Data are displayed as P value for univariate analysis (Spearman rank correlation coefficient). The correlation coefficient is shown only if P < .05. Bold entries denote the mononuclear cell (MNC) subsets that remained significantly associated with infection rates in multivariate analyses. The significance of association was tested as follows: Poisson regression models were fit using the SAS Genmod procedure using the log of the number of days at risk as a fixed predictor (offset). Days at risk were calculated as 365 or the posttransplant day of death or relapse (whichever occurred first) minus 100. Multivariate models were constructed by including all confounders significant in univariate models, and then applying a forward selection procedure to the MNC subset data. The most significant variable was added at each step until no variable entered at the .05 level of significance. Potential confounders (factors other than MNC subset counts that could influence the rate of late infections) considered in the multivariate analyses were donor relatedness/matching (HLA-matched siblings vs others), number of previous hematopoietic cell transplants (0 vs 1), disease and disease stage (diseases treated with multiple chemothrapies vs no chemotherapy, except for hydroxyurea before transplant), splenectomy (yes vs no), total body irradiation in conditioning regimen (yes vs no), administration of anticytomegalovirus T cells (yes vs no), use of corticosteroids in the first 3 months after transplant (yes vs no), occurrence of clinical extensive chronic graft-versus-host disease (GVHD) by day 365 (yes vs no), use of prophylactic antibiotics between day 100 and 365 apart from the routine trimethoprim/sulfamethoxazole to day 180 (yes vs no), use of intravenous immunoglobulin between day 180 and 365 (yes vs no), use of alpha-interferon between day 100 and 365 (yes vs no), and use of immunosuppressive drugs between day 100 and 365 apart from the routine cyclosporine taper by day 180 (yes vs no).

Definition of infection: Isolation of a pathogen from a site associated with symptoms or signs of infection, or, in the absence of a microbiologic isolate, the presence of symptoms and signs diagnostic of infection. For categorizing infections as viral, bacterial, or fungal, the infectious agent had to be microbiologically documented, except for shingles, where clinical diagnosis was considered sufficient. Presumed gastroenteritis and presumed respiratory tract infections were not counted because data on their presence or absence were not reliable. Sinusitis or pneumonia was counted only if verified by an imaging study. Cytomegalovirus (pp65) antigenemia was counted as an infection if more than 5 positive cells per slide were detected.23 A polymicrobial infection of one organ or several adjacent organs was counted as one infection. Infections with one microorganism in 2 nonadjacent organs were counted as 2 infections. Recurrent infections were counted as multiple infections only if clearly separated by a period of healed infection.

Treated in a hospital (not outpatient).

IgD⁺ B cells represent naive B cells, as most IgD⁺ B cells have been shown to lack somatic mutations.15-17 Naive CD4 T cells were counted as CD45RA^high CD4 T cells, as this subset has been shown to contain thymic emigrants.18,19 Naive CD8 T cells were defined as CD11a^low CD8 T cells, as virtually all cord blood CD8 T cells are CD11a^low and become CD11a^high after activation.20,21CD28⁺ T cells represent cells that can receive both the T-cell receptor-mediated signal and the CD28-mediated costimulatory signal.22