Prediction of overall survival in acute myeloid leukemia using supervised machine learning.
Machine learning models using patient data predict outcomes better than current genetic guidelines for untreated acute leukemia, suggesting clinical factors matter most.
Machine learning models integrating clinical, cytogenetic, and molecular features were trained and validated on 785 AML patients from multicenter Danish real-world practice, demonstrating C-index >0.75 for non-intensively treated patients versus approximately 0.57 for ELN2024. Clinical variables contributed most to prognostic accuracy, and adding genetic features did not consistently improve performance, suggesting guideline-based genomic risk stratification is insufficient for non-intensive AML prognosis.
What the study was
- Study design
- retrospective_ml_cohort
- Category
- hematologic_malignancies
- Maturity
- Validated
- Journal
- ESMO Real World Data and Digital Oncology
Why it surfaced
Multicenter Danish ML model outperforms ELN2024 in non-intensive AML; demonstrates clinical variables matter most; practical for personalized risk stratification in treatment-ineligible or elderly patients.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.