Introduction The number and location of human leukocyte antigen (HLA) mismatches associate with higher mortality after allogeneic hematopoietic cell transplantation (alloHCT). We sought to determine if prediction models using demographic and clinical data combined with genome-wide recipient-donor allele mismatching at non-HLA single nucleotide polymorphisms (SNPs) contributed independent information about patients’ cause-specific and overall mortality after alloHCT. Methods To develop and test our outcome prediction models, we use DISCOVeRY-BMT, a 2-cohort study of patients with acute leukemia (ALL and AML) and myelodysplastic syndromes (MDS) and their 8/8 HLA-matched unrelated donors reported to the Center for International Blood and Marrow Transplant Research. Donor-recipient pairs (DRP) were genotyped using the OmniExpress Chip. At each SNP, mismatch was defined as: 1) Host-versus-Graft (HvG) - where the recipient is homozygous at a SNP locus with the donor sharing one allele, 2) Graft-versus-Host (GvH) - where the donor is homozygous at a SNP locus, while the recipient shares one allele with the donor, and 3) Bidirectional - both alleles differ between DRP. Independent typed SNPs selected one representative SNP for each linkage disequilibrium region. For each DRP and for each type of mismatch, values were summed (1 = mismatch and 0 = match), multiplied by the weight based on expression quantitative trait (eQTL) loci in blood using Blood eqtlGEN (√-log10 (p) for an eQTL and 1 for a non-eQTL SNP), then divided by the number of total weighted SNPs to provide an estimated proportion of SNPs mismatched for HvG, GvH, and Bidirectional, respectively. Low, Medium, and High mismatch levels were assigned using a hierarchical method to determine the optimal cut points for the exponentiated β coefficients from the multivariate Cox regression. Risk groups were developed using DRPs in Cohort 1 and performance was validated in Cohort 2. Results The mean proportion and standard deviation of weighted mismatch SNPs for HvG, GvH and Bidirectional is 9.11 ± 0.14, 9.09 ± 0.11, and 3.0 ± 0.09, respectively, and do not differ across cohorts. A continuous mismatch score was constructed based on β coefficients trained using Cohort 1 DRPs (βHvG = -0.39, βGvH = -1.42, and βBidirectional = 0.74) and the score is normally distributed. The optimal cut-off values (x104) for the mismatch score for low/medium- and medium/high-risk groups were: 3.87 and 4.24, respectively. The mismatch risk score in testing data (Cohort 2) showed that the overall survival rate in the first year after transplant was 15.2% lower in the high mismatch risk group compared to the low mismatch risk group (log rank p = 0.0019) and did not significantly differ by disease. When further stratified by disease, the survival probability differed significantly in AML patients with high risk compared to low risk mismatch (p = 0.01). Analyses of Cohort 2 show patients in the medium mismatch risk group (HR = 1.47, 95% CI 1.08, 2.02, p = 0.02) and high mismatch risk group (HR = 1.58, 95% CI 1.17, 2.12, p = 0.003) have a significantly increased risk of overall mortality within 1-year after alloHCT compared to the low mismatch risk group when adjusted for disease, disease status, recipient age, donor age, and graft source (Figure 1). Cohort 2 analyses of cause specific death (transplant- vs. disease-related mortality, TRM vs. DRM) showed that models including genetic mismatch, disease, disease status, recipient age, donor age, and graft source, predicted patients in the medium mismatch risk group (HR = 2.19, 95% CI 1.36, 3.51, p = 0.0012) and high mismatch risk group (HR = 2.38, 95% CI 1.51, 3.74, p = 0.0002) have significantly increased risk of DRM within 1-year after alloHCT, compared to the patients in the low mismatch risk group, with no significant associations for TRM (Figure 2). Conclusions These data suggest that a reliable and reproducible genome-wide mismatch score can be used to assess risk of mortality after alloHCT. Recipients with a low mismatch genetic score showed significantly better survival irrespective of disease status; thus, this score is prognostic for patients in complete remission or with measurable disease. Understanding the contribution of non-HLA genetic mismatches between the recipient and their HLA-matched donor could provide better guidance for the donor selection process and improve the current donor-recipient matching algorithm. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal