OR10 Data-driven modeling of flow cytometric crossmatch: Enhanced virtual crossmatching

Abstract

Aim Increasingly HLA laboratories are using virtual crossmatching to predict recipient and donor compatibility using HLA antibody data and donor HLA type. However, virtual crossmatch interpretation is based on HLA experience and expertise of individual transplant centers. The purpose of this study is to develop data-driven algorithms (DDA) that predict flow cytometric crossmatch (FXM) outcomes using HLA antibody mean fluorescent intensity (MFI) data and donor HLA typing. These methods are independent of human intuition or experience and may provide insights that are otherwise imperceptible. Methods Two data sets consisting of 222 and 109 FXM with single antigen bead data for both HLA class I and II antibodies were used. Single antigen bead data was compiled against donor HLA antigens using the OneLambda assay and Luminex. The first DDA found the optimal MFI threshold using summation of mean fluorescent intensity data for class I and/or II that predicted either T cell or B cell mean channel shifts (MCS) above FXM cutoff. The second DDA applied a least-squares regression model to the HLA locus-specific data (second data set) to predict the actual T or B cell MCS. Results The threshold method yielded between 84.9% and 91.1% accuracy when using class I donor specific antibody (DSA) data to predict T cell outcome, and class I and II DSA data to predict B cell outcome. Optimal MFI thresholds of 3450 and 7560 were found for T and B cell prediction. For quality assurance, prediction of T cell MCS was attempted using class II data, resulting in 61.9% accuracy. The least-squares model increased accuracy to 93.6% and 97.2% for T and B cell MCS, respectively. Class I DSA influenced T and B cell MCS more than class II. The relative importance of an individual HLA locus on T cell prediction was found to be HLA-B > -A > -C. In the least-squares fitting, B64 had a large positive effect while A34 had a large negative effect on T cell MCS. Conclusions Utilizing DDA can expand accuracy of biologic systems beyond human experience and expertise. We showed how DDA can improve HLA crossmatch by generating high-level accuracy T and B cell FXM outcomes. Further improvements to the algorithm will incorporate HLA antigen expression and HLA antibody avidity.