3.1 More participants assigned to sertraline than placebo had >15% increased MA use during the last month of treatment
Of the 129 participants in the study who completed at least 8 weeks of the trial, 61 were in the sertraline group, and 68 in the placebo group. For all 129 participants, we compared their fraction of MA-positive urines during the pre-randomization baseline period to the fraction of MA-positive urines during weeks 8 through 12 of treatment to categorize those individuals who had increased their MA use during the last month of treatment. Based upon a histogram of these results, natural breakpoints at + and − 15% change in MA use with treatment were observed (data not shown).
We performed a 3 × 2 Chi-Square analyses of the number of subjects in each study condition who either: had a >15% increase in MA use with treatment (sertraline 13(21%), placebo 5(7%)), had a >15% decrease in MA use with treatment(sertraline 24(39%), placebo 24(35%)), or whose MA use changed less than 15% with treatment (sertraline 24(39%), placebo 38(57%); overall χ2(2)= 6.77, p=0.034). These results demonstrated that there was no difference between groups on the number of subjects who decreased MA use with treatment, but that the sertraline group had fewer subjects who remained unchanged during treatment, and more subjects who increased MA use during treatment than the placebo condition ().
Characterics of Subjects >15% Increased MA use with Treatment
3.2 Characteristics of participants who had >15% increase in MA use during the last month of treatment
Of the 18 participants in both groups (13 sertraline, 5 placebo) who had a >15% increase in MA use in the last month of treatment, there were no group differences in age, gender, ethnicity, length of lifetime MA use, assignment to contingency management, or baseline fraction of MA-positive urines (). Average baseline craving for MA for these subjects did not differ across treatment groups: sertraline group 5.2(3.2); placebo group 6.3(1.7); (F(3)=0.78, p=0.51).
3.3 Classification Trees
CT identified multiple factors from both the baseline/pre-randomization phase and in-treatment phase which were associated with a >15% increase in MA use during treatment (both groups; ). The baseline/pre-randomization phase data identified were the SCID Global Assessment of Functioning and ASI Drug Composite Score (). The in-treatment data identified were the average craving for MA over weeks 8–13, the week 13 Clinical Global Impression, and SCL-90 Positive Symptoms Total scale ().
3.4 Craving for MA Characterizes Subjects in the Sertraline group with increased MA use during treatment
We compared each of the CT-identified factors by group (sertraline or placebo) in all subjects who had >15% increase in MA use during treatment by either ANOVA (continuous variables) or Pearson’s Chi-Square (categorical variables; see ). Only average MA craving over weeks 8–13 was significantly higher in the sertraline group compared to placebo (F(1,125)=4.44, p=0.037; ).
3.5 Craving for MA distinguished sertraline condition participants who had >15% increased MA use during treatment
In order to better assess the evolution of craving for MA over time, we divided the dataset into four groups: 1. sertraline, <15% increase in MA use with treatment (n=48); 2. placebo, <15% increase in MA use with treatment (n=63); 3. sertraline, >15% increase in MA use with treatment (n=13); 4. placebo, >15% increase in MA use with treatment (n=5; ). We then compared these groups using a 3-week moving average MA craving and linear mixed effect modeling, using baseline MA craving as a covariate, across all 12 weeks of the study. This analysis demonstrated that those subjects in Group 3 (sertraline, >15% increase in MA use with treatment) were significantly different from all other groups in MA craving across the study versus all other groups (p<0.001; ); there was no effect of time in treatment for this group (p=0.35). A linear mixed effect model using raw (non-smoothed) MA craving data also showed that Group 3 had greater MA craving than all other groups (p<0.001), demonstrating that this result was not affected by smoothing.