Multi-omics integration of molecular genetics, cytogenetics and immunophenotyping: a novel prognostic model for immune landscape characterization and outcome prediction in Chinese patients with acute myeloid leukemia

Background Current genetic risk stratification systems for acute myeloid leukemia (AML), including the 2017 European Leukemia Network (ELN), 2022 ELN, and 2023 China (CN) stratifications, inconsistently categorize key genetic variants such as CEBPA and FLT3-ITD . These systems also demonstrate limited applicability to Chinese patients, complicating clinical decision-making. This study optimized genetic stratification for Chinese AML patients and developed a multi-omics prognostic model by integrating immunophenotypic and clinical characteristics to enhance outcome prediction. Methods We retrospectively enrolled 378 Chinese AML patients (298 for training, 80 for external validation) and compared the prognostic performance of the 2017/2022 ELN and 2023 CN systems (both integrating molecular genomic and cytogenetic data). Controversial genetic variants were re-evaluated to establish a novel 2023 CN (n-2023 CN) classification. A multi-omics model was constructed by combining n-2023 CN classification with CD15 positivity (leukemic immunophenotype), del(7q) (cytogenetic abnormality), and age ≥ 60 years (clinical factor). The model underwent external validation, was assessed for use in guiding hematopoietic stem cell transplantation (HSCT), and was compared with a published model. Results The 2023 CN demonstrated superior efficacy in predicting AML relapse and 3-year relapse-free survival compared to 2017/2022 ELN systems. The n-2023 CN reclassified CEBPA bZIP as favorable-risk, t(8;21)/inv(16) with KIT D816 / FLT3-ITD low as intermediate-risk, and FLT3-ITD high as adverse-risk among patients not receiving FLT3 inhibitors (regardless of NPM1 co-mutation). CD15 positivity, del(7q), and age ≥ 60 years were independent adverse prognostic factors for overall survival (OS). The model stratified patients into three risk groups with significantly distinct complete remission (CR) rates (93.2%, 71.2%, 35.7%) and OS ( P < 0.001), with a concordance index of 0.729 (internal) and 3-year OS AUC of 0.752 (external). High-risk patients receiving HSCT exhibited longer OS. The model outperformed a published comparator in predicting CR and 1- and 3-year OS (all P ≤ 0.001). Conclusions The n-2023 CN resolves uncertainty in genetic risk stratification for Chinese AML patients. The novel multi-omics model incorporates immunophenomics and clinical and cytogenetic factors, captures comprehensive AML characteristics, enhances prognostic precision, and provides reliable guidance for personalized treatment, including HSCT selection.