Real-world efficacy and prognostic modeling of venetoclax-enhanced intensive chemotherapy in fit AML

Abstract Objective: This study aimed to evaluate the efficacy and safety of Venetoclax combined with intensive chemotherapy in Fit AML patients, and to construct a prognostic prediction model. Methods: We retrospectively analyzed 158 newly diagnosed Fit AML patients admitted to Sichuan Provincial People's Hospital from January 2022 to May 2024. Four regimens were compared: DAV, IAV, HAV, and MAV. Primary endpoints included composite complete remission (CRc), measurable residual disease (MRD)-negativity rate, overall survival (OS), and adverse reaction rates. Seven machine learning algorithms (XGBoost, Random Forest, LightGBM, Logistic Regression, Decision Tree, MLP, SVM) were used to build prognostic models, assessing feature importance, model performance, and risk stratification. Results: The cohort had a median age of 47.5 years, with ELN-2022 risk stratification showing 11.4% favorable, 58.9% intermediate, and 29.7% adverse prognosis. Regimen distribution comprised DAV (44.9%), IAV (36.1%), HAV (10.8%), and MAV (8.2%). The overall CRc rate was 77.2% and MRD-negativity rate was 56.56%. While no significant CRc difference existed among treatment regimens (P=0.348), the adverse-risk ELN group exhibited significantly lower CRc than the intermediate-risk group (61.7% vs 83.8%, P=0.010). Survival analysis demonstrated superior OS in CRc patients versus non-responders (HR=0.42, P=0.046), and hematopoietic stem cell transplantation significantly improved survival (HR=0.31, P=0.008). For prognostic modeling, XGBoost achieved optimal performance in the entire cohort (validation AUC=0.94, sensitivity=0.75, specificity=0.91), identifying bone marrow blast percentage and CD56 expression as key negative predictors, with CD4+/CD8+ T-cell ratio and NPM1 mutation as positive predictors. In the post-remission cohort, Random Forest showed the best performance (AUC=0.92, sensitivity=0.81, specificity=0.88), where CD4+/CD8+ T-cell ratio and IDH2 mutation were key negative variables, while age and ECOG score were positive predictors. Calibration and decision curves validated clinical utility, and risk stratification effectively discriminated high-risk patients, with both models demonstrating high generalizability for personalized treatment guidance. Conclusion: Venetoclax-enhanced intensive chemotherapy achieves high response rates in Fit AML. Our machine learning models provide novel tools for prognostic prediction and personalized treatment.