Age-specific impact of apoptosis gene expression and recurrent mutations on remission and survival in AML: A machine learning-based analysis

Abstract Patient's age remains one of the most important prognostic factors in acute myeloid leukemia (AML), but the extent to which age influences the clinical impacts of molecular markers is not well understood. In this study, we analyzed how a patient's age affects interaction with key gene mutations and apoptosis-related gene expression to influence remission and survival outcomes in newly diagnosed AML. We analyzed 852 AML patients from the BeatAML2 dataset and grouped them into the following age groups: 18-30 (n=82), 30-45 (n=106), 45-60 (n=218), and 60+ (n=510). We focused on five recurrent mutations (TP53, NPM1, RUNX1, ASXL1, FLT3) and a set of 12 genes (BAX, CCNG1, GADD45A, CDKN1A, MDM2, TP53, PMAIP1, MCL1, BCL2L11, BCL2L1, BCL2, CHEK2) central to p53 signaling and apoptotic regulation, selected based on prior literature. We used XGBoost models to predict complete remission (CR) and 2-year overall survival (OS), incorporating age, mutations, and genes. Baseline models using only age and mutations achieved AUCs of 0.704 for CR and 0.668 for 2-year OS. However, after we included the 12 gene expression, the model's performance improved significantly to AUCs of 0.846 for CR and 0.850 for 2-year OS. In multivariable Cox regression, TP53 mutation was associated with significantly worse 2-year OS, with a hazard ratio (HR) of 3.07 (95% CI: 2.29-4.12; p<0.005), indicating that patients with TP53 mutations had more than three times the risk of death within two years compared to those without the mutation. This effect was more pronounced in patients aged ≥60. Mutations in ASXL1 (HR 1.53, CI: 1.18-1.99; p<0.005) and FLT3 (HR 1.39, CI: 1.13-1.71; p<0.005) were also independently associated with poorer survival. Among the apoptosis-related genes, higher MCL1 expression was linked to worse survival (HR 0.80, CI 0.70-0.90; p<0.005), meaning that each unit increase in MCL1 expression was associated with a 20% reduction in the likelihood of survival. In contrast, higher BCL2 expression was modestly associated with improved outcomes, though the effect was not statistically significant. Logistic regression showed that NPM1 mutation was associated with greater odds of achieving complete remission (odds ratio (OR): 2.47, CI 1.16-5.30; p=0.02), but it did not translate into improved long-term OS. TP53 and ASXL1 mutations had the worst impact in patients aged ≥60, while CDKN1A and BAX expression predicted better outcomes in younger patients (18-59 years). SHAP analysis revealed that apoptosis-related genes contributed more strongly to prediction in younger patients, suggesting age-dependent biological patterns. These results show that combining targeted gene expression with age and mutational status can improve risk stratification in AML. This study supports building age-aware predictive models to guide clinical decision-making, particularly for older adults where treatment strategies are less standardized.