The non-cell-type variables such as patient age at transplantation, patient/donor sex, patient/donor cytomegalovirus serostatus, risk of disease, use of total body irradiation for conditioning, presence of major ABO mismatch, and graft-versus-host disease (GVHD) prophylaxis if significantly or suggestively associated with outcome, were used to create a “base” model for each outcome. Each cell-type variable was then added to assess the improvement to the base model.

Each cell-type variable was modeled as a continuous variable, a log-transformed variable, or according to quartiles.

P value estimated by the likelihood ratio test.