Machine learning–based scoring models to predict hematopoietic stem cell mobilization in allogeneic donors

Abstract

AbstractMobilized peripheral blood has become the primary source of hematopoietic stem cells for both autologous and allogeneic stem cell transplantation. Granulocyte colony-stimulating factor (G-CSF) is currently the standard agent used in the allogeneic setting. Despite the high mobilization efficacy in most donors, G-CSF requires 4–5 days of daily administration, and a small percentage of the donors fail to mobilize an optimal number of stem cells necessary for a safe allogeneic stem cell transplant. In this study, we retrospectively reviewed 1361 related allogeneic donors who underwent stem cell mobilization at Washington University. We compared the standard mobilization agent G-CSF with five alternative mobilization regimens, including GM-CSF, G-CSF+GM-CSF, GM-CSF + Plerixafor, Plerixafor and BL-8040. Cytokine-based mobilization strategies (G-CSF or in combination with GM-CSF) induce higher CD34 cell yield after 4–5 consecutive days of treatment, while CXCR4 antagonists (plerixafor and BL-8040) induce significantly less but rapid mobilization on the same day. Next, using a large dataset containing the demographic and baseline laboratory data from G-CSF–mobilized donors, we established machine learning (ML)–based scoring models that can be used to predict patients who may have less than optimal stem cell yields after a single leukapheresis session. To our knowledge, this is the first prediction model at the early donor screening stage, which may help identify allogeneic stem cell donors who may benefit from alternative approaches to enhance stem cell yields, thus ensuring safe and effective stem cell transplantation.