Correlate the flow crossmatch results with the single antigen results using logistic regression

Aim: The Flow crossmatch (XM) test is widely utilized in HLA labs for donor selection. XM results are correlated with strength of recipient’s anti-HLA antibody and donor HLA typing. However, establishing a statistical correlation between antibody levels and Flow XM outcomes is challenging due to variable HLA locus expression and shared epitopes among different antigens. This study aims to explore the relationship between One Lambda single antigen bead (SAB) assay results and XM outcomes based on known donor HLA typing.

Methods: The XM dataset spans from 2016 to 2022 and includes all donor cell types from various sources, such as PBL, lymph nodes, and spleen. To eliminate XM results possibly affected by non-HLA factors, we excluded patients with known rituximab infusion and positive T-cell XM results that were only positive after pronase treatment (where the untreated control was negative). Donor serological typings were matched with normalized SAB MFI values. For antigens represented by multiple beads, the maximum MFI was used.

Results: We obtained 1585 T-cell XM results, of which 247 were positive, and 1782 B-cell XM results, with 322 being positive. HLA-A, B, C were selected to construct the T-cell XM model, while HLA-A, B, C, DR, and DQ were chosen for the B-cell XM model. HLA-DP was excluded due to limited data. We allocated 85% of the dataset for training and 15% for testing. Using logistic regression, we established correlations between SAB MFI values and Flow XM results, achieving 87% and 88% accuracy for T-cell and B-cell XM models, respectively. Subsequently, we utilized the model to simulate data by feeding it with different MFI values for each locus, incrementing by 500, to generate T-cell and B-cell XM simulations (Figure 1 and Figure 2). The results showed that T-cell XM results were likely to be positive when anti HLA-B MFI exceeded ~1800, anti HLA-A > ~2000, and anti HLA-C > ~10,000. For B-cell XM, >50% positivity was associated with anti HLA-B and DR > ~2000, anti HLA-A > ~4000, anti HLA-DQ and DRw > ~6000, and anti HLA-C > ~12,000 MFI.

Conclusion: Establishing clear MFI thresholds for positive XM results is difficult due to multiple DSAs and shared epitopes. However, statistical modeling can help clarify the relationship between antibody strength and XM outcomes. This approach highlights the importance of high-quality, standardized lab data in achieving correlation analysis.