Machine learning-based prediction of G-CSF-induced hematopoietic stem cell mobilization outcomes in healthy volunteers.

Abstract

Haematopoietic stem cell transplantation (HSC-T) has been established as a fundamental therapeutic intervention for a wide range of hematological malignancies and disorders, with a proven record of efficacy spanning over five decades. Peripheral blood (PB) has become the haematopoietic stem cell (HSC) source of choice, surpassing bone marrow, owing to its cost-effectiveness, reduced invasiveness, higher cell yields, and shorter hospitalizations. Clinically, granulocyte-colony stimulating factor (G-CSF) administration is the standard procedure for inducing HSC mobilization. Nevertheless, a significant proportion of potential donors, ranging from 5% to 10%, exhibit suboptimal mobilization responses to G-CSF. To investigate this, we carried out a retrospective analysis of mobilization data from 1056 donors who underwent G-CSF-induced HSC mobilization over the 5-year period from 2018 to 2023. This comprehensive study elucidated the complex interplay between mobilization efficacy, as measured primarily by CD34+ cell yield, and a variety of influencing factors. Our data indicate that better mobilization outcomes are achieved in male donors than female donors. Additionally, we found a positive correlation between increased body weight and improved mobilization efficiency, implying a more favourable response to G-CSF in obese donors. Moreover, the implementation of a split-dose regimen for the mobilization agent significantly improved outcomes, highlighting the critical role of dosing strategies. Notably, younger donors exhibited better mobilization responses, underscoring age as a pivotal determinant of mobilization outcomes. Leveraging machine learning (ML) algorithms, we developed seven predictive models designed to forecast G-CSF-induced HSC mobilization outcomes on the basis of these variables.