Machine Learning Approach to Predicting Stem Cell Donor Availability

Area of Science:

• Hematopoietic stem cell transplantation
• Medical informatics
• Machine learning in healthcare

Background:

• Unrelated donor stem cell transplant success relies on donor availability, not just genetic matching.
• Approximately 50% of potential donors in the National Marrow Donor Program are unavailable post-match, with disparities across demographic subgroups.
• Existing methods for assessing donor availability are laborious and often inaccurate at the individual level.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting individual unrelated donor availability.
• To improve the efficiency of donor selection in unrelated donor stem cell transplantation.
• To reduce the time to transplant for patients requiring hematopoietic stem cell transplants.

Main Methods:

• Utilized a machine learning approach to predict the availability of registered stem cell donors.
• Trained and evaluated the predictive model on a large dataset of 44,544 donor requests.
• Assessed model performance using the area under the receiver-operating characteristic curve (AUC).

Main Results:

• The machine learning model achieved a predictive performance of 0.77 AUC on the test cohort.
• Demonstrated the capability of machine learning to accurately estimate individual donor availability.
• Highlighted the potential for enhanced donor data to improve availability predictions.

Conclusions:

• A machine learning-based predictor can accurately estimate individual stem cell donor availability.
• Implementing this predictor during donor selection can significantly reduce the time to transplant.
• This approach offers a more precise alternative to extrapolating group averages for individual donor assessment.