Real-world validation of ELN2024 and BEAT-AML2024 risk-stratification in patients with Acute Myeloid Leukemia unfit for intensive chemotherapy in a multicenter italian cohort.

Abstract Introduction: Acute myeloid leukemia (AML), a clonal neoplastic disease, mainly affects older adults, peaking at age 68, with two-thirds of cases diagnosed after age 55. Treatment is influenced by disease biology and patient-specific factors like comorbidities and performance status. The current risk stratification system, the European LeukemiaNet (ELN) 2022 risk model, was developed for fit patients receiving intensive therapy and is less effective for unfit patients. Therefore, two new models, ELN 2024 and Beat-AML 2024, have been proposed for patients who are candidates for non-intensive treatments. Our study aims to assess which model best predicts median Overall Survival (OS) in unfit AML patients in Italy, highlighting strengths and limitations and exploring strategies to identify patients suitable for further reduced-intensity therapies. Materials and Methods: We analyzed 171 unfit AML patients from 8 Italian centers. Molecular and cytogenetic data were used to apply ELN2022, ELN2024, and Beat-AML 2024 stratifications. Genetic alterations were assessed using myeloid NGS panels and conventional molecular analysis; cytogenetics followed ISCN guidelines. Patients received hypomethylating agents (HMA) with/without venetoclax. Risk distributions were analyzed with contingency tables; OS was assessed using Kaplan-Meier, log-rank tests, and Cox models; model performance was evaluated with the C-index. Results: The study population had a median age of 74.8 years. Among all patients, 132 received HMA-Ven, 31 received HMA alone and 8 received other low-intensity therapies. Median OS was 14 months (mos) (CI 10-17), similar to the one observed in de VIALE-A study (14,7 mos). Patients were stratified according to the ELN2022, ELN2024, and BEAT-AML2024 risk models. Based on ELN2022, the majority of patients (63%) were classified as adverse risk, while 17% and 20% were categorized as intermediate and favorable risk respectively. In contrast, using the ELN2024 model, 16% of patients fell into the adverse risk category, 22% into intermediate, and 62% into the favorable risk group. Finally according to BEAT-AML2024, 20%, 43% and 37% were observed for the same categories respectively. The curves based on the ELN2022 and the ELN2024 models demonstrated a statistically significant difference only between the adverse risk groups and the others but proved unable to discriminate between favorable and intermediate risks. The BEAT-AML2024 model, instead, showed the strongest statistical significance in differentiating between all risk groups (p-value <0.0001). The 2-years OS was 48.7% (CI 35.5-66.9) and median OS was 21 mos (CI 15-NA) for favorable risk, 25.6% (CI 15.8- 41.6) and 13 mos (CI 12-20) for intermediate risk, 10.33% (CI 2.8- 37.7) and 8 mos (CI 4-11) for adverse risk. In contrast ELN2024 showed a 2-year OS of 35.06% (CI 25.5-48.7) and median OS of 16 mos (CI 13-24) for favorable risk, 28.9%(CI 15.4-54.3) and 9 mos (CI 8-NA) for intermediate risk, 17.51% (CI 6.3-48.3) and 9 m (CI 5-16) for adverse risk. The C-index was highest for BEAT-AML24, confirming its better predictive performance (0.603, SE 0.031 vs 0.573, SE 0.028). Within the limits of our sample, we evaluated the influence of age, type of therapy, and mutation burden on survival, without obtaining statistical significant predictions. Conclusion: The BEAT-AML2024 model emerged as the best statistically significant model, demonstrating a superior C-index compared to the others. As shown by our data, BEAT-AML2024 better discriminates the 2-years OS and median OS compared to ELN2024. In our opinion, the more balanced distribution in the 3 risk classes of BEAT-AML2024 prevent the overestimation of favorable risk patients observed in ELN2024. Furthermore, patients stratified as adverse risk by BEAT-AML2024 show a lower OS compared to ELN2024-defined counterparts, suggesting a better discriminative capacity. However, the study has some limitations. Only two centers routinely analyze mutations in the DDX41, BCOR, and STAG2 genes. Moreover, these models rely on both karyotyping and NGS analysis, whose additional costs are not yet justified by the current therapeutic implications. Further research is needed to identify which patients might benefit from a further reduction in treatment intensity.