Predictive analytics and cord blood banking: toward utilization-based unit selection.

Background Aims: Total nucleated cell (TNC) and CD34+ cell doses are considered among the most important parameters when assessing the suitability of a human leukocyte antigen-matched cord blood unit (CBU) for allogeneic hematopoietic stem cell transplantation (HSCT). Cord blood banks therefore frequently select CBUs for cryopreservation based on pre-process TNC content. However, cell loss during processing can lead to a significant quantity of CBUs that do not meet desired post-process quality criteria, and such grafts are less likely to be selected by transplant centers for HSCT. Here the authors present a multi-parameter linear regression (MLR) model capable of identifying CBUs that would process poorly, despite meeting established pre-process TNC and CD34+ quality thresholds.

Methods: Historically processed CBUs were graded from A+ to D depending on post-process cell content, and the utilization rate of each grade category was examined. Eight pre-process predictors of post-process cell content were used to train the MLR model, including red blood cell (RBC) content; CBU volume; age of CBU when received; and TNC constituent cell subsets. The selection efficacy of this model was then compared to that of methods conventionally used to select CBUs for processing, with receiver operating characteristic (ROC) and mean inventory quality analysis forming the basis of assessment.

Results: Within the Anthony Nolan Cell Therapy Centre, CBUs graded 'D' accounted for 37% of processing expenditures despite providing only 11% of grafts shipped for HSCT. The MLR model significantly improved pre-process identification of 'D' grade CBUs relative to thresholds based primarily on CD34+ cell content (P < 0.0001) and TNC content (P < 0.0001). At a comparable financial investment, this translated to a banked graft inventory of significantly higher quality than that produced by CD34+ (+8.8% mean increase, P = 0.007) and TNC (+9.9% mean increase, P = 0.010) selection methods.

Conclusions: A predictive modelling approach to pre-process CBU selection is a simple and effective means to increase graft inventory quality and potentially future graft utilization, at no additional financial investment.