Assuming h follows an arbitrary distribution with a mean of 0 and variance ��K, testing the null hypothesis H0: h(Z)=0 is equivalent to testing H0: ��=0, which inhibitor Imatinib Mesylate can be accomplished using a variance-component score statistic [55]: where logit . To obtain a p-value, we can compare Q to a scaled ��2 distribution with scale parameter �� and degrees of freedom ��, which are modified to account for correlation between SNPs in the same SNP-set (for further explanation, see Appendix A in Wu et al [54]). In this analysis, we opted to use a kernel that models identity-by-state (IBS), or the number of alleles shared by a pair of individuals. This kernel is the most powerful option when epistatic effects may be present. Results Descriptive analyses are shown in Table 1. The median age for included participants was 58.

0 years (range 18�C85). Approximately half of the population was male (51%) and the majority were white (82%). Most tumors were located in the stomach (66%) or small intestines (31%) and were between 5 and 10 cm in diameter. Table 1 Demographic information and tumor characteristics of patients included in genotyping ancillary study. 70% of evaluated tumors had exon 11 KIT mutations, 10% had PDGFRA mutations and 13% had no identified KIT or PDGFRA mutations. Non-white participants were younger, on average (53.0 years vs. 59.0 years), and more likely to have stomach tumors (74% vs. 64%) and exon 11 KIT mutations (84% vs. 67%). The most common exon 11 KIT mutation was a deletion at codons 557�C558 (34%).

Compared with other ACOSOG Z9001 participants, the individuals included in this genotyping GSK-3 substudy have similar demographic and tumor characteristics (Table 2). A somewhat higher proportion of participants in this ancillary study were white (82% versus 76%), but our subpopulation had nearly identical age, gender, tumor size, mitotic rate, tumor location, and tumor mutation type distributions to the full patient pool. Table 2 Comparison of patients included in the genetic ancillary study to the remainder of the Z9001 clinical trial patients. Genotype distributions of the 208 variants varied substantially by race (Table S1), but genotype frequencies among whites in our study population were very similar to the HapMap CEU sample for the 204 SNPs available in both populations. Notable discrepancies included SNPs on several aldehyde dehydrogenase genes, ALDH1A3, ALDH1A2, ALDH1L1 and ALDH1L2, and two DNA repair genes, ERCC2 and XPC. The associations between each genetic variant and possible outcome are depicted in Figure 1, with the strength of the association quantified by the inverse of the log of the p-value.