Optimal donor selection with multiple outcomes for hematopoietic stem cell transplant recipients by Bayesian nonparametric machine learning

Allogeneic hematopoietic stem cell transplantation is among the few treatment options for patients suffering from life-threatening hematologic malignancies; yet, the possible adverse complications can be serious and even fatal. Matching an unrelated donor with a recipient by human leukocyte antigens (HLA) is widely accepted. However, there is lesser agreement about other donor characteristics like age, sex/parity, additional HLA loci, etc. This leads to varying donor selection practices. Furthermore, the donor characteristic impact on mortality may differ from other critical outcomes. In this case study, we develop a general methodology for optimal donor selection with Bayesian nonparametric machine learning. Our findings suggest a simple clinical guideline: the youngest available male donor should be prioritized for all patients, except when there is a female donor considerably younger.