Refined genetic risk model with complex karyotype predicts outcomes of newly diagnosed acute myeloid leukemia treated with decitabine plus venetoclax

Abstract Introduction: The European LeukemiaNet (ELN) developed a new genetic risk score (ELN2024) for patients (pts) with newly diagnosed (ND) acute myeloid leukemia (AML) receiving less intensive therapy(ies). In this observational study, we evaluated the prognostic impact of ELN2024 when compared to ELN2022, as well as a newly developed modified ELN2024 (mELN2024) on outcomes of pts with ND AML receiving decitabine plus venetoclax (DEC/VEN). Methods: A total of 128 consecutive pts were treated between July 2019 and March 2024 at the University of Alabama at Birmingham. Pts and disease characteristics were collected through chart review. We conducted a univariate Cox regression analysis (UVA) of baseline factors and different risk models. We performed separate multivariate Cox regression analysis (MVA) for baseline characteristics and risk models to avoid multicollinearity between ELN risks and their components. Only mutations with 10 or more pts were included in the UVA. Variables with p <0.1 in the UVA were included in the MVA. To evaluate the ability of each classification score to predict a relapse free survival (RFS) and overall survival (OS), c-indices were estimated, where a c-index equal to 0.5 indicates predictive ability equal to chance and a c-index equal to 1.0 indicates perfect predictive ability. No adjustment for multiple testing was made in these exploratory analyses. The study was approved by the institutional review board. Results: Median age was 69 (range, 39-75) years. Seventy (58%) patients were male, and 90 (70%) were of White race. Secondary and therapy-related AML were diagnosed in 29 (23%) and 24 (19%) pts. Complex karyotype (CK) was present in 50 (39%) pts. TP53 was the most commonly detected mutation (n=47, 37%), followed by NRAS (n=22, 17%), NPM1 (N=18, 14%), and IDH2 (n=17, 13%). Pts were classified as favorable (n=17, 13%), intermediate (n=11, 8%), or adverse (n=100, 78%) risk per ELN2022. When reclassified according to ELN2024, 56 (44%) pts were downstaged and 7 (5%) were upstaged. The median follow-up was 29.5 (range, 1-60) months. The median OS was 10.26 (95%CI 1.497-19.023) months for favorable, 12 (95%CI 5.807-18.193) months for intermediate, and 5.25 (95%CI 3.017-7.483) months for adverse risk per ELN2022 (p=0.276). When using ELN2024 the respective median survival were 9.21 (95%CI 4.916-13.50) months for favorable, 6.46 (95%CI=4.094-8.826) months for intermediate, and 4.36 (95%CI 0.973-7.747) months for adverse risk (p=0.09). In the MVA of baseline characteristics, IDH2 was significantly associated with improved RFS (HR=0.423 [95%CI=0.219-0.820), p=0.011) and OS (HR=0.429, [95%CI 0.217-0.850], p=0.015), while CK (HR=2.014 [95%CI 1.094-3.706], p=0.025) was associated with worse OS. MVA for risk classifications showed that neither ELN2022 nor ELN2024 significantly predicted RFS or OS. Given the predictive value of CK on OS in MVA, we developed a modified ELN2024 (mELN2024) risk model adding CK to TP53 in the adverse risk group. In the mELN2024 model, 33 (26%) patients had favorable, 31 (24%) intermediate, and 64 (50%) adverse risk disease. In the MVA, mELN2024 adverse risk was associated with worse RFS (HR=2.213 [95%CI 1.115-4.394], p=0.023) and OS (HR=2.248 [95%CI 1.126-4.486], p=0.022) versus favorable risk. Two-year RFS and OS for favorable/intermediate/adverse groups were as follows: ELN2022, 25%/45%/13% and 25%/45%/18%, ELN2024, 20%/18%/7% and 24%/18%/14%, and mELN2024 25%/20%/4% and 31%/20%/10%, respectively. We calculated the c-index for all risk classifications to evaluate the fitness of these models to predict RFS and OS. mELN2024 was associated with the best predictive ability for both RFS and OS: for RFS, ELN2022 c-index=0.65 (se=0.055, p=0.005), ELN2024 0.61 (se=0.041, p=0.004), and mELN2024 0.67 (se=0.039, p <0.001); for OS, ELN2022 0.64 (se=0.056, p=0.009), ELN2024 0.60 (se=0.042, p=0.01), mELN2024 0.66 (se=0.041, p <0.001). Conclusion: Our findings indicate that genetic mutations alone may not adequately predict outcomes following DEC/VEN therapy for ND AML treated in the real-world setting. Importantly, cytogenetics (CK) maintains its critical role as a prognostic factor for survival in these patients. This newly developed mELN2024 risk score exhibits superior predictive power for survival outcomes compared to existing ELN2022 and ELN2024 models. Further larger-scale investigations are necessary to validate the predictive capacity of this novel mELN2024 risk score.