Computer program to predict likelihood of finding and HLA-matched donor: methodology, validation, and application.

Abstract

Approximately 65% of the patients requiring bone marrow transplantation do not have an HLA-A, -B, -DR identical sibling and therefore need to find a phenotypically matched unrelated donor. As of June 30, 1996, the National Marrow Donor Program maintains a registry of 2.31 million volunteer donors, 35% of whom are fully typed for HLA-A, -B, -DR loci. Because a majority of the donors has not been DR typed, a patient who does not find a complete match at the time of the preliminary search may elect to prospectively DR type A, B matched and A, B one-antigen mismatched donors. An efficient strategy is therefore needed for determining the likelihood that an appropriate donor exists and for deciding which of the donors that have not yet been DR typed should be tested for DR matching with the candidate. We developed a mathematical algorithm and computer program to facilitate the search for a suitable donor by donor race and phenotype. The program provides information on the likelihood of 1) finding at least one HLA-A, -B, -DR phenotypically matched donor, 2) the likelihood of finding at least one DR match among m A, B matched donors who have not yet been DR typed, and 3) the likelihood of an A, B one antigen mismatched donor of a specific phenotype being a DR match with the patient. The mathematical models underlying the program are based on basic population genetics theory and utilize HLA-A, -B, DR haplotype frequencies derived from the NMDP registry. The results of the validation study show that the prediction is highly accurate at the level of broad antigens. The algorithm and program have the potential to assist patients and physicians in optimizing their decisions regarding clinical management and resource allocation on the process of searching for a suitable unrelated bone marrow donor.