Predicting In-Hospital Mortality After Acute Myeloid Leukemia Therapy: Through Supervised Machine Learning Algorithms

Area of Science:

• Oncology
• Biostatistics
• Health Informatics

Background:

• Induction chemotherapy for acute myeloid leukemia (AML) carries a significant risk of treatment-related mortality (5%-20%) despite careful patient selection.
• Predicting this mortality risk is crucial for optimizing patient care and resource allocation.

Purpose of the Study:

• To evaluate machine learning (ML) algorithms for predicting in-hospital mortality in acute myeloid leukemia (AML) patients.
• To identify predictive factors available at admission for AML therapy.

Main Methods:

• Utilized a State Inpatient Database (2008-2014) including 29,613 AML hospitalizations (age > 17) receiving chemotherapy.
• Compared logistic regression (LR), decision tree, and random forest ML algorithms using pre-chemotherapy covariates.
• Assessed model performance using the area under the curve (AUC) metric.

Main Results:

• Random forest and LR models achieved an AUC of 0.78, outperforming the decision tree (AUC 0.70).
• A baseline LR model using only age had an AUC of 0.62.
• At a decision threshold of 0.7, 51 potential treatment-related deaths could have been averted in the test set.

Conclusions:

• Machine learning algorithms can effectively predict inpatient mortality for AML patients using readily accessible variables.
• The developed ML model shows promise for clinical application and warrants further research.