Since there is no known functional CRHR1 polymorphism we used a haplotype-driven approach to capture potential CRHR1 variation. This approach is likely to detect the disease association of any allele, known or unknown, of moderate abundance and effect size for areas of the genome with conserved block structure, such as across the CRHR1 gene. Within each of the two haplotype blocks, we performed one analysis -- a logistical regression analysis -- and this identified the haplotypes that had an independent main effect as well as a G × E effect on outcome (see tables 4 and 6). In secondary analyses we then went further to determine whether any of the tightly linked SNPs provided the signal for the haplotype effects. Since at the outset we could not hypothesize which haplotype block might be associated with disease, for the purpose of Type I error correction a nominal p-value threshold of p = 0.025 was set (p = 0.05 divided by the number of haplotype blocks identified (i.e., 2).
