The effectiveness of the two interventions was established by using intent-to-treat analyses, that is, all participants assigned to each group were retained in analyses whether they actually received the intervention or not. There were no missing data for linkage status.
4.1. Baseline sample
A total of 678 CIU clients consented to participate in the study and completed the baseline interview. The sample included 429 men (63.3%) and 249 women (36.7%). The sample was 51.0% Caucasian and 48.1% African American. The average age of the clients was 33.40 years (SD = 10.04, range 18−64). Only 58.0% of the clients were employed in the six months before coming to the CIU, and 20.2% of the clients were homeless. Half (50.3%) of the respondents described themselves as currently involved in the criminal justice system. The majority of the clients (70.1%) had prior treatment experience. Participants identified their most serious substance problem as: crack or powder cocaine (45.6%), heroin (22.0%), marijuana (14.9%), alcohol (11.6%), or other drugs (5.9%). The CIU referred approximately half (49.7%) of the participants to drug free outpatient treatment and another 31.9% were referred to residential programs. Only 18.4% of participants were referred to an outpatient methadone maintenance program. Despite random assignment, the intervention conditions did differ on two of the variables. Participants assigned to the MI condition were lower in psychiatric severity and were also more likely to say that they did not have a substance abuse problem. There were no other significant pre-existing differences among the groups. See for a complete description of sample characteristics broken down by intervention condition.
Comparison of Sample Characteristics by Total Sample and Intervention Condition
There were no significant differences in the three month follow-up interview rate for the three study groups: SOC, 85.2%; MI, 83.6%; SBCM, 88.3%. Participants in the SBCM group attended a mean of 2.3 sessions; linkers attended 2.9 sessions and non-linkers, 1.6 sessions.
4.2. Linkage to treatment by condition
The linkage rate for the total sample was 46.0%. The relationship between intervention condition (SOC, MI, or SBCM) and linkage was examined using a Chi-Square test of association. There was a significant relationship between the type of intervention participants received and whether they linked with treatment, χ2 (2, N = 678) = 12.26, p < .01. To test the specific a priori hypotheses about differences between the interventions, a series of z-tests of proportions was conducted. Results revealed that there was no significant difference between the proportion of participants in the SOC group who linked (38.7%) when compared with participants in the MI group who linked (44.7%), z = 1.30, p > .05. However, participants in the SBCM group (55.0%) did link at a significantly higher rate than the SOC group, z = 3.46, p < .001. There was also a significant difference in linkage rate between the MI group and the SBCM group, z = 2.17, p < .05. contains the linkage rate for each study condition.
Percent of Participants Who Linked by Intervention Condition and Treatment Modality
4.3. Linkage by treatment modality
After assessment at the CIU, participants received referrals to one of three types of programs: inpatient residential treatment (n = 216), outpatient drug-free (n = 337), or methadone maintenance (n = 125). Linkage rates for the three modalities were: inpatient residential (46.3%), outpatient drug free (41.2%), and methadone maintenance (58.4%). A Chi-Square test of association showed that there was a relationship between treatment modality and linkage with treatment, χ2 (2, N = 678) = 10.81, p < .01.
Chi-Square analysis was also used to look at the effect of group assignment (SOC, MI, or SBCM) on linkage in each of the treatment modalities. For the residential treatment modality there was no relationship between group assignment and whether participants linked, χ2 (2, N = 216) = 4.67, p > .05. Nor was there a relationship between assignment and linkage for participants assigned to methadone maintenance, χ2 (2, N = 125) = 3.35, p > .05. There was, however, a significant relationship in the outpatient treatment modality, χ2 (2, N = 337) = 13.17, p < .01. These results were further examined with a series of post hoc z-tests of proportions with Bonferroni correction for familywise error. There was no significant difference in the proportion of participants in the SOC group who linked with outpatient treatment (28.7%) and participants in the MI group who linked (43.4%), z = 2.31, p > .05. Participants in the SBCM group, however did link at a significantly higher rate (52.3%) than participants in the SOC group, z = 3.60, p < .01. Finally, there was no significant difference between the MI group and the SBCM group in linkage to outpatient treatment, z = 1.33, p > .05. See for modality linkage rates.
4.4. Predictors of linkage to treatment
A logistic regression analysis was conducted to identify the predictors of linkage to treatment. Predictor variables included: gender, race, age, education level, homelessness, involvement in the criminal justice system, problem severity, and prior treatment experience. Participant perceptions of barriers to treatment and pre-treatment motivation were also included in the model. Finally, dummy variables were included to indicate whether or not participants received case management or motivational interviewing.
The results of the analysis, presented in , clearly indicated that the set of predictors successfully distinguished between participants who linked with treatment and those who did not, χ2 = 85.58, p < .001, Nagelkerke R2 = .16. The model also demonstrated the ability to correctly predict participants' outcome category (linkage vs. no linkage). Specifically, the model correctly predicted the outcome category of 73.0% of the participants who did not link and 59.9% of those who did link. The overall prediction success rate was 67.0%.
Significant Predictors of Linkage for Overall Sample and Each Intervention Condition
Six different variables included in the model were significant predictors of linkage across the entire sample. The strongest predictor was whether participants were in the SBCM condition [z = 13.45, p < .001, odds ratio (OR) = 2.13]. Participants in the SBCM condition were 2.13 times more likely to link with treatment than those who were not in the SBCM condition. MI, however, was not a significant predictor of linkage (z = 1.08, p > .05). Men were less likely to link (z = 4.04, p < .05, OR = .69). Participants with more years of education were more likely to link (z = 4.43, p < .05, OR = 1.11). Participants who were homeless were less likely to link with treatment (z = 3.98, p < .05, OR = .68) while participants who had prior treatment experience were more likely to link with treatment (z = 8.06, p < .01, OR = 1.75). None of the scales from the BTI or the PRS were significant predictors in this model; however, the ASI composite score which measures alcohol use was, (z = 14.77, p < .001, OR = .20).
4.5. Predictors of linkage by type of intervention
To examine whether the pattern of predictors of linkage differed depending on the type of intervention participants received, three separate logistic regressions were conducted. The same set of predictors used for the entire sample was also included in logistic regression analyses for each study condition. Results are presented in .
4.5.1. SOC group
For participants who were randomly assigned to receive the standard of care by the CIU, the logistic regression model indicated that the set of predictors did not successfully distinguish between those who linked to treatment and those who did not, χ2 = 39.72, p = .05; Nagelkerke R2 = .22. There were only two significant predictors of linkage for participants in the SOC group. The first of these was gender, z = 4.54, p < .05, OR = .48. Men were less likely to link with treatment than were women. The second predictor was Problem Recognition, one of the scales from the PTR, z = 5.61, p < .05, OR = .927. Participants who showed less problem recognition were more likely to link with treatment.
4.5.2. MI group
There were two significant predictors of linkage for the MI group. First, participants with higher levels of education were more likely to link with treatment, z = 4.32, p < .05, OR = 1.23. Second, participants who had higher alcohol use problems, as measured by the ASI, were less likely to link with treatment, z = 7.28, p < .01, OR = .064.
4.5.3. SBCM group
There was only one significant predictor of linkage to treatment in the SBCM group. Participants who had more severe alcohol use problems, as measured by the ASI, were less likely to link with treatment, z = 6.46, p < .05, OR = .16.