BackgroundHuman leukocyte antigen (HLA) matching is a critical factor in allogeneic unrelated hematopoietic cell transplantation (HCT) due to its impact on post-transplant survival and quality of life. Umbilical cord blood transplantation (UCBT) offers unique advantages but the optimal approach to graft selection and immunosuppression remains challenging. Unsupervised clustering, a machine learning technique has potential in analyzing transplant outcomes but its application in investigating leukemia outcomes has been limited. ObjectiveTo identify optimal combinations of HLA/KIR donor-patient pairing, conditioning, and immunosuppressive regimens in pediatric patients with acute lymphoblastic (ALL) or acute myeloblastic (AML) leukemia undergoing umbilical cord blood transplantation (UCBT). Study designOutcome data for single, unmanipulated UCBT in pediatric AML (n=708) and ALL (n=1034) patients from the Eurocord/EBMT registry were analyzed using unsupervised clustering. Resulting clusters were used to inform post-hoc competing risks and Kaplan-Meier analyses. ResultsIn AML, single HLA-C mismatches with other loci fully matched (7/8) associated with poorer relapse-free survival (RFS) (p=0.039), but a second mismatch at any other locus counteracted this effect. In ALL, total body irradiation (TBI) effectively prevented relapse mortality (p=0.007). KIR/HLA-C match status affected RFS in AML (p=0.039) but not ALL (p=0.8). Anti-thymocyte globulin (ATG) administration substantially increased relapse, with no relapses occurring in the 85 patients not receiving ATG. ConclusionsOur unsupervised clustering analyses generate several key statistical and mechanistic hypotheses regarding the relationships between HLA matching, conditioning regimens, immunosuppressive therapies, and transplantation outcomes in pediatric AML and ALL patients. HLA-C and killer immunoglobulin receptor (KIR) combinations significantly impact RFS in pediatric AML, but not ALL. ATG use in fully matched pediatric patients is associated with late-stage relapse. TBI regimens appear beneficial in ALL, with efficacy largely independent of histocompatibility variables. These findings reflect the distinct genetic and biological profiles of AML and ALL.