Abstract
ABSTRACT
Neutrophils are unintentionally co‐collected during HPC(A) procedures using mononuclear cell collection protocols. Although non‐target granulocyte co‐collection is generally clinically acceptable, excessive neutrophil content may compromise certain translational research applications. This study aimed to develop a predictive model for neutrophil yield during routine HPC(A). A retrospective single‐center analysis of 377 HPC(A) procedures performed between January 2023 and April 2026 was conducted. Donor demographics, hematological parameters, procedural characteristics, and collection efficiency (CE) metrics were analyzed. Correlations between CE
1
and CE
2
were evaluated using Spearman's rank correlation coefficient. A multivariable linear regression model incorporating pre‐apheresis neutrophil count, processed blood volume (pBV), and neutrophil CE
1
was developed to predict neutrophil apheresis yield. Model performance was assessed using
R
2
, RMSE, MAE, Bland–Altman analysis, and 5‐fold cross‐validation. Strong positive correlations were observed between CE
1
and CE
2
for both CD34
+
cells (
ρ
= 0.836,
p
ρ
= 0.865,
p
R
2
= 0.198). Compared with the conventional fixed CE
1
approach, the regression‐based model showed lower RMSE, reduced systematic bias, and improved agreement between predicted and observed neutrophil yields on Bland–Altman analysis. Neutrophil yield during HPC(A) can be estimated using a regression model that integrates pre‐apheresis neutrophil count, pBV, and CE
1
. Although predictive performance remains modest, regression‐based approaches improve estimation accuracy compared with simplified fixed CE models and may support procedural planning and translational research applications involving non‐target granulocyte co‐collection in HPC(A) products.