Retrospective study on the clinical outcomes and characteristics of acute myeloid leukemia: different outcomes in the same risk group

Introduction

Acute myeloid leukemia (AML) is a molecularly heterogeneous clonal disorder characterized by dysregulated myeloid differentiation and dismal survival rates in high-risk subgroups (Kantarjian et al., 2021). Contemporary risk stratification, guided by the National Comprehensive Cancer Network (NCCN) and European Leukemia Network (ELN) criteria, integrates cytogenetic and molecular abnormalities to inform therapeutic decisions (Lachowiez et al., 2023; Pollyea et al., 2023). The prognostic significance of these classifications underscores their critical role in optimizing AML management.

The 2022 ELN guidelines represent a paradigm shift in AML risk classification, revising key criteria from the 2017 version. Notably, FLT3-ITD mutations—previously stratified based on allelic ratio—are now uniformly classified as intermediate-risk when occurring without adverse genetic features. This reclassification primarily addresses methodological challenges in standardizing allele ratio assessment, the emerging role of measurable residual disease (MRD) dynamics in NPM1-mutated AML, and the therapeutic impact of targeted kinase inhibitors (Falini & Dillon, 2024). In addition, FLT3-ITD MRD analysis, which is based on next-generation sequencing (NGS), has improved the ability to predict survival outcomes (Ediriwickrema, Gentles & Majeti, 2023; Peroni et al., 2023). Mutations in AML with mutations in nine myelodysplasia-related (AML-MR) genes (ASXL1, BCOR, EZH2, RUNX1, SF3B1, SRSF2, STAG2, U2AF1, or ZRSR2) are designated as high-risk, independent of other cytogenetic abnormalities. CEBPA biallelic or monoallelic mutations within the bZIP domain are classified as favorable, irrespective of their allelic status. Additionally, hyperdiploid karyotypes with multiple trisomies are no longer considered complex or high-risk, reflecting advances in understanding their prognostic implications (Bouligny, Maher & Grant, 2023; Lachowiez et al., 2023).

Despite these updates, the 2022 ELN framework faces unresolved limitations (Lo et al., 2023). The combination of MRD and risk stratification can guide the treatment of AML (Cloos, Ngai & Heuser, 2023). In addition, studies have shown that age should also be considered in the risk stratification of AML patients (Pogosova-Agadjanyan et al., 2020). Clinical and molecular heterogeneity persist within risk groups, particularly among patients with “normal” karyotypes or co-occurring mutations (Dohner, Weisdorf & Bloomfield, 2015; Mrozek et al., 2012). For instance, patients classified as favorable or intermediate-risk by ELN 2022 exhibit marked survival variability, underscoring the inadequacy of current stratification criteria. Furthermore, NPM1 and CEBPA mutated AML, while generally sensitive to induction chemotherapy, demonstrate inconsistent long-term outcomes due to high relapse rates (Shi et al., 2024; Su et al., 2022; Suzuki et al., 2005; Wang et al., 2020). The integration of novel therapies—such as venetoclax (VEN)-based regimens and hematopoietic stem cell transplantation (HSCT)—has improved survival even in historically high-risk cohorts, yet these advances challenge traditional risk models (Socie, 2022). AML-RC are classified into an adverse risk group, but the prognosis for these patients varies.

However, persistent outcome disparities within risk strata highlight unmet needs in prognostic precision. This study conducts a comprehensive analysis of clinical characteristics and survival outcomes in newly diagnosed AML patients, addressing unresolved controversies in current risk stratification.

Materials & Methods

Results

Genomic profiling results

NGS was performed on 154 patients, covering 56 genes. NGS detected mutations in 94.2% of patients. Complex variations involving more than three gene mutations were observed in 56.5% (87/154). The genetic landscapes of these patients are shown in Fig. S1. Of the 56 genes analyzed, six were not detected: CDC25C, ETNK1, GNB1, PIGA, PPM1D, and UBA1. The most frequently mutated genes were FLT3 (26.62%), DNMT3A (21.43%), NPM1 (18.18%), CEBPA (17.53%), TET2 (15.58%), ASXL1 (14.29%), TP53 (13.64%), NRAS (13.64%), and WT1 (12.34%), as illustrated in Fig. 1A. The comutation and exlusivity patterns were examined (Fig. 1B). FLT3 co-occurred with NPM1 and DNMT3A but was mutually exclusive with TP53 and CEBPA. TP53 mutations were associated with DDX41 but excluded NRAS, WT1, FLT3, and NPM1. CEBPA was strongly associated with WT1, but mutually exclusive to both TP53 and FLT3 mutations.

Survival

Survival data were available for 146 patients. Median follow-up was 16.8 months (range: 0.5–57.4). At the last follow-up, a total of 79 (54.1%) patients had died. Median OS and PFS were 22.9 months and 14.07 months, respectively. Three-year OS and PFS rates were 41% and 36% (Figs. 2A–2B).

Survival with HSCT

Of the 146 patients under follow-up, 67 underwent HSCT, which significantly improved survival outcomes. HSCT significantly improved OS (not reached vs. 11.4 months, P < 0.0001; Fig. 3A) and PFS (not reached vs. 9.3 months, P < 0.0001; Fig. 3B). Our transplant group consisted of 54 allo-HSCT and 13 ASCT. To determine whether the type of transplantation influenced survival, we conducted a further analysis. No significant difference was observed between allo-HSCT and ASCT (Figs. 3C–3D).

Clinical factors and prognosis

We evaluated whether other risk factors, including age, sex, induction chemotherapy regimen, chromosome karyotype, proportion of BM blasts % (BM%), and proportion of peripheral blood (PB) blasts % (PB%) at initial diagnosis, might affect survival outcomes (Fig. 4). Female sex (OS, 32.4 months vs 15.1 months, P = 0.02; PFS, 17.7 months vs 10.9 months, P = 0.049; Figs. 4A–4B), age < 60 years (OS, 27.9 months vs 10.8 months, P = 0.007; PFS, 17.7 months vs 9.5 months, P = 0.009; Figs. 4C–4D), and a normal karyotype (OS, 38.5 months vs 16.9 months, P = 0.02; PFS, 19.1 months vs 11.7 months, P = 0.15; Figs. 4G–4H) were associated with longer OS and PFS. However, chromosome karyotype seemed to have little effect on PFS (P > 0.05). The chemotherapy regimen was also an important factor, and the IA ± E had longest OS and PFS, the AZA+VEN had the shortest survival (Figs. 4E–4F). BM% ≥ 50% and PB% ≥ 50% were associated with longer OS and PFS (Figs. 5A–5D).

Discussion

AML prognosis is influenced by genetic and clinical factors, necessitating individualized risk stratification (Venugopal & Sekeres, 2024). To our knowledge, our retrospective study is leveraging real-world data to evaluate the prognostic utility of NGS-based risk classification under the 2022 ELN framework. We investigated the prognostic implications of genetic mutations, clinical features, and treatment modalities under the 2022 ELN framework. Notably, while the 2022 ELN guidelines enhance the accuracy of intermediate/adverse risk stratification, they exhibit limitations in capturing intragroup variability, particularly among patients harboring FLT3, CEBPA, or TET2 mutations. Mutations in TP53 and KIT emerged as strong predictors of dismal outcomes, underscoring their necessity for inclusion in future prognostic models. The 2022 ELN framework demonstrated superior discriminatory power for OS and PFS compared to 2017 ELN, particularly in refining adverse-risk classification (Dohner et al., 2022).

HSCT confers a survival advantage in AML patients, with no significant differences observed between allo-HSCT and ASCT (Chen et al., 2023). At present, there are still controversies regarding the selection criteria and therapeutic efficacy of allo-HSCT and ASCT in AML. Patients with intermediate to high risk may be more likely to benefit from allo-HSCT. However, some studies have found that for low-risk patients such as Core-binding factor acute myeloid leukemia (CBF-AML), the efficacy of ASCT is superior to that of allo-HSCT (Al Hamed et al., 2024). Due to the limitations of our sample size, this conclusion still requires further verification. Our study identified female sex, age < 60 years, and normal karyotype as independent predictors of prolonged OS and PFS. Notably, conflicting results emerged between univariate and multivariate analyses for age, suggesting that its prognostic impact may be modulated by co-factors such as mutation profiles or treatment intensity. Although age remained a significant predictor in the univariate analysis, its marginal effect in multivariate modeling likely reflects the limited representation of elderly patients (15.6%) in our cohort. In addition, our study revealed that BM% ≥50% and PB% ≥50% are not factors for poor prognosis; in contrast, they are factors for good prognosis, which is not quite what we thought. This result needs to be further verified.

According to the Chinese guidelines for diagnosis and treatment of adult AML, for all AML patients who are eligible to participate in clinical studies, it is recommended that they primarily participate in clinical studies. For patients who are suitable for intensive treatment, the IA or DA regimens are the mainstream options. For those who are not suitable for intensive treatment or are elderly and frail, the AZA + VEN regimen is the mainstream option (Leukemia Lymphoma Group CSoHCMA, 2023). There are no uniform OS or PFS values applicable to all Chinese AML patients. This is usually significantly influenced by factors such as age, genetic risk, physical condition and treatment tolerance. One retrospective, real-world study demonstrated that patients with AML treated with VEN+AZA at a single centre had a median OS of 11.93 months (Bai et al., 2025). Our study revealed that chemotherapy prolonged survival, with the IA ± E regimen resulting in the longest OS and PFS and AZA+VEN resulting in the shortest OS and PFS, which is consistent with the findings of one study (Bell et al., 2019). The reason for this result may be that AZA+VEN is mainly used for elderly patients, those with multiple comorbidities or poor physical condition, or those who cannot tolerate intensive chemotherapy. These patients have a poorer prognosis due to their underlying diseases and physical conditions, which limit the intensity and efficacy of the treatment. The AZA+VEN regimen is a low-intensity chemotherapy approach, which is unable to completely eliminate leukemia cells, has a high recurrence rate, and a low long-term survival rate. In addition, HSCT is also an important factor in prolonging the survival prognosis of patients.The 2022 ELN and 2017 ELN are not suitable for elderly patients receiving nonintensive therapy, especially AZA+VEN (Jahn et al., 2023). These patients can be referred to 2022 ELN patients receiving low-intensity therapy (Dohner et al., 2024a). Our results are consistent with reports that the IA regimen prolongs OS and PFS in AML patients more than the DA regimen does (Ding et al., 2024). These results emphasize the need for risk-adapted treatment strategies, particularly in non-intensive settings where 2022 ELN guidelines may not apply.

HSCT significantly improved survival, though its benefit was limited in TP53-mutated patients (He et al., 2024; Kim et al., 2024). Notably, HSCT failed to confer survival benefits in this subgroup, with all five patients undergoing transplantation dying post-transplant (Zhao et al., 2023). OS and PFS were the shortest for TP53, followed by KIT. The occurrence rate of the WT1 gene in our study was 12.34%. TP53 is mutually exclusive with WT1, while CEBPA is prone to coexist with WT1. The median OS of WT1 was not reached, and the median PFS was 10.6 months. In addition, our study found that the CEBPA group indicated a good prognosis, but if combined with WT1 gene mutation, the prognosis might be poor (Wang et al., 2022). KIT mutations, not included in ELN criteria, demonstrated prognostic significance comparable to TP53, necessitating their reevaluation in risk models. Interestingly, we found a strange phenomenon, 2022 ELN risk stratification is the same, but the actual prognosis is completely different. Statistical analysis of their baseline characteristics revealed no significant differences in Group A, except for LDH level and HSCT. In Group B, the only significant difference was PB%. Co-mutation analysis revealed distinct patterns—Group A (OS < 18 months) was enriched for NPM1, FLT3, CEBPA and TET2 co-mutations, whereas Group B (OS > 18 months) exhibited frequent TET2, ASXL1, WT1 and CEBPA associations. These findings highlight the inadequacy of current risk models in accounting for mutation interactions and suggest that 2022 ELN stratification may misclassify patients with specific co-mutation profiles. As proposed by Dohner et al. (2024b), a four-gene model (TP53, FLT3-ITD, NRAS, KRAS) demonstrates superior predictive accuracy for elderly AML patients receiving non-intensive therapy. Our study emphasizes the need for integrated models that incorporate both ELN criteria and mutation-specific signatures. These results advocate for a refined risk model integrating ELN criteria with mutation-specific signatures, particularly for elderly or non-intensive therapy patients. Future studies should validate the prognostic role of KIT and explore gene-environment interactions to optimize personalized care. The retrospective design and single-center cohort limit generalizability. Larger prospective studies are needed to validate prognostic role and optimize co-mutation scoring systems.

Conclusions

In conclusion, AML prognosis is influenced by genetic and clinical factors, necessitating individualized risk stratification. Our study confirms that the 2022 ELN guidelines improve intermediate/adverse risk classification but underestimate variability within groups, particularly in patients with FLT3, CEBPA, or TET2 mutations. TP53 and KIT mutations were strongly associated with poor outcomes, warranting inclusion in prognostic models. HSCT significantly improved survival, though its benefit was limited in TP53-mutated patients. Our findings underscore the need for molecular data integration to refine AML risk stratification.

Supplemental Information