To account for high rates of comorbid substance dependence in these subjects, we chose a method that could simultaneously model these disorders and their association with genotype (17). We utilized a logistic regression method in which genotype (coded 0, 1, or 2 to represent the number of coded risk alleles) is expressed as the dependent outcome variable: log(P1+P21−P1−P2)=α1+∑i=1kβiDi+demographic covariates In this model P1 and P2 represent an individual’s probability of carrying one or two copies of the risk allele, respectively, and the Di represent diagnoses for dependence on the substances evaluated in this study: nicotine, alcohol, marijuana, and cocaine. This model makes a “proportional odds” assumption, which, in this case, is equivalent to assuming an additive genetic model. The demographic covariates used were sex and age quartiles. This model was then extended to include covariates for “use” of each substance as well as dependence. Inclusion of “use” covariates creates a more complex model, but has the advantage of distinguishing between individuals who have used a substance, but not become dependent from those who cannot be dependent because they have never
