Machine Learning Predicts 30-Day Outcome among Acute Myeloid Leukemia Patients: A Single-Center, Retrospective,...

Area of Science:

• Hematology
• Computational Biology
• Medical Informatics

Background:

• Acute myeloid leukemia (AML) presents a critical medical challenge due to high 30-day mortality.
• Effective prediction of short-term survival is crucial for timely clinical intervention.

Purpose of the Study:

• To develop and validate a machine learning model for predicting 30-day survival in AML patients.
• To identify key clinical and laboratory variables influencing 30-day mortality in AML.

Main Methods:

• Utilized a cohort of 1830 AML patients (1700 survivors, 130 non-survivors at 30 days).
• Collected 50 variables (8 clinical, 42 laboratory) at diagnosis.
• Applied feature selection to identify six key predictors: induction chemotherapy (CTx), hemorrhage, infection, C-reactive protein, blood urea nitrogen, and lactate dehydrogenase.
• Trained and tested various machine learning algorithms, focusing on the decision tree (DT) model.

Main Results:

• The decision tree (DT) model achieved high accuracy (90.6%), sensitivity (70.4%), and specificity (92.1%) in predicting 30-day survival.
• DT classified patients into eight distinct groups based on prognostic factors.
• Patients receiving induction chemotherapy (Group 1) had a superior survival rate (97.8%).
• Patients with hemorrhage and low fibrinogen (Group 6) exhibited the poorest survival (45.5%) and shortest survival time (20.5 days).

Conclusions:

• Machine learning, specifically the decision tree algorithm, effectively predicts 30-day survival in AML patients.
• The identified key variables and patient classifications offer valuable insights for clinical decision-making and risk stratification.
• This approach supports personalized treatment strategies to improve outcomes for AML patients.