Monte Carlo simulation of HLA-DQ heterodimers and their interactions to predict relapse and GVHD risk in cord blood transplantation

Abstract

Abstract Background The importance of HLA-DQ heterodimers on outcomes in unrelated and haploidentical stem cell transplant has been recently described. However, in the setting of cord blood transplantation, their potential interaction with broader immunogenetic variables—such as mismatch burden, B leader matching, and HLA-C expression—has not been systematically characterized. Due to limited sample sizes in trials, these relationships are often underpowered for traditional analysis. We developed a Monte Carlo simulation framework to expand the effective sample size and investigate the effects of DQ heterodimers and their interactions on relapse, GVHD, and transplant-related mortality (TRM). Methods We analyzed omidubicel data provided by Gamida Cell from its Phase III and Expanded Access Program (N = 85). Empirical distributions to simulate outcomes were generated from the dataset. Bernoulli distributions were used for simulations with binary outcomes such as relapse, TRM, acute, and chronic GVHD [Kroese et al., 2014]. Time to relapse was modeled using a log-normal distribution to reflect right-skewed clinical timing [Limpert et al., 2001]. Lastly, a Poisson-like distribution was used for mismatch burden [Cameron & Trivedi, 2013]. Then, we proceeded to generate a 10,000-sample simulation that incorporated DQ heterodimer (G1G1, G1G2, G2G2), mismatch burden, DQ mismatch, B-leader match, as well as HLA-C expression. The simulated data were then compared to the original omidubicel cohort using histograms and kernel density overlays, Kolmogorov–Smirnov (KS) tests for time to relapse and HLA mismatch burden [Massey, 1951], as well as summary comparisons of proportions and standard deviations for categorical outcomes. Multivariate logistic regression was used to model relapse, TRM, aGVHD, and cGVHD. Predictor variables used were DQ heterodimer, HLA mismatch burden, B-leader matching, and HLA-C expression. Results The Monte Carlo simulated population closely matched the real dataset across all key variables (KS p = 0.32 for relapse timing; p = 0.48 for mismatch burden). Importantly, the presence of G2G2 DQ heterodimers was associated with a significantly higher risk of relapse (OR 1.61, p < 0.001). We also found a directly proportional correlation in relapse risk with donor DQ group 2 burden (p-trend < 0.01), not present for GVHD or TRM. DQ mismatch using G1G1 donors decreased relapse risk among G2G2 recipients (interaction p < 0.05), suggesting a biologically important effect. There was no impact of HLA mismatch burden, B leader, and C expression on risk of relapse nor did they alter the effect of DQ heterodimers on outcomes. Conclusion Monte Carlo simulation offers a robust and biologically grounded approach to studying the interactions in underpowered transplant datasets. Applied to the omidubicel cord blood cohort, our model identified a relapse gradient across DQ heterodimer subtypes, with worse outcomes with higher donor group 2 burden. These findings show the utility of Monte Carlo simulation for advancing donor selection science and hypothesis generation in cord blood transplantation.