Internet-Delivered Treatment for Substance Abuse: A Multisite Randomized Controlled Trial

Patients seeking treatment for drug or alcohol problems at 10 community-based outpatient treatment programs nationwide (affiliated with the National Drug Abuse Clinical Trials Network) were enrolled between June 2010 and August 2011. Details of program selection and characteristics have been reported elsewhere (4). Outpatient addiction treatment programs were selected for geographic and patient diversity, and they varied in programming, consistent with the goals of an effectiveness trial to promote external validity. Programs had to offer at least two face-to-face therapeutic group or individual sessions per week, lasting at least 2 hours, with most offering two to six sessions per week. Each program was asked to enroll approximately 50 patients (range=38–60).

After completing a 2–3 hour baseline assessment, patients were randomly assigned, in a 1:1 ratio, to receive 12 weeks of either treatment as usual or treatment as usual plus TES, with the intervention substituting for approximately 2 hours of clinician time (the equivalent of two Internet-delivered modules, twice a week; i.e., four modules per week, or 48 modules over 12 weeks). All participants were asked to provide urine drug and breath alcohol screens twice a week. Randomization was conducted by an independent statistician at a centralized data center in randomly permuted blocks, stratified by treatment site, patient’s primary substance of abuse (dichotomized as stimulant versus nonstimulant, since contingency management has been tested most for cocaine dependence [17, 18]), and whether or not the patient was abstinent at the baseline assessment and study entry based on urine drug and breath alcohol tests. Abstinence at study entry is a strong predictor of outcome (19) and thus is arguably an important covariate in primary outcome analyses (20). Patients and staff were not blind to treatment arm. Additional details of the study design and rationale have been reported elsewhere (4).

Patients were eligible if they were age 18 or older; indicated by self-report that they had used illicit substances in the 30 days before study entry, or 60 days for those exiting a controlled environment (alcohol could be the primary problem, but patients had to have used at least one illicit drug as well); had entered the treatment episode within the past 30 days (randomization occurred on average 9.5 days [SD=7.4] after treatment entry); were planning to remain in the area and in the treatment program for at least 3 months; and were proficient in English. Patients were excluded if they were being treated with opioid replacement therapy (e.g., buprenorphine, methadone) or were unable to provide informed consent. The study was approved by the institutional review boards of the New York State Psychiatric Institute and all participating clinical sites. Patients provided written informed consent after receiving a complete description of the study.

TES (12) includes contingency management and 62 interactive multimedia modules, based on the community reinforcement approach, requiring approximately 20–30 minutes each to complete. The community reinforcement approach is grounded in the premise that drugs compete with more delayed prosocial reinforcers; hence, the treatment promotes skills training to teach, encourage, and increase satisfaction with drug-free sources of reinforcement (21). An initial training module that teaches patients how to use the program is followed by modules on basic cognitive-behavioral relapse prevention skills (e.g., drug refusal, managing thoughts about using, conducting functional analyses). Subsequent modules teach skills aimed at improving psychosocial functioning (e.g., communication, mood management, family/social relations, time management), as well as prevention of HIV, hepatitis, and sexually transmitted infections. Video clips show actors modeling the skills being taught. Short quizzes assess the patient’s grasp of material; the pace and level of repetition of material is adjusted accordingly to maximize individual mastery of the skills and information being taught.

Each clinic received computers for on-site delivery of the intervention; patients could also access the intervention outside the clinic via the Internet. Treatment program clinicians, who had patients in the Internet-based condition, were asked to incorporate brief discussion of module completion into individual counseling sessions. An electronic reporting system allowed clinicians to view summaries of their patients’ computer activity. According to clinicians’ documentation, most individual treatment-as-usual sessions (85.3%) included discussion of the patient’s participation in the computer intervention.

TES includes a flexible system for delivering contingency management according to the prize-based incentive system developed by Petry and colleagues for delivering low-cost contingency management in community-based treatment settings (17, 18). Incentives take the form of opportunities to draw vouchers from a virtual “fishbowl.” Some vouchers provide congratulatory messages (e.g., “Good job”), while others are exchangeable for prizes of mostly modest value (usually around $1, occasionally around $20, rarely $80–$100). In the present study, draws were awarded for abstinence (based on negative urine or breath alcohol screens) and for completion of modules (up to the recommended four per week, although there was no cap on the number of modules that could be completed). Research staff entered target behaviors into the computer and oversaw prize distribution.

Twice a week during the 12-week treatment phase (i.e., 24 half weeks), and again at 3- and 6-month follow-up visits, urine was collected and screened for 10 drugs with standard lateral flow chromatographic immunoassays (QuickTox dip card) and temperature and adulterant test strips, and self-report drug and alcohol use data were collected using the timeline follow-back calendar method (22). Each half week of treatment was categorized as abstinent if the urine screen was negative and the self-report indicated no drug use or heavy drinking days (according to the National Institute on Alcohol Abuse and Alcoholism guidelines: >4 drinks a day for men and >3 drinks a day for women), and not abstinent otherwise. If self-report was missing but the urine screen was positive, the half week was scored as not abstinent. Abstinence during a given half week was considered missing if 1) self-report indicated no use but the urine screen was missing, 2) the urine screen was negative but self-report was missing, or 3) both urine and self-report were missing. Abstinence at the 3- and 6-month follow-up visits was scored similarly based on the urine screen and the last 4 days of self-report data.

Research staff tracked each patient’s participation in the community-based treatment program. If a patient dropped out of treatment, the event was scored as the last week that a patient attended a face-to-face group or individual therapy session at the treatment program (range=0–11 weeks). Patients who attended treatment in week 12 were considered censored at that point.

Means and standard deviations or frequencies and percentages were calculated for baseline characteristics of the randomized sample. Two primary outcome measures—abstinence from drug or heavy alcohol use in the last 4 weeks of treatment and retention in treatment (time to dropout)—were prespecified in the study protocol, along with their respective data analysis plans (4). Sample size computations were based on a Bonferroni adjustment approach (i.e., each hypothesis test has a significance level of 0.025) with 80% power to detect an odds ratio of 1.5 for the abstinence outcome (12) and 90% power to detect 50% versus 35% (Internet-based intervention versus treatment as usual) retention (17) with 500 participants. The dichotomous abstinence scores for each of the 24 half weeks in the 12-week treatment phase were analyzed using a repeated-measures piece-wise logistic model, where a linear time-by-treatment interaction was allowed during the first 16 half weeks (8 weeks) but a constant study intervention effect was assumed during the last 8 half weeks (4 weeks). Generalized estimating equations (23) were utilized to adjust for the correlation of half weeks within patients. Missing half week data were excluded; the median number of missing half weeks during the last 8 weeks was 1 (interquartile range=4) for both treatment arms. The stratification factors (treatment site, primary substance [stimulant versus nonstimulant], and abstinence at baseline/study entry) were included in the model as main effects. During subsequent model building, the interaction of each covariate with treatment was tested and considered significant if the p value was <0.100.

The primary retention outcome (time to dropout) was analyzed with survival methods stratified by site, using a log-rank test and a proportional hazards model to consider effects of the stratification factors as covariates (24). Schoenfeld residuals were used to test the assumptions of proportional hazards (25).

Two summary outcome measures typically reported in other clinical trials—the total number of abstinent half weeks and the greatest number of consecutive abstinent half weeks—were prespecified as secondary outcome measures and were analyzed using a Wilcoxon rank-sum test. Missing half weeks were imputed as not abstinent. At the 3- and 6-month follow-up points, the log-odds of abstinence was modeled as a function of visit, treatment assignment, and stratification factors, using generalized estimating equations to adjust for the correlation within patients. Missing half week data were excluded. All analyses used the intent-to-treat sample and were conducted using SAS, version 9.2 (SAS Institute, Cary, N.C.).

Patients in the TES group completed a mean of 36.6 computer-delivered modules (SD=18.1) out of a recommended 48 (range=0–72); 22% of Internet sessions were completed off-site. TES patients earned a mean of 118 (SD=90) voucher draws (out of a possible 252 draws) contingent on abstinence or module completion, resulting in a mean of $277 (SD=226) worth of prizes over 12 weeks.

Patients in the treatment-as-usual and TES groups attended similar numbers of treatment-as-usual therapy sessions at their treatment programs (TES group: mean=21.2, SD=17.5; treatment-as-usual group: mean=20.4, SD=17.5) and similar numbers of sessions per week in the weeks prior to dropout (TES group: mean=1.4, SD=0.9; treatment-as-usual group: mean=1.3, SD=0.9). Notably, these average numbers of sessions attended are lower than the two to six sessions per week typically recommended across the participating treatment programs.

Results of the logistic regression modeling abstinence are summarized in Table 2. Model 1 includes the main effects of treatment and stratification factors. Compared with patients in the treatment-as-usual group, those receiving TES had greater odds of abstinence at the end of treatment, by a factor of 1.62 (p=0.010). Main effects of abstinence at baseline/study entry and treatment site were also significant. Abstinence at baseline/study entry strongly predicted abstinence at the end of treatment. Sites varied in the overall rates of abstinence their patients achieved. There was no significant main effect of primary stimulant use. Interactions of primary stimulant use by treatment and site by treatment were not significant. The interaction of abstinence at baseline/study entry by treatment (p=0.068) was included in model 2, in which the effects of each treatment condition were estimated separately in the abstinent and nonabstinent strata. Among patients who were not abstinent at baseline/study entry, those in the TES group had more than twice the odds of abstinence compared with those receiving treatment as usual, whereas among those who were abstinent, there was no significant difference between conditions. Figure 1 presents the observed rates of abstinence by half week across the 12-week trial, along with rates at the 3- and 6-month follow-ups, stratified by abstinence at baseline/study entry.

Compared with patients receiving treatment as usual, those receiving TES achieved significantly more total half weeks of abstinence during the 12-week trial (mean=11.1 [SD=9.0] compared with mean=8.8 [SD=8.2]; p=0.008) and more consecutive abstinent half weeks (mean=8.0 [SD=8.1] compared with mean=5.1 [SD=6.1]; p=0.001).

At the 3- and 6-month follow-ups, abstinence at baseline/study entry continued to significantly predict abstinence (odds ratio=2.39, 95% CI=1.67, 3.42, p<0.001). The effect of TES compared with treatment as usual was no longer significant.

There was less dropout from treatment among patients in the TES group than in the treatment-as-usual group (log-rank p=0.017) (Figure 2). The proportional hazards model yielded a main effect of treatment (hazard ratio=0.72, 95% CI=0.57, 0.92, p=0.010) similar to, and an effect of abstinence at baseline/study entry analogous to, the primary abstinence outcome: patients who were abstinent were less likely to drop out (hazard ratio=0.66, 95% CI= 0.51, 0.86, p=0.002). There was no main effect of the stratum of primary stimulant use, nor were there significant stratum-by-treatment interactions.

Strengths of this effectiveness study include the randomized controlled design, the prespecification of primary outcome measures and analyses, the high follow-up rates, and the relatively low rates of missing outcome data. The outcome measures chosen are germane, as abstinence is the primary goal of treatment and dropout from treatment is a substantial problem that limits the effectiveness of outpatient addiction treatment (31). As befits an effectiveness trial, the study was conducted with a large, demographically and geographically diverse sample, and eligibility criteria were kept broad. These features suggest that the sample is likely to be representative of patients seeking community-based treatment for substance abuse problems across the United States, with findings reflective of how the intervention performs when integrated into real-world treatment settings.

A main limitation is that the study tested TES as a package, compared with treatment as usual, in a two-arm design. Thus, it is not possible to disentangle the unique effects of the computerized community reinforcement approach and of contingency management. The two-arm design has the advantage of simplicity, which is a consideration in community-based effectiveness trials. Furthermore, previous research with clinician- and computer-delivered community reinforcement approach and contingency management techniques have suggested that both contribute to beneficial treatment effects (15, 32–35). However, future research should attempt to disentangle the effects of the two components in both community-based addiction settings and non-specialty settings.

The superiority of TES over treatment as usual was not sustained at longer-term follow-up. The effects of contingency management interventions may diminish once the contingencies end (18). In contrast, some (36), although not all (37), studies of CBT for addictions have observed that the benefits of the intervention actually increased after treatment, suggesting, as one would hope, that the patients learned skills that they continue to practice and benefit from over time. It may be that the beneficial effect of TES observed during the active treatment phase was mainly attributable to the contingency management component of the intervention, although this was not tested in the present study. A diminishing intervention effect over time is consistent with the chronic, relapsing nature of addiction (38) and the need for ongoing monitoring and treatment. In the present study, TES was available to patients only during the 12-week trial, but since it is Internet accessible, it could be made available to patients indefinitely, an option that should be studied.

Finally, a number of patients were eligible for the study but did not enroll. This raises the generalizability question of whether the unenrolled patients might have responded differently to the intervention. A previous analysis of the screening data (39) showed that the unenrolled patients reported more drug use compared with their enrolled counterparts. The greater relative benefit of TES among patients who were not abstinent at baseline/study entry (associated with more drug use) suggests that those not enrolled might have benefited. Engaging patients at the outset of an episode of outpatient treatment remains a challenge that needs to be addressed.