For that purpose, we generated a virtual donor file with d = 2,600,000 donors based on 5-locus high-resolution (HR) haplotype frequencies (HF) of the German population. HF had been estimated from a very large sample (n = 370,856). Each virtual donor was assigned randomly to one of 5 typing levels, ranging from level 1 (28% of all donors, HLA loci A, B, C, DRB1 and DQB1 typed at HR) to level 5 (17%, only HLA-A and -B typed at low resolution). The resulting file is a simplified model of DKMS Bone Marrow Donor Center (reference model). With the same HF, we also generated p = 10,000 virtual patients. Within donor searches for these patients, virtual HLA typings of incompletely typed donors could be requested based on matching probabilities. As typing result, the donor’s known HLA information was uncovered.
To assess their importance, we varied input parameters of our simulation, such as the number of allowed virtual typings and of donors in the file, the distribution of typing levels, and the diversity of the HF-distribution.
In the reference model, no HLA-matched donor(s) could be identified for 2.7% (1.8%) of the patients if 3 (10) virtual typings were allowed.
We found that these ratios are most strongly influenced by the distribution of typing levels and least by the donor file size. We also observed that when donors of level 5 are ignored during virtual typings, some patients remain without donors irrespective of the number of virtual typings allowed.
In order to minimize the number of “worst-case searches”, registries should type new donors with the highest resolution available and for all relevant HLA-Loci.