3.1. Participant characteristics
Across the five sites, a total of 640 individuals were screened; of these, 423 were determined to be eligible for the protocol and provided informed consent. The primary reasons for ineligibility were no substance use in the last 28 days (n = 95, 51.9%), seeking detoxification, inpatient treatment or methadone maintenance (n = 34, 18.6%), lack of sufficient housing to participate in outpatient treatment (n = 15, 8.2%), moving or going to jail within 60 days (n = 12, 6.6%), insufficient psychiatric stability for outpatient treatment (n = 11, 6%), not willing to be randomized for treatment or be reached for follow-up (n = 5, 2.7%), not interested in participating (n = 5, 2.7%), less than 18 years of age (n = 3, 1.6%), did not speak English (n = 2, 1.1%), or previously participated in the study (n = 1, 0.5%). Thirty-four individuals were screened but dropped out during the evaluation process.
A total of 423 participants were randomized to treatment condition (198 to MI, 202 to standard intake/evaluation for the four sites who reached 100). Baseline characteristics by site are presented in . Although randomization was successful in that there were few significant differences between conditions within sites, there were several statistically significant differences in participant characteristics across sites, including gender (the proportion of female participants ranged from 10% to 67% across sites), education (mean years of education ranged from 11.8 to 12.8), legal system involvement in treatment seeking (the proportion of participants with legal problems that prompted or mandated treatment seeking ranged from .31 to .91 across the sites), and primary reported substance use problem. Regarding the latter, although alcohol was the most frequent primary substance abuse problem reported across the sites (ranging from 30% to 60% of participants), for each site the second more prevalent type of drug use varied widely; these included marijuana, cocaine, and methamphetamines. Across sites, 38% of the participants had had previous alcohol and 47% had had previous drug abuse treatment.
Baseline demographic characteristics and substance use variables by site
Overall, of the 423 randomized participants, 377 (89%) completed their protocol session, 323 (76%) completed the 1-month (28 day) follow-up and provided a urine or breath specimen (81% of those who completed their protocol session), and 307 (73%) completed the 3-month (84 day) follow-up (77% of those who completed their protocol session). Three hundred and forty-seven participants (82%) were interviewed at least once. Rates of follow-up did not differ by condition within sites, but did vary across sites (completion rates for the 84-day follow-up across the four sites that randomized 100 participants were 65%, 81%, 81%, and 69%).
3.2. Treatment implementation, fidelity, and skill
Session audiotapes were available from 315 of the 377 sessions delivered (59 sessions were either not taped, inaudible, or taped incorrectly). All 315 audiotapes were rated by the independent evaluators to evaluate: (1) the degree to which MI was implemented as intended and could be discriminated from the standard intervention and (2) the level of variation in intervention delivery across sites and therapies, for both the MI and standard intervention conditions. As shown in , there were consistent, sharp differences across the two conditions, in the expected directions, in ratings of the frequency with which interventions and strategies associated with MI were present in the sessions (MI mean = 3.8, standard mean = 2.2), with statistically significant differences in all sites (NB: site identities are masked). For those sessions in which at least one MI strategy or technique was rated as present (100% of all MI sessions, 44% of all standard sessions), clinicians delivering MI were rated as significantly more skillful in delivering MI interventions (MI mean = 4.6, standard mean = 3.4), with statistically significant site effects as well.
Treatment adherence and skill levels by condition and site
As expected, the items tapping interventions associated with general counseling activities were not significantly different by condition (MI mean = 4.2, standard mean = 4.5). Again, clinicians delivering MI were rated as delivering these significantly more skillfully (MI mean = 4.6, standard mean = 4.3). As shown in Table 2, interventions which were antithetical to MI were rarely seen in either condition, as very low mean scores were seen on this scale. MI therapists were, however, rated as using these interventions significantly less frequently than standard treatment therapists (MI mean = 1.4, standard treatment mean = 1.5), but significant differences in the skill level with which these were implemented did not differ by condition.
Although there were statistically significant differences in MI frequency and skill ratings across conditions, there were also significant site effects for most of these dimensions that were likely to reflect variability in the nature of the interventions typically delivered at these sites. To put these differences into context, a multivariate ANOVA analysis (Harris, 1985
) of the adherence/frequency ratings from all three scales (MI, non-MI, and general) simultaneously suggested significant effects for condition (F
(3,305) = 112.30, p
= .00) and site within condition (F
(18,921) = 6.11, p
= .01). However, the theta values, which provide an estimate in the amount of variance accounted for by each of these effects, suggested condition (θ
= .52) accounted for substantially more variance in adherence scores than did site within condition (θ
= .23). Similarly, although there were significant effects of both condition and site within condition for the skill scores, the theta values suggested most of the variance in skill scores was associated with condition (θ
= .30), rather than condition within group (θ
= .10). A similar analysis evaluated the magnitude of therapist effects, and suggested that 47% of variance in the tape rating adherence scores were associated with intervention condition, and only 8% attributable to therapists overall.
3.3. One-month outcomes: retention and substance abuse
Primary outcome variables (retention in treatment and frequency of substance use), by treatment condition and site, are presented in . As noted above, two approaches were used to evaluate effects of the study treatment on the continuous measure of retention (number of treatment sessions completed). The mixed effect ANOVA model, with the effect of site nested within treatment conditions, evaluated condition effects in the context of variability across the participating sites. This model indicated that across the five sites, participants assigned to MI completed significantly more sessions in the 28 days after randomization than those assigned to standard treatment (mean 5.0 versus 4.0, F(1,334) = 3.8, p = .05). The effect size, expressed as Cohens d, was .24. When each of the sites was evaluated separately, retention was higher in MI than the standard intervention in three of the four sites. Using the dichotomous measure, participants assigned to MI were significantly more likely to be enrolled in treatment at the clinic 28 days after randomization than those assigned to the standard evaluation (84% for MI versus 75% for standard, X2(1) = 3.5, p = .05). In the cases where there was some delay in providing the protocol session, results were similar.
Retention and substance use by site and condition
The primary outcome measure for evaluating the effects of the study conditions on substance use was the total number of days on which the participant reported using his or her identified primary substance of use in the 28 days following randomization. Both the ANOVA model and the mixed effects model indicated no significant effect of condition on days of substance use for the sample as a whole (ANOVA: F(1,334) = .10, p = .75; mixed effects: F(1,328) = 0.15, p = .70). When sites were evaluated separately, MI was associated with fewer days of substance use in three of the four sites, but these effects were not statistically significant.
3.4. Three-month outcomes: retention and substance abuse
At the 84-day follow-up, retention in treatment remained high overall. Participants assigned to the standard evaluation had completed a mean of 13.2 (S.D. = 13.0) sessions with a mean of 56.5 days of treatment (S.D. = 31.2) and those assigned to MI had completed a mean of 15.2 sessions (S.D. = 14.6) sessions and a mean of 60.7 days of treatment (S.D. = 32.7). However, these differences were not statistically significant, using either model. Overall, 96 (61.5%) of those assigned to MI and 91 (56%) of those assigned to the standard evaluation were still enrolled in the clinic at the 84-day follow-up (X2(1) = 1.1, p = .3). There were no significant differences between groups on substance use outcomes at the 84-day follow-up (ANOVA: F(1,291) = .97, p = .33; mixed effects: F(1,288) = .05, p = .83).
3.5. Subgroup analyses: alcohol users
One advantage of large multisite trials is that they allow some analyses of outcome within specific populations of interest. Given that MI was initially developed and validated as an intervention for alcohol use disorders, and that recent studies suggest that MI may be more effective among alcohol, rather than drug-using, populations (Miller et al., 2003
), additional exploratory analyses were conducted to evaluate outcomes for the large subpopulation whose principal substance used was alcohol (n
= 177). For this subgroup, those assigned to MI completed significantly more sessions in the 28 days following randomization compared with those assigned to the standard evaluation session (MI mean = 5.1 sessions (S.D. = 5.1), standard mean = 3.3 (S.D. = 3.2)), for both models (ANOVA: F
(1,175) = 8.1, p
= .01, d
= .56; mixed effects: F
(1,164) = 10.33, p
= .002). The positive effect of MI on treatment retention was also significant at the 84-day follow-up (F
(1,154) = 3.79, p
= .05, d
= .32). Regarding the substance use outcome (i.e., frequency of alcohol use in the 28 days following randomization), the standard ANOVA model including participants from all sites did not suggest significant intervention effects overall (F
(1,107) = .6, p
= .44). However, the mixed effects model suggested that participants assigned to MI used alcohol less frequently than those assigned to standard treatment (F
(1,164) = 3.07, p
3.6. Secondary outcome measures
The ASI, HRBS, and URICA were included as measures of change in psychosocial problems, HIV risk behaviors, and intention to change, respectively. For the ASI composite scores, repeated measures ANOVA for the aggregate sample indicated significant reductions in intensity of problems in all seven areas (medical, legal, employment, alcohol, drug, family, and psychological) over time, for both the 28-day and 84-day assessment points. However, there were no significant effects of intervention or intervention by time. For the HRBS, there were significant reductions in both the drug-risk and sex-risk subscales at the 28-day and 84-day assessment points, but no significant effects of condition or condition by time. Finally, for the URICA, there were no significant effects of time, condition group, or condition by time at the 28-day follow-up for the precontemplation, contemplation, action, or maintenance scores. At the 84-day follow-up, there were significant effects of time only for the contemplation scale, indicating a significant decrease in contemplation scores for participants overall.