Prognostic impact of measurable residual disease in AML patients treated frontline with azacitidine and venetoclax: results from the French VENAURA registry

Abstract

Measurable residual disease (MRD) is a key prognostic marker in acute myeloid leukemia (AML), but its significance in patients treated with azacitidine and venetoclax (AZA/VEN) outside clinical trials remains unclear. We retrospectively analyzed 220 newly diagnosed AML patients from the French VENAURA registry who achieved composite complete remission and underwent MRD evaluation by multiparametric flow cytometry (MFC, LAIP/LSC) and/or NPM1 RT-qPCR. Cumulative MRD negativity was achieved in 62–67% of patients. Attaining MRD negativity at any time was strongly associated with superior overall survival (OS: 31.3 months vs 15.7 months (HR = 0.47) for LAIP, not reached vs 10.8 months (HR = 0.38) for NPM1; all p < 0.001) and lower cumulative incidence of relapse. Dual LAIP/LSC negativity (NEG/NEG) conferred the best outcomes compared to NEG/POS (HR = 0.36, p = 0.02), POS/NEG (HR = 0.24, p < 0.001) and POS/POS (HR = 0.26, p = 0.26) status. Importantly, MRD response mitigated the adverse prognostic impact of ELN 2024 intermediate/poor risk, with MRD-negative patients achieving outcomes comparable to favorable-risk cases. MRD kinetics (early vs late responders) did not affect survival, while G-CSF use improved MRD conversion and OS. In real-world AZA/VEN–treated AML, achieving deep MRD negativity, by MFC or NPM1 RT-qPCR, emerges as the dominant prognostic determinant, overriding baseline risk and supporting its integration into response-adapted strategies.

Introduction

In patients treated intensively for acute myeloid leukemia (AML), measurable residual disease (MRD) has become a critical standardized endpoint that is highly predictive of relapse-free survival (RFS) and overall survival (OS) [1–4]. However, MRD predictive value at specific time points during intensive chemotherapy (IC) courses in AML has been mostly evaluated by quantitative polymerase chain reaction (RT-qPCR) for NPM1^mut, CBF-rearranged, and PML::RARA AML [1, 3, 5]. Yet, only 30–35% of patients present molecular abnormalities that can be assessed by RT-qPCR, and even less in elderly patients. Multiparametric flow cytometry (MFC) MRD is applicable to a majority of patients after treatment-induced morphologic remission, even for those who do not harbor an appropriate molecular target for RT-qPCR. Tremendous efforts have been made to standardize MFC-MRD procedures and define specific thresholds according to DfN (Different from normal)/LAIP (leukemia associated immunophenotype) and more recently LSC (Leukemic Stem Cell) approaches [6]. In this context, LSC-MRD by MFC appeared to be highly predictive for survival and relapse risk in a large prospective cohort of patients treated with intensive chemotherapy (IC) [7]. While MFC-MRD is integrated into response assessment in pediatric AML, its use in adult AML to guide preemptive risk stratification, therapeutic interventions, and/or hematopoietic stem cell transplantation allocation in patients treated with intensive chemotherapy remains heterogeneous and not yet uniformly standardized.

For patients unfit for IC, low intensity therapy based on azacitidine are usually offered. A retrospective study reported that in patients age >60 years and treated with hypomethylating agents, the cumulative incidence of relapse was shown lower among patients who had achieved a MRD-negative (MRD^neg) response, but no relationship between MRD and overall survival (OS) or relapse free survival (RFS) was demonstrated [8]. More recently, azacitidine and venetoclax (AZA/VEN) combination regimen have emerged as a new standard of care in frontline unfit AML patients [9]. In the VIALE-A prospective trial, patients treated with AZA/VEN were evaluated for MRD by MFC at the end of cycle 1, and every three cycles thereafter. Median OS was not reached in patients reaching MRD^neg at any time with a predicted 12 months OS of 94% [10]. Similarly, Othman et al. reported on a large retrospective cohort of NPM1^mut patients treated with VEN based regimens that NPM1-MRD^neg was associate with a favorable outcome [11]. Despite these advances, MRD negativity either by MFC or RT-qPCR (i.e. NPM1^mut) determinants and their impact on survival outside clinical trial remains limited in AZA/VEN treated patients. Identifying patients with favorable outcome might be the corner stone for future de-escalation strategy.

We aimed in this retrospective study to evaluate the predictive value of MRD negativity by MFC or NPM1 RT-qPCR and its kinetics on survival in AML patients treated upfront with AZA/VEN in real-life settings.

Methods

Study design

VENAURA is an observational, multicenter registry collecting retrospective data on AZA/VEN-treated AML patients from 12 centers in the French Auvergne-Rhône-Alpes (AURA) region between January 2019 and February 2024 (IRB 00013204). All patients included in this study provided informed consent.

Inclusion criteria were patients with AML according to the WHO 2022 classification (including low-blast AML/MDS) receiving AZA/VEN as frontline therapy, with at least one MRD assessment during the first six cycles. AZA/VEN response and MRD assessments were evaluated as described below. Patients were stratified according to the ELN 2024 risk classification [12].

Composite complete remission (CRc) was defined as in the VIALE-A trial (complete remission (CR) or complete remission with incomplete hematologic recovery (CRi)). Regarding MFC-MRD, its assessment was based on the LAIP approach according to the European LeukemiaNet (ELN) recommendations (bulk lysis and at least 500,000 recorded events to achieve a sensitivity threshold of at least 10⁻³ [0.1%]). MFC-MRD was performed on bone marrow samples using an 8-color panel and considered positive when detectable above the 0.1% threshold for LAIP and 0.01% for LSC assays [6, 13].

A “backbone” of CD34/CD38/CD45/CD117 was used in both tubes, supplemented by CD7, CD56, CD13, CD33, HLA-DR, and CD19 for LAIP assessment, and by CD90 (Thy-1), Mix (CD97/CLL1/TIM3), CD45RA, and CD123 for LSC analysis [14]. LSC^low was defined as an LSC population <1% at baseline, as previously reported [7].

NPM1 RT-qPCR MRD was performed as previously described [3]. MRD negativity (MRD^neg) was defined as ≤10⁻⁴ for NPM1 by RT-qPCR on peripheral blood (or bone marrow when available). For NPM1^mut patients evaluated by both flow cytometry and RT-qPCR, those negative by flow but positive by RT-qPCR were considered MRD^pos.

As timing of MRD evaluation was highly heterogenous, we defined, only for patients that were evaluated at least twice during follow-up, MRD_first as the first time point of MRD assessment and MRD_second the second assessment. For patients evaluated only by MFC, an early MRD response was defined as negativity on the first MRD assessment (MRD^first) confirmed at the second evaluation (MRD_second). Late responders were defined by conversion from MRD^first positivity to MRD^second negativity between two time points. Poor MRD responders were defined as persistent MRD positivity at both MRD^first and MRD^second, or loss of MRD^first negativity at MRD^second .

For Sankey diagrams, patients were censored at the time of the last received cycle, allogeneic hematopoietic stem cell transplantation, or toxic death. If patients were not re-evaluated at subsequent cycles, their MRD response for the next cycle was based on the last available MRD evaluation.

Statistical analysis

For comparisons between patient characteristics, Mann–Whitney and Kruskal–Wallis tests were used for continuous quantitative variables, and the Chi-square test for categorical variables. Probabilities of OS were estimated using the Kaplan–Meier method. Stratified log-rank tests were used for pairwise intergroup comparisons of OS.

To study the cumulative incidence of relapse, a cumulative incidence model was applied. Univariate analyses were performed using the log-rank test for OS and Gray’s method for cumulative incidence outcomes. Multivariate regression was carried out using a Cox proportional hazards model, including all variables significant in univariate analyses.

To assess the overall predictive value on survival of different MRD approaches, we used Receiver Operator Characteristic (ROC) curve assay to compare MRD assays performance by calculating the area under the ROC curve (AUC).

Statistical analyses were performed using GraphPad Prism software version 8.0.1 for Windows (GraphPad Software, San Diego, California, USA, www.graphpad.com) and R software version 4.1.1 (R Core Team, 2021. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org).

Results

Patient characteristics

Of the 438 frontline patients included in the registry, 220 achieved CRc and had at least one MRD assessment by MFC or RT-qPCR (NPM1) during follow-up (Supplementary Fig. 1). Regarding timing of first marrow assessment, 90% (198/220) of patients were evaluated at the end of cycle 1, 5.9% (13/220) at the end of cycle 2 and 4.1% thereafter. Overall, median number of marrow evaluation during 6 first cycles was 2 (range 1–6). Patients’ characteristics are summarized in Table 1. Median age at diagnosis was 74.4 years (range: 31–88), and 15.4% of patients were aged over 80.

Table 1.

Patients’ characteristics.

Patients characteristics	N = 220
Median age, years (range)	74.4 (31–88)
Male, n (%)	137/220 (62.3)
Prior hematological history, n (%)
- MDS	44/220 (20)
- MPN	8/220 (3.6)
- CMML	19/220 (8.6)
Treatment for prior hematological malignancies, n (%)
- Azacitidine*	9/220 (4.1%)
- Hydroxyurea	2/220 (0.9)
- Lenalidomide	2/220 (0.9)
ECOG, n (%)
- 1	150/220 (68.1)
- ≥ 2	70/220 (31.9)
Morphological characteristics
Monocytic bias, n (%)	56/220 (25.4)
Morphological myelodysplastic related changes, n (%)	64/220 (29.1)
Median marrow blastic infiltration, % (range)	49 (10–95)
Biolgoical characteristics
Hemoglobin, g/dL (IQR)	8.8 (7.8–10.1)
Platelets, G/L (IQR)	62 (34–119.5)
WBC, G/L (IQR)	3.32 (1.81–9.86)
ANC, G/L (IQR)	1 (0.42–2.46)
Genetic characteristics
ELN 2024 risk classification, n (%)
- Favorable	86/165 (52.1)
- Intermediate	45/165 (27.3)
- Adverse	34/165 (20.6)
Cytogenetics, n (%)
- Normal	100/220 (45.4)
- CBF	10/220 (4.5)
- Isolated -5/-7	18/220 (8.1)
- Isolated del(5q)/del(7q)	26/220 (11.8)
- Complex karyotype	45/220 (20.4)
Molecular abnormalities, n (%)
- NPM1	58/217 (26.7)
- FLT3-ITD	32/217 (14.7)
- IDH1/IDH2	48/212 (22.6)
- TP53	36/170 (21.2)
Treatment characteristics
Median number of AZA/VEN cycles, n (range)	5 (1-52)
Median number of VEN days/cycle, n (range)	19 (5-28)
Median VEN dose/cycle, mg (range)	400 (100-400)
VEN dose reduction after CRc, n (%)	38/220 (17.3)
VEN duration reduction after CRc, n (%)	129/220 (58.7)
AZA dose reduction after CRc, n (%)	24/220 (10.9)
Azole concomittant use, n (%)	126/220 (57.3)
HSCT, n (%)	34/220 (15.4)

* Patients that received azacytidine in monotherapy prior AZA/VEN onset were treated for an high risk MDS

ANC Absolute Neutrophil Count, CMML chronic myelomonocytic leukemia, CRc composite complete remission, ELN European leukemia net, HSCT hematopoietic stem cell transplantation, MDS myelodysplastic syndrom, MPN myeloproliferative neoplasm, VEN venetoclax, WBC white blood cells.

A prior myeloid neoplasm was reported in 32.3% of cases, most frequently myelodysplastic syndromes (20%), followed by chronic myelomonocytic leukemia (8.6%) and myeloproliferative neoplasms (3.6%). Therapy-related AML (t-AML) was reported in 7.2% (16/220) of cases.

Morphologically, 29.1% of patients had cytological myelodysplasia-related changes, and 25.4% had a monocytic bias (defined as FAB M4/5). Extramedullary disease was reported in 4.5% (10/220) of patients (five leukemia cutis, four central nervous system infiltrations, and one isolated myeloid sarcoma).

Cytogenetic and molecular risk according to the 2024 ELN classification was available for 165/220 patients: 52.1% were favorable, 27.3% intermediate, and 20.6% adverse risk.

After a median follow-up of 12.8 months, median OS of patients reaching CRc at any time during the first six cycles and evaluated for MRD was 17.8 months. Toxic death rate was similar in patients reaching MRD^pos (12.5% 12/96) and MRD^neg (8.1% 10/124) patients (p = 0.54).

LAIP MRD response determinants and impact on survival

Bone marrow LAIP-MRD assessment was performed at least once during the first six AZA/VEN cycles in 184/220 patients (83.6%), with a median time to first LAIP-MRD assessment (LAIP-MRD_first) of one cycle (range: 1–6). At the end of cycle 1, 75.5% (139/184) were evaluated for LAIP-MRD, and 51.9% were LAIP-MRD_first negative. LAIP-MRD_first positive patients evaluated a second time during the first six cycles (n = 66) converted to MRD_second negative in 45.4% (30/66) of cases by the end of cycle 6. Conversely, 16.6% of LAIP-MRD_first negative patients lost their MRD response during the first six cycles. The cumulative LAIP-MRD ^neg rate was 61.7% (Fig. 1A).

Fig. 1. LAIP-MRD response rates and its impact on survival.

A LAIP-MRD response rate by the end of cycle 1 and at best LAIP-MRD response; B Sankey diagram of LAIP-MRD status at each cycle; C overall survival according to LAIP-MRD at the end of cycle 1 (MRD1); D overall survival according best LAIP-MRD response; E overall survival according to LAIP-MRD timing of responses defined as early (MRD_first negative and MRD_second negative), late (MRD_first positive and MRD_second negative), and poor (MRD_first negative or positive and MRD_second positive) MRD responses in patients evaluated at two different time points during 6 first cycles.

Overall, 27% of patients relapsed by the end of the first six cycles, mostly from the LAIP-MRD^pos population (Fig. 1B), while the LAIP-MRD^neg rate remained stable. A prior history of MPN or MDS (HR = 3.1, p = 0.008), NPM1^mut (HR = 0.1, p = 0.009), granulocyte colony-stimulating factor (G-CSF) use at any time (HR = 0.35, p = 0.031), and complex karyotype (HR = 2.32, p = 0.025) were significantly associated with LAIP-MRD^neg probability, while the ELN 2024 classification or single mutations (i.e., IDH^mut, TP53^mut) were not (Supplementary Table 1).

Venetoclax dose or exposure duration during cycle 1 and subsequent cycles did not influence MRD response probability (data not shown). In patients who received G-CSF after CRc, the overall LAIP-MRD^neg rate was 74.6% compared to 51.9% in those who did not (p = 0.002). After logistic regression, only NPM1^mut, prior history of MPN/MDS, and G-CSF use at any time were independently associated with LAIP-MRD^neg probability (Table 2).When evaluated at the end of cycle 1, LAIP-MRD_first negativity was associated with improved OS compared to LAIP-MRD_first positivity (31.2 vs 17.8 months, p = 0.044) (Fig. 1C). However, LAIP-MRD_first was not predictive of relapse (Supplementary Fig. 2A), and different LAIP-MRD thresholds at the end of cycle 1 were not predictive of relapse (Supplementary Fig. 3A). Median OS for patients who achieved best LAIP-MRD^neg at any time during the first six AZA/VEN cycles was 31.3 months, with an estimated 2-year OS of 58.9%. Conversely, median OS was significantly lower (15.7 months) in LAIP-MRDpos patients (HR = 0.43, p < 0.001) (Fig. 1D).

Table 2.

Multivariate analysis evaluating variables influencing LAIP-MRDneg probability.

Variables	Comparison	Odds ratio	CI95%	p value
Complex karyotype	yes vs no	1.95	0.82–4.71	0.13
Prior MDS/MPN history*	yes vs no	2.2	1.01–4.9	0.048
NPM1mut	yes vs no	0.25	0.06–0.85	0.04
GCSF use after cycle 1	yes vs no	0.46	0.21–0.99	0.049

* As there was a strong overlap between MRC status and MDS/MPN history, we selected most impactful variable for multivariate analysis.

GCSF Granulocyte-Colony Stimulating Factor, MDS myelodysplastic syndrome, MPN myeloproliferative neoplasm.

At 24 months, patients with best LAIP-MRD^pos status had a higher CIR of 70.9% compared to 29.1% in those with best LAIP-MRD^neg (Supplementary Fig. 2B). Deeper LAIP-MRD responses at best MRD assessment were predictive of relapse. The 12-month CIR was 28.3%, 38%, 67.1%, and 70% for negative, 0.1–0.9%, 1–4.9%, and ≥5% LAIP-MRD, respectively (Supplementary Fig. 3B).

Regarding timing of MRD evaluation, among patients evaluated at two time points (LAIP-MRD_first and LAIP-MRD_second) during the first six cycles, early LAIP-MRD responders had similar outcomes to late responders (p = 0.32). In contrast, poor LAIP-MRD response was associated with lower OS (p < 0.001) (Fig. 1E).

While type of best cytological response (CRi vs CR) has no impact on MRD^neg probability, patients with CR and LAIP-MRD^neg best response during 6 first cycles have a significantly longer OS compared to patients with a CRi as a best cytological response but LAIP-MRD^neg (p = 0.016). Conversely, quality of cytologic response (CR vs CRi) did not influence OS in patients with LAIP-MRD^pos best response (Supplementary Fig. 4A).

LSC MRD response determinants and impact on survival

Among the 175 patients assessed for LAIP-MRD by MFC, 61.1% (126/175) were also evaluated using the LSC approach at diagnosis. Median LSC levels at diagnosis were 4% (95% CI: 2–6%); 42% (53/126) and 58% (73/126) were LSC^low and LSC^high, respectively. Patients with LSC^low populations at diagnosis were more frequently favorable according to ELN 2024 and enriched in NPM1mut (Supplementary Table 2).

Of the 126 patients evaluated at baseline, 107 were subsequently followed for MRD. The cumulative LSC-MRD^neg rate during the first six cycles was 67.4% (Fig. 2A). Among patients evaluated at two time points during these cycles, conversion from LSC-MRD_first positive to LSC-MRD_second negative occurred in 13/27 (48.1%). By the end of cycle 6, 77.8% (21/27) of LSC-MRD^pos patients who did not convert to LSC-MRD^neg relapsed (Fig. 2B). Conversely, relapse within the first six cycles occurred in only 7.1% (4/56) of patients who achieved LSC-MRD^neg (Fig. 2B).

Fig. 2. LSC-MRD response rates and its impact on survival.

A LSC-MRD response rate by the end of cycle 1 and at best LSC-MRD response; B Sankey diagram of LSC-MRD status at each cycle; C best LSC-MRD response according ELN 2024 risk classification; D LSC and LAIP MRD kinetics during 6 fist AZA/VEN cycles; E overall survival according LSC-MRD at the end of cycle 1 (MRD_first); F overall survival according best LSC-MRD response; G overall survival according LSC-MRD timing of responses defined as early (MRD_first negative and MRD_second negative), late (MRD_first positive and MRD_second negative), and poor (MRD_first positive or negative and MRD_second positive) MRD responses in patients evaluated at two different time points during 6 first cycles.

The cumulative LSC-MRD^neg rate was significantly lower in patients with adverse ELN risk (40%) compared to favorable (73%) and intermediate (67%) groups (Fig. 2C). Monocytic bias at diagnosis (OR = 0.2, p = 0.045), complex karyotype (OR = 4.06, p = 0.018), and adverse ELN 2024 classification were significantly associated with lower LSC-MRD^neg probability (Supplementary Table 3). In the multivariate model, only prior MDS/MPN history was independently associated with LSC-MRD^neg probability (Supplementary Table 4), while G-CSF use and ELN classification were not.

We then compared LAIP and LSC kinetics across cycles. Using a two-way ANOVA mixed-effects model assuming sphericity, MRD technique had a significant effect on MRD values (F(1,334) = 6.84, p = 0.0093), indicating that the predicted mean LAIP-MRD (M = 3.46) was consistently higher than LSC-MRD (M = 0.86; difference = 2.60, 95% CI [0.64, 4.55]). However, the kinetic decay profiles of LAIP and LSC MRD were similar across groups (F(5,277) = 0.58, p = 0.71) (Fig. 2D).

When considering LSC levels at diagnosis, patients with LSC^low had superior median OS compared to LSC^high, but this did not reach statistical significance (NR vs 14.1 months, p = 0.071) (Supplementary Fig. 5). At the end of cycle 1, LSC-MRD_first negative status was associated with a favorable outcome compared to LSC-MRD_first positive (31.3 vs 8.2 months, p = 0.044) (Fig. 2E). Unlike LAIP-MRD, LSC-MRD_first was predictive of relapse (Supplementary Fig. 2C). Median OS for patients who achieved LSC-MRD^neg at any time during the first six AZA/VEN cycles was 31.3 months, with a 2-year OS of 56.4% (Fig. 2F). As for LAIP-MRD, best LSC-MRD^pos status was associated with poor outcomes, mainly due to a 24-month CIR of 83.3% compared to 45.1% in best LSC-MRD^neg (Supplementary Fig. 2D). Deeper LSC-MRD thresholds at cycle 1 or best MRD response were significantly associated with relapse, as any MRD^pos level correlated with a higher relapse risk (Supplementary Fig. 3C, D). When comparing timing of MRD evaluation, among patients assessed at two time points (LSC-MRD_first and LSC-MRD_second), late MRD responders had outcomes similar to poor MRD responders (Fig. 2G).

Finally, as LSC and LAIP-MRD by MFC are complementary approaches to evaluate residual AML cells, both were combined to assess their joint impact on survival. Patients achieving dual LAIP-MRD^neg/LSC-MRD^neg exhibited very favorable outcomes, with median OS not reached and a 4-year estimated OS of 57%. Conversely, all other MRD response patterns were associated with significantly shorter survival (Fig. 3A, B). Using ROC curve approach, dual LAIP-MRD^neg/LSC-MRD^neg (AUC = 0.83) demonstrated superior prognostic discrimination compared to LAIP-MRD^neg (AUC = 0.7) or LSC-MRD^neg (AUC = 0.78) alone.

Fig. 3. Dual evaluation of LAIP and LSC-MRD is critical for survival assessment.

A Overall survival and B cumulative incidence of relapse according to best dual LAIP/LSC MRD response at any time during 6 first AZA/VEN cycles.

As for LAIP-MRD, type of best cytological response (CRi vs CR) has no impact on MRD^neg probability. However, median OS in patients with CR and LSC-MRD^neg best response during 6 first cycles was longer (not reached) compared to patients with a CRi as a best cytological response but LAIP-MRD^neg (median OS = 12.4 months, p = 0.032). Similarly, quality of cytologic response (CR vs CRi) was significantly associated with OS in patients with LAIP-MRD^pos best response (Supplementary Fig. 4B).

NPM1 MRD response rate and impact on survival

A total of 56 patients with NPM1mut were evaluable by RT-qPCR. The NPM1 mutational burden showed a median NPM1mut /ABL1 transcript ratio at diagnosis of 524% (range: 51–3452). By the end of cycle 1, only 7% of NPM1^mut patients were NPM1-MRD^neg (Fig. 4A). Overall, cumulative NPM1-MRD^neg at any time was 62.5%, with a median time to MRD^neg of six cycles (range: 1–6). Relapse rate within the first six cycles was low (10.7%) and occurred only in patients who did not achieve MRD^neg by the end of cycle 6 (Fig. 4B). No variables significantly influenced the probability of reaching NPM1-MRD^neg. Achieving NPM1-MRD^neg at any time during the first six AZA/VEN cycles was associated with median OS not reached and an estimated 4-year OS of 69.7%, whereas median OS in NPM1-MRD^pos patients was significantly lower (10.8 months, p < 0.001) (Fig. 4C).

Fig. 4. NPM1-MRD response rates and its impact on survival.

A. NPM1-MRD response rate by the end of cycle 1 and at best MRD response in NPM1^mut patients; B Sankey diagram of NPM1 RT-qPCR MRD status at each cycle in NPM1^mut patients; C overall survival of NPM1^mut patients according best NPM1 RT-qPCR MRD response; D overall survival of NPM1^mut patients according to 3 log reduction status of NPM1/ABL1 transcript ratio compared to baseline.

While discrepancies between NPM1-MRD and LAIP-MRD were infrequent (Supplementary Table 5), the depth of molecular response provided additional prognostic information. As NPM1 MRD negativity was typically achieved later than MFC MRD, a ≥ 3-log reduction (3-logRED) in NPM1 transcript levels by the end of cycle 3 was strongly associated with a favorable prognosis compared to those not reaching this response (Fig. 4D). Using ROC curve approach, >3-logRED NPM1 (AUC = 0.85) demonstrated superior prognostic discrimination compared to MRD^neg status within the first six AZA/VEN cycles (AUC = 0.8).

Impact of MRD on survival according to ELN 2024 risk groups

To further evaluate the overall impact of MRD response during the first six AZA/VEN cycles on outcome according to the ELN 2024 classification, we analyzed patients evaluated for both LAIP and NPM1-MRD. For NPM1^mut patients, MRD status (NEG vs POS) was based on RT-qPCR rather than LAIP/LSC-MRD.

According to ELN 2024, median OS was 22.9, 13.8, and 10.9 months in favorable, intermediate, and adverse risk groups, respectively (Fig. 5A). As there was no significant difference in outcome, intermediate and adverse groups were merged (INT/ADV) for further analysis.

Fig. 5. MRD negativity abrogates negative impact of ELN 2024 classification.

A Overall survival according to ELN 2024 risk classification; B Overall survival according to ELN 2024 risk classification and best LAIP/NPM1-MRD response during 6 fist cycles.

In the ELN favorable (FAV) group, median OS was not reached in MRD^neg patients versus 16.7 months in MRD^pos (p = 0.002) (Fig. 5B). Similarly, in the ELN INT/ADV group, median OS was 31.3 versus 9.3 months in MRD^neg and MRD^pos, respectively (p = 0.008) (Fig. 5B). MRD response abrogated the negative impact of ELN classification, as no significant differences were observed between FAV and INT/ADV MRD^neg patients (HR = 0.65, p = 0.15), nor between FAV and INT/ADV MRD^pos patients (HR = 1.77, p = 0.22) (Fig. 5B). When using LSC (and NPM1 in evaluable patients), results were similar, with no significant difference between FAV and INT/ADV MRD^neg patients (HR = 0.54, p = 0.09) (Supplementary Fig. 5).

Univariate analysis of variables influencing OS are summarized in Supplementary Table 6. After logistic regression, ANC at diagnosis (continuous, HR = 1.04, p = 0.001), diploid karyotype (yes vs no, HR = 0.58, p = 0.03), G-CSF use after cycle 1 (yes vs no, HR = 0.6, p = 0.03), and best MRD response (POS vs NEG, HR = 2.44, p < 0.001) were independently associated with OS, while HSCT and ELN 2024 classification were not (Table 3).

Table 3.

Multivariate analysis for overall survival.

Variables	Comparison	Odds Ratio	CI95%	p value
ANC at diagnosis (G/L)	Continuous	1.04	1.02–1.06	0.001
Diploid Karyotype	yes vs no	0.58	0.35–0.96	0.03
ELN 2024	FAV vs ADV	0.62	0.34–1.15	0.18
	INT vs ADV	0.79	0.4–1.55	0.48
HSCT	yes vs no	0.15	0.02–1.12	0.064
GCSF use after cycle 1	yes vs no	0.58	0.36–0.94	0.03
Best cytological response	CRi vs CR	1.4	0.86–2.27	0.17
Best MRD response	POS vs NEG	2.3	1.41–3.72	<0.001

ANC absolute neutrophil count, ELN European Leukemia Net, GCSF Granulocyte-Colony Stimulating Factor, HSCT hematopoietic stem cell transplantation, MRD measurable residual disease.

Discussion

In intensively treated AML, achieving deep MRD-negative complete remission is a well-established prognostic factor associated with prolonged OS and RFS compared with morphological remission alone. In older or unfit patients, azacitidine combined with venetoclax (AZA/VEN) not only improves complete remission rates but also increases the depth of response compared with azacitidine monotherapy [9]. In VIALE-A, 32% of patients in CRc under azacitidine alone achieved MRD negativity ( < 10⁻³ by DfN/LAIP flow cytometry), but MRD status had limited evaluability due to small sample size. Similarly, a retrospective analysis of patients >60 years treated with hypomethylating agents alone found no significant correlation between MRD negativity and OS [8]. In contrast, AZA/VEN-treated patients achieving MRD-negative CRc showed longer remission durations and improved survival [10]. However, limited real-world data exist on the clinical impact of MRD and its kinetics in treatment-naive, lower-intensity, VEN-based therapy.

Our study, one of the largest real-world series to date, confirms the strong association between MRD negativity (by MFC or NPM1 RT-qPCR) and survival outcomes, aligning with VIALE-A findings but with slightly lower OS, possibly reflecting broader real-life heterogeneity [10]. These observations suggest that MRD negativity retains its predictive value beyond controlled trial settings and supports its use as a clinically meaningful endpoint in routine practice.

Different approaches can be used for MRD assessment by multiparametric flow cytometry, including the different-from-normal (DfN) and LAIP strategies, each with specific strengths and limitations. A major limitation of the DfN approach is that residual leukemic cells may be difficult to distinguish from treatment-related or regenerative myelopoiesis, which is frequent due to the continuous nature of AZA/VEN exposure or GCSF use. The LAIP approach provides higher specificity by tracking patient-specific diagnostic immunophenotypes but may miss clonal evolution or immunophenotypic switches during follow-up. Notably, phenotypic evolution from a primitive to a monocytic profile appears to occur preferentially in AZA/VEN-treated patients compared with those receiving intensive chemotherapy [15]. This phenomenon has been reported in approximately 8% of patients and is enriched in IDH2- and RUNX1-mutated AML. Although we did not formally assess phenotypic switch in our cohort, more than half of the patients were monitored using a combined LAIP and leukemic stem cell (LSC) approach. As LSC analysis targets a more stable leukemic initiating compartment, it may be less affected by treatment-induced differentiation. Consistently, dual LAIP/LSC MRD negativity showed superior prognostic discrimination compared with LAIP or LSC-MRD alone.

While the prognostic impact of LSC-MRD is well established in intensively treated AML (14–16), its relevance in lower-intensity regimens has been less explored. In the HOVON-SAKK135 trial, LSC persistence under azacitidine monotherapy correlated with adverse outcomes [16–18]. Consistent with these findings, we observed that LSC-MRD negativity after one cycle of AZA/VEN predicted superior OS (31.3 months), and failure to convert to MRD negativity within six cycles was almost invariably associated with relapse. These results align with recent data showing that LSC and LAIP integration refines prognostic stratification independently of ELN risk [7, 19]. Together, they suggest that VEN-based regimens may effectively target not only the leukemic bulk but also stem-like compartments, translating into durable remissions.

In contrast to some reports in intensively treated AML,MRD timing of response (early vs. late responders) assessed by LAIP or LSC approaches were not different in terms of OS and RFS predictive value in AZA/VEN treated patients [20, 21]. Similarly, among NPM1^mut patients, a ≥ 3-log MRD reduction within the first three cycles was more predictive for long-term survival than a binary cutoff, consistent with Othman et al. and previous data on NPM1 MRD as a dynamic marker [11, 22]. This discrepancy may be explained by fundamental differences in treatment biology between intensive chemotherapy and lower-intensity venetoclax-based regimens. AZA/VEN induces a more gradual cytoreduction, with continuous drug exposure, delayed hematologic recovery, and prolonged bone marrow regeneration, which may blunt the prognostic value of early MRD assessment. In this context, delayed MRD clearance may reflect progressive eradication of leukemic and stem-like compartments rather than primary treatment resistance. Moreover, heterogeneity in MRD assessment timing across cycles in real-life practice may further limit the ability of MRD kinetics to discriminate outcomes. Consistent with observations in intensively treated patients, ultimate MRD negativity appeared to be more clinically relevant than the timing of MRD conversion, supporting the use of MRD status as a dynamic but not strictly time-dependent surrogate endpoint in AZA/VEN-treated AML [23].

Beyond NPM1 RT‑qPCR, ultra‑high‑sensitivity (UHS) NGS assays, particularly for FLT3‑ITD, are emerging as valuable molecular MRD tools capable of detecting residual clones below conventional PCR thresholds. Early studies indicate that UHS‑NGS MRD positivity correlates with relapse risk in intensively treated AML, while not yet reported with non-intensive regiments [24]. Although not assessed in our cohort, such approaches may complement MFC and RT‑qPCR, especially in patients lacking trackable markers, and further refine risk stratification under AZA/VEN.

A key finding is that MRD negativity appeared to mitigate the adverse prognostic impact of intermediate and poor ELN 2024 risk groups. Similar to VIALE-A and Maiti et al., MRD-negative patients in these risk categories achieved survival outcomes approaching those of favorable-risk patient [9, 25]. These results highlight MRD’s potential to refine ELN-based risk stratification in lower-intensity regimens.

Post-cycle 1 G-CSF use independently increased MRD negativity probability and improved OS, corroborating post-hoc VIALE-A analyses [26]. We can hypothesize that G-CSF may promote AZA/VEN efficiency through the old concept of priming that have been previously explored in several prospective trials based on intensive chemotherapy with mitigated results [27, 28]. While G-CSF itself is unlikely to provide direct anti-leukemic benefits, it may transiently recruit quiescent leukemic stem and progenitor cells into the cell cycle, thereby increasing their susceptibility to cytotoxic or targeted agents such as azacitidine and venetoclax. Nevertheless, our findings support proactive supportive care to preserve treatment intensity and optimize outcomes.

Our study has limitations inherent to its retrospective design, including heterogeneity in MRD methodology and timing across centers. Nonetheless, the consistency of results with prospective data supports the robustness of MRD as a surrogate endpoint in real-world AZA/VEN settings.

In this large real-world cohort of AML patients treated frontline with AZA/VEN, achieving MRD negativity—by multiparametric flow cytometry or NPM1 RT-qPCR—was strongly associated with improved survival and mitigated the adverse prognostic impact of ELN 2024 risk. Dual LAIP/LSC MRD assessment and dynamic NPM1 log reduction provided additional discriminative power, identifying patients with particularly durable outcomes. These findings establish MRD as a robust and actionable surrogate endpoint in lower-intensity AML therapy, supporting its integration into personalized treatment algorithms, including response-adapted de-escalation or its potential utility as a standardized endpoint for regulatory evaluation and drug development.

Author contributions

MH treated patients, performed analysis and wrote paper, AP/DM realized flow cytometry analysis. ZG, AB, UTJ, ET, GAR, MC, AC, NBFHCS, CR, SL, GMP, MM³, JC, MM⁷, NM, SC, UT, AD and GLM treated patients. LC, ST, EB, SC, PF, TT and SH performed RT-qPCR and NGS analysis.

Data availability

Data might be accessible on request to corresponding author

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

All methods were performed in accordance with the relevant guidelines and regulations, including the principles of the Declaration of Helsinki. All patients were enrolled in the VENAURA database and provided informed consent for the collection and use of their personal data. The study protocol, including data extraction and analysis, was approved by the local Ethics Committee (Institutional Review Board, IRB 00013204).

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41375-026-03024-y.