Machine learning-guided risk stratification in elderly AML based on genomic, immunophenotypic and therapeutic...

Area of Science:

• Hematology
• Oncology
• Machine Learning in Medicine

Background:

• Elderly patients with acute myeloid leukemia (AML) present significant heterogeneity, making prognosis challenging.
• Current prognostic systems for AML are insufficient for elderly patients, lacking crucial immunophenotypic and therapeutic data.
• There is a need for a tailored prognostic model for elderly AML patients.

Purpose of the Study:

• To develop and internally validate a machine learning-based prognostic model for elderly AML patients.
• To improve individualized risk stratification and treatment optimization in this population.

Main Methods:

• A two-stage modeling strategy analyzed 156 elderly AML patients.
• Clinical, genomic, and immunophenotypic variables were assessed using multilayer perceptron (MLP), random forest (RF), and multivariate Cox regression.
• Internal validation involved 1000 bootstrap iterations for model performance assessment.

Main Results:

• The model achieved a strong concordance index (C-index) of 0.702.
• It accurately predicted 1-, 3-, and 5-year overall survival, validated by time-dependent AUC and calibration plots.
• Key prognostic factors identified include TP53 mutations, high CD13 expression, and IDH2 mutations.

Conclusions:

• The developed model offers a robust tool for personalized risk assessment in elderly AML.
• Integration of diverse data types enhances its utility for optimizing treatment strategies.
• External validation and dynamic biomarker integration are recommended for future research.