Subsequently, comparisons of each clinical variable in each pair of clusters in each cluster solution were undertaken. For example, age of onset of use was compared in each pair of clusters in the two-cluster solution (1 vs. 2), the three-cluster solution (1 vs. 2, 1 vs. 3, and 2 vs. 3), and the four-cluster solution (1 vs. 2, 1 vs. 3, 1vs. 4, 2 vs. 3, 2 vs. 4, 3 vs. 4). The three-cluster solution generated the best separation of the data (see Results, below). Demographic variables, which had not been used to generate the clusters, were then compared in each pair of clusters in the three-cluster solution. Continuous variables, including all clinical variables, age, and years of education, were compared using analysis of variance (ANOVA). Dichotomous variables, including gender, current employment (yes vs. no), annual household income (<$20K vs. >=$20K), Native American Heritage (<50% vs. >=50%), and whether currently married (yes vs. no), were compared using chi-square and Fisher exact test. Significance was set at p<0.05.
