lated donor versus cord blood stem cell sources and explicitly modeling the cost structure of a cord blood bank.
Cord blood inventory size affects patient survival times through its impact on the likelihood that a transplant candidate matches a stored unit. The first step in estimating the benefit associated with various inventory levels is to determine the corresponding match probabilities by match level (six of six, five of six, or four of six human leukocyte antigen [HLA] matches [referred to as 6/6, 5/6, and 4/6 matches, respectively]). These were calculated by (1) estimating the population frequency of HLA types by racial/ethnic group (“racial groups” for short, hereafter), based on the distribution of HLA phenotypes in the National Marrow Donor Program registry, (2) calculating separately for each HLA type the likelihood of matching to an adult unrelated donor or cord unit, assuming that the HLA types in the registry and cord bank are the same as in the general population, and (3) summing over HLA types separately for each racial group (Kollman et al.  provide a detailed description of the algorithm). We calculated nationally representative estimates by taking a weighted sum of the racial group-specific probabilities based on the racial distribution of patients conducting searches of the National Marrow Donor Program registry in the second half of 2001, as reported in Kollman et al. (2004). These are shown in Table E-1. Matching is defined at the antigen level for HLA-A and B and at the allele level for HLA-DRB1.
For marrow match probabilities, the model assumes that only a fraction of matched donors will be available and willing to donate, based on historical patterns among each racial group. For cord unit match probabilities, the model uses the cell count distribution of cord units in the National Marrow Donor Program database and the empirical weight distribution of transplant candidates to estimate probabilities for a given minimum cell dose threshold. In the baseline analysis, we assume a threshold of 2.5 × 107 total nucleated cells per kilogram of body weight (TNC/Kg) and that cord units are collected from each racial group in proportion to the number of births among each group in 2002 (Martin et al., 2003). We further assume that transplants occur with one and only one cord unit. In some cases, heavier transplant candidates may receive more than one cord unit to achieve the minimum cell dose, but this is not standard practice.
Table E-1 displays estimated match probabilities for patients ≥20, averaged over weight deciles and racial groups, for cord blood inventories in the range of 50,000 to 300,000 units. Table E-2 displays probabilities for