerations [3, 5, 9,10,11,12]. The proportion of MRD-positive patients was 53% (17/32), 36% (10/28), and 70% (23/33) across ELN 2022 favorable, intermediate, and adverse risk groups, respectively (p = 0.03). Patients with MRD positivity had a significantly higher 2-year cumulative incidence of relapse (CIR 73% vs 38%, p = 0.003), inferior relapse-free survival (RFS 23% vs 55%, p = 0.008), and reduced overall survival (OS 54% vs 79%, p = 0.041) (Fig. 1A–C). These results indicate that UHS-NGS MRD retains a clinically meaningful prognostic impact in older AML patients despite their biological features.

Fig. 1: Prognostic impact of ultra-high-sensitivity next-generation sequencing–based measurable residual disease (UHS-NGS MRD) in older patients with acute myeloid leukemia.
Fig. 1: Prognostic impact of ultra-high-sensitivity next-generation sequencing–based measurable residual disease (UHS-NGS MRD) in older patients with acute myeloid leukemia.The alternative text for this image may have been generated using AI.
Full size image

A Cumulative incidence of relapse (CIR) according to UHS-NGS-MRD status using a DTAI-excluded multi-target approach. B Relapse-free survival (RFS) according to UHS-NGS-MRD status using a DTAI-excluded multi-target approach. C Overall survival (OS) according to UHS-NGS-MRD status using a DTAI-excluded multi-target approach. D CIR according to the number of persistent MRG mutations among patients with exclusively MRG markers. E RFS according to the number of persistent MRG mutations among patients with exclusively MRG markers. F OS according to the number of persistent MRG mutations among patients with exclusively MRG markers. G CIR according to an integrated molecular MRD classification combining persistence of non-MRG mutations and MRG mutation burden. H RFS according to the integrated molecular MRD classification. I OS according to the integrated molecular MRD classification.

Assessing MRD in older AML patients raises specific challenges related to disease ontogeny. While clinically defined secondary AML (sAML) accounts for approximately 20–30% of cases in this age group, nearly half of older patients with de novo AML harbor mutations in myelodysplasia-related genes (MRG), as defined by ELN 2022 [6]. We therefore explored the prognostic impact of MRG mutations (BCOR, EZH2, STAG2, RUNX1, SF3B1, U2AF1, SRSF2, ZRSR2), focusing on their number rather than their mere presence, to better capture their biological significance and improve MRD interpretation in older AML patients.

Among patients in whom only MRG mutations were detectable in remission, those harboring more than one such mutation had a markedly adverse outcome compared with patients carrying a single mutation (Fig. 1D–F), with a 2-year CIR of 100% versus 50% (p < 0.001), 2-year RFS of 0% versus 50% (p < 0.001), and 2-year OS of 43% versus 84% (p = 0.038). These findings suggest that persistence of multiple MRG mutations is more indicative of residual leukemic disease than innocuous age-related clonal hematopoiesis, thereby supporting a quantitative interpretation of these markers in older patients with AML.

Based on this observation, we implemented an integrated molecular risk stratification strategy, considering as MRD-positive patients either those with persistence of at least one non–MRG mutation (excluding DNMT3A, TET2, ASXL1, and IDH1/2 variants) or those harboring more than one MRG mutation detected at diagnosis and persisting in remission (Fig. 1G–I). Using this classification, MRD-positive patients had a markedly inferior outcome, with a 2-year CIR of 81% versus 41% (p < 0.001), a 2-year RFS of 13% versus 54% (p < 0.001), and a 2-year OS of 42% versus 81% (p < 0.001).

In the whole cohort, multivariable analyses adjusted for age, baseline white blood cell count (categorized), ELN 2022 risk classification, and allogeneic transplantation in first complete remission as a time-dependent covariate confirmed the prognostic value of MRD positivity. Cumulative incidence of relapse was modelled using Fine–Gray regression, considering death without relapse as a competing event, although few such events occurred in the analyzed cohort (n = 5), whereas RFS and OS were analyzed using extended Cox proportional hazards models. MRD positivity was associated with a higher risk of relapse (sHR 3.49, 95% CI 1.88–6.50; p < 0.001) and inferior RFS (HR 3.26, 95% CI 1.87–5.70; p < 0.001) and OS (HR 2.75, 95% CI 1.52–4.97; p < 0.001) (Table 1).

Table 1 Multivariable analyses of cumulative incidence of relapse, relapse-free survival, and overall survival according to an integrated UHS-NGS MRD classification in older patients with acute myeloid leukemia.
Full size table

This study provides the first dedicated evaluation of UHS-NGS MRD in a multicentric cohort of older AML patients treated within a homogeneous therapeutic framework. While our results support the strong prognostic value of molecular MRD in this population, they also indicate that MRD interpretation in older AML patients may benefit from adapted frameworks. In particular, given the high prevalence of MRG mutations in this age group, a quantitative rather than purely binary consideration of such markers appears clinically relevant, and the persistence of multiple markers during remission carries prognostic significance.

Interpretation of persistent MRG mutations remains challenging in older AML, as they may reflect residual leukemia or age-related clonal hematopoiesis. We did not aim to fully resolve this biological ambiguity. Instead, we used a quantitative approach, showing that multiple (≥2) MRG mutations are associated with adverse outcomes, consistent with residual disease. Further longitudinal and clonal studies are needed to better distinguish these entities.

Several limitations should be acknowledged. MRD was assessed at a single post-consolidation time point after the second IDAC course, corresponding to the end of intensive chemotherapy in this treatment framework (Supplementary Fig. 2). However, other time points, including early MRD assessment after induction, pre-transplant evaluation, and longitudinal monitoring, may provide complementary prognostic information and better capture disease kinetics. Recent recommendations emphasize the importance of dynamic MRD assessment across multiple time points rather than relying on a single evaluation [13]. The cohort size remains modest, reflecting the inherent constraints of molecularly trackable markers in older AML patients. Although UHS-NGS enables quantitative MRD assessment, our analysis primarily relied on a binary MRD classification. This approach was driven by the limited number of patients with low-level MRD (below 0.01%), precluding robust evaluation of MRD burden. Multiparameter flow cytometry MRD, which has shown complementary prognostic value in several studies [4, 14], was not available in the present analysis, and direct comparisons between UHS-NGS–based MRD and other MRD assessment techniques could not be performed. UHS-NGS-based MRD assessment was performed retrospectively and was not used to guide therapeutic decisions in this cohort. Prospective studies evaluating MRD-guided treatment strategies will be required to determine the clinical utility of UHS-NGS MRD in older AML patients. Analyses were restricted to bone marrow samples, and the potential value of alternative matrices such as peripheral blood or cell-free DNA was not evaluated. Finally, external validation in independent cohorts and across different therapeutic backbones will be necessary.

In conclusion, our findings confirm the high prognostic value of error-corrected UHS-NGS MRD in older AML patients, while highlighting the need for adapted interpretative frameworks in this setting. Accounting for both the number and nature of residual molecular markers, particularly those related to MRG mutations, may refine MRD-guided risk stratification in elderly patients and inform future therapeutic strategies.