Six programs participated: Ayundantes, Española, New Mexico; Brandywine Counseling, Newark, Delaware; Duke Addictions Program, Durham, North Carolina; Mercy Recovery, Westbrook, Maine; Mountain Manor Treatment Center, Baltimore, Maryland; and the University of New Mexico Addiction and Substance Abuse Programs, Albuquerque. Four were methadone programs and 2 were adolescent programs that started using buprenorphine-naloxone for the study. Recruitment was stopped at the Newark (n = 3 patients) and Española (n = 8 patients) sites midway through the study due to slow enrollment; however, treatment and follow-up of randomized patients continued. The numbers of patients at other sites ranged from 29 to 52. The institutional review boards at the University of Pennsylvania and at each trial site approved the study.
Participants and Consent/Assent
The study was open to individuals aged 14 to 21 years who met DSM-IV
criteria for opioid dependence with physiologic features18
and who sought outpatient treatment. Participants aged 18 to 21 years had to provide written consent and correctly answer 9 of 10 questions testing their understanding of the study; for participants aged 14 to 17 years, written assent and written parental consent were required and both participants and their parents had to pass the quiz. Exclusion criteria were having medical or psychiatric conditions likely to make participation difficult or unsafe; abusing alcohol or sedatives or using benzodiazepines for more than 15 days in the last 28 days; having had a sedative overdose in the past 6 months; being unable to provide a urine test result negative for benzodiazepine and methadone (in up to 3 attempts); receiving other addiction treatment; being likely to be incarcerated or to leave the area; breastfeeding or being pregnant; being unable or unwilling to use effective birth control; or receiving psychotropic medication other than a selective serotonin reuptake inhibitor. Participants defined their race and ethnicity using a demographic form standardized for the Clinical Trials Network according to National Institutes of Health policy.
Enrollment and Randomization
Patients were enrolled between July 2003 and December 2005 and randomized to 14-day outpatient detoxification (detox) or 12 weeks of treatment with buprenorphine-naloxone. Randomization occurred through an automated 24-hour service at the Veterans Affairs Cooperative Studies Program in Perry Point, Maryland, that was programmed to randomize patients separately by site. At each site, a biased-coin randomization19
protected against severe imbalance of sex, ethnicity, route of administration, and age across the treatment groups. Age was dichotomized as 14 to 18 years or 18 to 21 years, ethnicity as the majority ethnic group vs all others within the site, and route of administration as injecting or noninjecting. Balance was assessed by comparing the group sum of the binary indicators as each new patient was randomized. If both groups were balanced when a new patient was being randomized, then each group had an allocation probability of 1/2; if there was an imbalance, then the group with the higher score on the sum of indicators received an allocation probability of 1/3 and the other group a probability of 2/3. The indicator data were analyzed by K.D. and K.G.L.
Medication and Dosing
Reckitt Benckiser Pharmaceuticals Inc (Richmond, Virginia) provided medication, and the NIDA coordinated its distribution. Patients receiving buprenorphine-naloxone were instructed to not use heroin or other opioids for at least 6 hours and to be experiencing mild/moderate withdrawal prior to the first dose. The properties of buprenorphine-naloxone were explained during the consent/assent process and reviewed again prior to the first dose so patients understood they needed to hold the medication under the tongue until it dissolved and that it was likely to cause withdrawal if dissolved and injected by someone who was opioid dependent. Medication was administered on site 5 to 7 days per week (patients received take-home doses on days they were not medicated on site if a site was not open 7 days a week), and research assistants or site physicians directly observed it. The first dose was 2-mg buprenorphine with 0.5-mg naloxone. Study personnel observed the patient for 1.5 to 2 hours, and a second dose of 2 to 6 mg (expressed as buprenorphine) was administered if appropriate. On day 2, patients received the dose from day 1 unless considered overmedicated or undermedicated by a clinical assessment, were observed for 1.5 to 2 hours, and the dose adjusted by 2 to 6 mg as needed. On day 3, patients were given the dose from day 2 unless it needed adjustment, observed for 1.5 to 2 hours, and given another adjustment if needed.
Patients in the 12-week buprenorphine-naloxone group received up to a maximum amount of 24 mg per day and began a taper at week 9 that ended by week 12. Patients in the detox group received up to a maximum amount of 14-mg buprenorphine per day and ended their taper by day 14. If a patient missed 3 consecutive days of doses, medication was stopped; it was not restarted for patients in the detox group. Medication was restarted for patients in the 12-week buprenorphine-naloxone group if they returned within 7 days of the last dose. Patients who restarted were given half the amount of the last dose received and observed for 1.5 hours. If the medication was tolerated, they received a portion or the remainder of the dose. Patients who dropped out for missing medication were encouraged to continue in counseling treatment. Adverse events were assessed by weekly vital signs, assessments for sedation and withdrawal, and questions about additional medications received and adverse effects in weeks 1 through 12; similar assessments were done at months 6, 9, and 12. Electrocardiograms and liver enzyme levels were analyzed at baseline and at 4 and 12 weeks.
Patients were scheduled for 1 individual and 1 group session per week with more frequent sessions if needed. Most counselors were licensed clinical addictions specialists or had master's degrees in counseling or social work. Counseling used methods in NIDA manuals20,21
and was standardized by a 3-hour training. One to 3 counselors treated study patients at each site and were supervised using local procedures. Counseling encouraged making positive relationships and stopping drug use, taking medication as prescribed, tolerating stressful events without using drugs, keeping appointments, teaching ways to avoid drug-using situations, educating about addiction, giving positive feedback for achieving goals, referring for treatment of associated problems, and participating in age-appropriate self-help groups.
Primary and Secondary Outcomes
The primary outcome was opioid-positive urine test results at weeks 4, 8, and 12. Urine samples were tested for adulteration (color, specific gravity, temperature), although most patients were not observed during the collection because it was difficult to match female staff with female patients and vice-versa. Two tests were used: the Sure-Step (Inverness Medical Innovations, Bedford, England) that identifies amphetamine, barbiturate, benzodiazepines, cocaine, methadone, methamphetamine, morphine, hydrocodone, hydromorphone, oxycodone, phencyclidine, and tetrahydrocannabinol; and the Rapid One OXY (American Bio Medica Corp, Kinderhook, New York), which is more sensitive to oxycodone.
Secondary outcomes were dropout from the assigned condition, self-reported use, injecting, enrollment in addiction treatment outside the assigned condition, other drug use, and adverse events. Patients were considered dropouts if they missed medication for 3 consecutive days if in the detox group or 7 consecutive days if in the 12-week buprenorphine-naloxone group, did not have an individual or group session lasting 30 minutes or more for 14 consecutive days, enrolled in other addiction treatment, asked to be withdrawn, went to jail, or died. Follow-up visits at months 6, 9, and 12 included assessing self-reported use of opioids, alcohol, marijuana, and cocaine and injecting in the past month and determining whether patients were receiving other addiction treatment. Research assistants likely knew group assignments because the study was not blinded. Patients were paid $5 each for weekly assessments and $75 each for assessments at weeks 4, 8, and 12 and months 6, 9, and 12.
General estimating equation (GEE) models compared groups on longitudinal outcomes using a compound symmetry, working correlation structure and empirical standard errors that can accommodate dichotomous dependent variables.22
Explanatory variables in models examining urine test–confirmed opioid use were baseline status, site, treatment group, time (as a categorical variable), and treatment group × time interactions. Sample sizes for 2 sites (those with 3 and 8 patients) prevented assessment of group × site interactions. In analyses excluding these sites, group × site interactions were not observed; thus, the models presented include data from all sites and do not include a group × site interaction term.
A pattern-mixture model23
was used to assess the impact of missing data on urine test results. Pattern mixture models extend the basic repeated measures by including a variable that describes the main patterns of missing data as a main effect and an interaction with other variables (week and group). Significant interactions with the missing data indicator on the main variables suggest that its effects differ across levels of missing data and that missing data may not be ignorable. Following suggested guidelines,23
we used time of last data provision (a categorical variable representing week 4, 8, or 12) as the missing variable. Another approach often taken is to impute missing tests as positive. If results obtained for the original and imputed models differ substantially, missing data may not be ignorable. Both methods were used to evaluate the effects of data on the primary outcome wherein missing urine test results were counted as opioid positive.
General estimating equation models examined group differences for binary secondary outcomes (retention, self-reported drug use, injecting). Models were similar to those outlined previously except that baseline status was not included in the self-reported opioid and retention analyses due to lack of variability. When models failed to converge (ie, self-reported cocaine and marijuana use, injection use), Mantel-Haenszel analyses were performed that examined use during the whole time period and stratified on site. To assess group differences on cross-sectional outcomes, logistic regression analyses were used for binary outcomes (nonstudy treatment, received other treatment), and a generalized linear model was used for number of counseling sessions attended. These models included terms for condition and site.
The study was designed to have 80% power to detect a difference of 18% between the groups at each of the 3 time points at a significance level of 5% and, assuming a 30% loss to attrition, a within-subject repeated-measures correlation of 0.5. With an additional adjustment to allow for nesting effects due to multiple sites, this yielded a required sample size of 120 per group. The study randomized only 78 patients to detox and 74 patients to receive 12 weeks of buprenorphine-naloxone, rather than the 120 originally planned. With the same assumptions as used for the original design, this would yield a power of only 58% for the original target effect. In the study, the attrition, within-person correlation, and site effects were comparable with the design assumptions. However, the effect sizes at weeks 4 and 8 were larger than expected (35% and 31%, rather than the planned 18%) while the effect at 12 weeks was smaller (8% rather than 18%). Thus, although power was lower for the designed effect, the observed effects were larger. Statistical analyses were performed with SAS version 9.1 (SAS Institute, Cary, North Carolina).