For this analysis only,

the model was reduced to a logistic regression. The outcome variable in the analysis was the presence or absence of a history of inhibitors to FVIII, and the effect of interest was the endogenous F8 haplotype. The use of genetic information allows for a more powerful method of classifying the ancestry of an learn more individual than data obtained by self-reported cultural identity or race. Principal components were constructed with EIGENSOFT [20], using an additional 13 331 SNPs spaced across the genome to describe population structure. F8 gene mutations were categorized as high risk: inversions, large deletions, nonsense, small deletions/insertions (outside A-runs), missense (Arg593Cys, Tyr2105Cys, Arg2150His, Arg2163His, Trp2229Cys, Pro2300Leu and Asn2286Lys) and splice site (at conserved nucleotides at position + or −1 and 2); or low risk: small deletions/insertions (within A-runs), splice site (at position + or −3 or more remote), missense (other regions) or other mutation types. A third category contained those for

whom no mutation was found. A multiple imputation (MI) procedure [21] was carried click here out using SAS PROC MI [22] to identify risk category for two genetically black individuals missing this variable. The MI procedure allows imputation of categorical and ordinal variables through the use of logistic regression models. The variables used to predict mutation risk were haplotype, HLA allele count, severity of haemophilia, principal components, year of birth and family relatedness. Imputations were performed by inhibitor status to allow for possible distributional differences in mutation risk for those with and

without inhibitors. For the HLA class II alleles of interest, DRB1*15 and DQB1*0602, analysis was performed based on the number of copies of the allele (i.e. 0, 1 or 2 alleles). Models DOK2 were analysed using SAS 9.2 (SAS Institute, Cary, NC, USA). Descriptors of the population are shown in Table 1. The H1, H2 and H3 haplotypes were observed in all four racial categories (Table 2), with the highest prevalence of H3 occurring in the genetically determined black population (28.6%). The H4 and H5 haplotypes were not observed in any participants. The haplotype distribution by race in the HIGS Combined Cohort was generally similar to that reported by Viel et al. [10]. A consideration when using the EM algorithm to impute missing genotypes is to establish that the missingness is at random, and not confounded with other predictors such as F8 mutation. To ensure that the imputation of missing genotypes using the EM algorithm was appropriate, the missing genotypes were compared with F8 mutation type. None of the marker positions showed an association with mutation type, indicating that the missingness was not likely due to a particular type of mutation.