Association tests were performed for SNPs with MAF >3% using linear association models embedded in generalized estimating equations to correct for correlations among related individuals (23). We modeled the FTND score as a continuous variable that was analyzed in a standard linear regression and adjusted for age, sex, and three PCs of ancestry. Although the FTND is an ordinal variable ranging from 0 to 10, we did not use ordinal logistic regression models, which assume the same beta across each ordered transition, which was not the case for the FTND data. The distributions of FTND scores are shown in Figure 1. To investigate the possibility that true association signals may have been obscured by or confounded with comorbid dependence on other substances, we also tested models for moderately associated SNPs (p < 1 × 10−4) that contained ordinal measures for dependence on cocaine, opioids, and alcohol. Details regarding the derivation of these measures are provided elsewhere (1–3).
