Using an expectation-maximization algorithm in SAS Genetics, we estimated frequencies for the five most common (>5% frequency) haplotypes in each population and imputed subject-specific expected haplotypes.[34], [35] Haplotype-specific odds ratios based on the additive model were calculated using unconditional logistic regression adjusting for matching factors with the most frequent haplotype set as the reference. The likelihood ratio test was used to calculate the global p-value comparing the model with haplotypes to the model without haplotypes. We present otherwise uncorrected p-values, and emphasize the 95% confidence intervals around risk estimates from two independent populations.
