Brief Behavioral Interventions for Substance Use in Adolescents: A Meta-analysis

We used established methodologies as outlined in the Agency for Healthcare Research and Quality (AHRQ) Methods Guide for Effectiveness and Comparative Effectiveness Reviews.²²  The key questions and protocol were discussed in depth with a panel of key informants and technical experts.

A medical librarian with expertise in designing searches for systematic reviews conducted literature searches in Medline, the Cochrane Central Register of Controlled Trials, Embase, and Cumulative Index to Nursing and Allied Health Literature from inception to October 31, 2019. The overall search was designed to cover the full range of key questions in the full report and is broader than the scope of interventions and outcomes addressed here. In the searches, the we included terms for substances (including a list of known substances) crossed with terms for disorders and also terms for treatments (including a list of known treatments). The search was further crossed with terms for the adolescent age group and terms for primary studies and systematic reviews. All searches were designed to be maximally sensitive. Full search strategies can be found in the online supplement. We further searched ClinicalTrials.gov and the US Food and Drug Administration websites for unpublished and ongoing studies. Reference lists of available clinical practice guidelines and existing systematic reviews were scanned for eligible studies. The results of all studies were deduplicated and screened by 2 reviewers using the online program abstrackr (http://abstrackr.cebm.brown.edu/). Accepted citations were retrieved and reviewed in full text.

For the research question addressed here, studies had to be randomized controlled trials (RCTs) in which researchers compared ≥2 interventions with at least 10 participants per arm. RCTs had to have been focused on adolescents, aged 12 to 20 years inclusive (ie, ≥80% of the population had to be within this age range), who met criteria for at least 1 SUD or for problematic SU (excluding tobacco). Problematic use was operationally defined as meeting at least 1 of the following criteria: (1) referral for treatment by self, parent, school, other professional, or the justice system; (2) screened by using a validated tool, with a BI given to those who met prespecified criteria indicating elevated risk; (3) reported SU at least once per month; or (4) identified after a substance-related consequence, such as an alcohol-related emergency department visit. Although in the full systematic review, a range of interventions was evaluated, to be eligible for the current analysis at least 1 of the evaluated interventions had to be designated as brief, defined as 1 or 2 sessions. We excluded interventions focused on drinking in the college setting because (1) this population is developmentally distinct from adolescents with problematic SU and (2) and such studies have been the focus of multiple previous systematic reviews.^23–25 

Two members of the team with expertise in psychology and complex interventions created a categorization schema to capture the salient aspects of the behavioral interventions and independently assigned ≥1 intervention code(s) for each arm in each study. Disagreements were resolved by consensus.

An intervention was coded as MI if at least 1 session was focused on building the adolescent’s motivation to reduce SU and/or attain abstinence. Motivation enhancement therapy, a more structured and specific approach to building the adolescent’s motivation, was also categorized as MI. Interventions that referenced a general goal to build the adolescent’s motivation to change were not coded as containing MI unless there was a stand-alone, manual-guided MI component.

Interventions were categorized as psychoeducation if they explicitly aimed to reduce the adolescent’s SU through provision of education about substance-related harms. Because most SU interventions involve some degree of education, we coded an intervention as psychoeducation only if there was explicit reference to a stand-alone psychoeducation module.

Interventions designed to be comparators to active interventions that were not directed at treating SU were categorized as treatment as usual (TAU). Examples included wait lists or pamphlets regarding issues other than SU.

In 3-arm studies comprising 2 treatment arms that were not distinguishable by using our taxonomy (eg, computer-delivered MI versus therapist-delivered MI versus TAU), we pooled the nondistinguishable arms and included the pooled arm in meta-analyses.

We examined outcomes related to frequency of use of specific substances (eg, alcohol, cannabis), whether reported as continuous measures reflecting frequency of use or as categorical measures of abstinence. For alcohol, we considered both frequency of heavy alcohol use (eg, mean days of heavy alcohol use per 30 days) and frequency of any use (eg, percent days of alcohol use per 30 days). For cannabis, we considered only frequency of any use.

Aggregate outcomes that combined multiple substances were classified into 1 of 3 mutually exclusive categories: (1) alcohol and other drugs, (2) illicit drug use (excluding alcohol but including cannabis and other drugs, regardless of local laws), and (3) other drugs (excluding alcohol and cannabis).

We also examined scales that measured problems related to SU. We extracted the mean value of the various scales reflecting substance-related problems. When a study reported several scales, each associated with a specific substance (ie, alcohol or cannabis), we chose the scale with the highest mean severity.

Two senior investigators assessed the RoB (methodologic quality) of each study on the basis of predefined criteria, using the Cochrane RoB tool for RCTs.²⁶ 

We conducted pairwise meta-analyses, using both frequentist and Bayesian frameworks, and network meta-analyses, using the Bayesian framework. Analyses were done by using R,²⁷  with the metafor²⁸  and gemtc packages.²⁹  Network meta-analysis is an extension of pairwise meta-analysis that allows simultaneous comparison of multiple interventions by combining direct evidence (when interventions are compared head-to-head) and indirect evidence (when interventions are compared through other reference interventions across studies). In Bayesian analyses, we used empirical previous distributions for the between-study heterogeneity variance.^30–32 

Outcomes related to frequency of SU were converted to a common metric of mean number of use days per 30 days. We analyzed outcomes evaluated ∼3 months after baseline assessment. Effects were estimated for net mean differences (NMDs), that is, the between-arm differences in the within-arm changes from baseline (difference-in-differences). Standardized net mean differences (SNMDs) were calculated for SU-related problem scales. Abstinence outcomes were compared to odds ratios (ORs). As a sensitivity analysis, we included all available substance-specific outcomes (ie, use days, scales, and nonlinear transformations of use days) and calculated a summary SNMD effect. Statistical heterogeneity was explored qualitatively. The a priori subgroup analyses of interest (male versus female, racial and ethnic minorities, socioeconomic status, and family characteristics) were too sparse for meaningful meta-regression or subgroup analysis.

We graded the SoE as per the AHRQ Methods Guide for Effectiveness and Comparative Effectiveness Reviews.³³  On the basis of this multidimensional assessment, which accounts for study RoB, consistency across studies, precision, and other factors, we assigned a SoE rating of high, moderate, low, or insufficient.

In our search for the full AHRQ systematic review, 33 732 citations were returned. Of 1831 abstracts accepted in initial screening and retrieved for full-text review, 988 that were focused on older adults were excluded (Fig 1). An additional 626 articles were rejected for reasons provided in Figure 1. In the full systematic review, 118 studies were included, and the review was also comprised of studies of nonbrief behavioral interventions and pharmacologic interventions. Here, we report on the results most relevant for primary care practitioners, the 22 studies that were used to evaluate BIs for adolescents with problematic alcohol or cannabis use.^34–55  The baseline and intervention details of these 22 studies are summarized in Supplemental Table 1. Despite superficial heterogeneity of study inclusion criteria with respect to the targeted substance(s), in most studies, researchers enrolled adolescents using some combination of alcohol and cannabis, with a minority using other drugs.

Interventions were coded as MI, psychoeducation, or TAU. Although included in the network meta-analysis, there was insufficient evidence to evaluate effects of either psychoeducation versus TAU or MI versus psychoeducation. The outcomes reported by each study are detailed in Supplemental Table 2. Aggregate outcomes (ie, alcohol and other drug use, illicit drug and other drug use) were rarely reported, precluding meta-analysis.

RoB summaries are presented graphically (Supplemental Fig 7) for the 22 included studies. The most common methodologic concerns involved lack of blinding of participants (all RCTs had a high RoB), personnel (21 had a high RoB), and outcome assessors (12 had a high RoB). The evidence profile used to define overall SoE by outcome is provided in Supplemental Table 3.

In 7 studies with 2821 participants, researchers reported a measure of heavy alcohol use.^{34,35,37,40,41,49,54}  In all 7 studies, researchers directly compared MI with TAU (Supplemental Fig 8). Across the 5 studies that reported days of heavy alcohol use (1248 participants),^{35,37,40,41,54}  the NMD was −0.7 (95% credible interval [CrI] −1.6 to 0.0) days per month of heavy alcohol use, favoring MI (Fig 2). This corresponds to a Bayesian posterior probability of 97.3% that MI is superior to TAU. Results were similar in a sensitivity analysis that included 2 additional studies ^34,49  (N = 1573) that reported a scale reflecting heavy alcohol use. Because of moderate RoB and imprecision, we deemed there to be low SoE supporting this conclusion.

In 10 studies^* (N = 2153 participants), a measure of the frequency of alcohol use was reported. As shown in the evidence network (Supplemental Fig 9), in 9 studies, researchers directedly compared MI to TAU, and, in 1 study, researchers evaluated MI versus psychoeducation. In 7 of the 9 studies in which researchers directly compared MI to TAU, days of alcohol use was reported.^{37,39–41,46,47,54,55}  The summary estimate from the Bayesian network meta-analysis for the NMD used to compare MI to TAU was −1.1 (95% CrI −2.2 to −0.3) alcohol use days per month (Fig 3), favoring MI. Conclusions were unchanged in a sensitivity analysis in which the 2 studies^34,49  that reported a scale were included. Because of moderate RoB but adequate precision, we deemed there to be moderate SoE supporting this conclusion.

Abstinence from alcohol use was reported in 7 studies (N = 2482 participants).^{34,39,44–46,50,55}  As shown in the evidence network (Supplemental Fig 10), in 5 studies, researchers compared MI to TAU, and 2 studies were used to evaluate MI versus psychoeducation. The summary OR comparing MI to TAU was 1.9 (95% CrI 0.9 to 6.0), favoring MI (Supplemental Fig 11). The CrI for this effect is wide and compatible with no effect. Because of moderate RoB and imprecision that included a null effect, we rated the SoE as insufficient.

In 13 studies, the effect of BI on cannabis use days was analyzed (N = 2386).^† In all studies in the network, the number of cannabis use days was reported. As shown in the evidence network (Supplemental Fig 12), 1 study had 3 arms (MI, psychoeducation and TAU), MI was compared to TAU in 9 studies, and MI versus psychoeducation was evaluated in 3 studies. The estimate from the Bayesian network meta-analysis for the effect of MI versus TAU was −0.05 (95% CrI −0.26 to 0.14) days per month (Fig 4). This result is used to support the conclusion that MI is not more effective than TAU in reducing cannabis use days. Because of moderate RoB, we rated the SoE as moderate.

In 6 studies, cannabis abstinence outcomes were reported (N = 1119).^{36,39,45,46,50,55}  Of these, MI was compared to TAU in 4 studies and MI versus psychoeducation was evaluated in 2 studies (Supplemental Fig 13). The summary OR for MI versus TAU was 1.5 (95% CrI: 0.7 to 3.4) (Supplemental Figs 13 and 14). Because of moderate RoB and imprecision compatible with a null effect, we rated the SoE as insufficient.

In 9 studies, 1 of 8 SU problem scales (N = 1854) was reported.^{35,40–43,46,53–55}  Of these, psychoeducation was compared to MI in 3 studies and MI versus TAU was evaluated in 6 studies (Supplemental Fig 15). After standardizing the scales, the pooled estimate from the network meta-analysis was –0.5 (95% CrI –1.0 to 0) SNMD units, ie, MI is more effective than TAU to reduce SU-related problems (Fig 5). Because of moderate RoB and imprecision, we rated the SoE as low.

We found that the use of MI reduces heavy alcohol use (low SoE) and overall alcohol use (moderate SoE) compared to TAU in adolescents with problematic alcohol use. By contrast, we found that the use of MI does not reduce days of cannabis use compared to TAU (moderate SoE). One interpretation of these findings is that adolescent cannabis use is more resistant to change than adolescent drinking. We also found evidence that MI may reduce overall substance-related problems relative to TAU (low SoE).

Our findings for alcohol use outcomes are generally consistent with those of previous systematic reviews and meta-analyses that have been focused on adolescent drinking. In a systematic review of experimental and quasi-experimental studies, Tanner-Smith and Lipsey¹⁷  pooled data across multiple outcomes and interventions using standardized effect size metrics. They concluded that adolescents aged 11 to 18 years who received BIs for alcohol use had lower levels of self-reported alcohol consumption and alcohol-related problems than those who received TAU and that MI strategies were the most effective BIs for adolescent drinking.¹⁷  We use our replication of these findings using network meta-analysis, “natural units” of days of use, and rigorous SoE assessments to speak to the robustness of the results.

Comparison of our findings for cannabis use with previous meta-analyses is complicated by the fact that cannabis use has often been conflated with “illicit drug use.” For example, Tanner-Smith et al¹⁹  concluded that MI explicitly targeting alcohol use in adolescents and young adults did not reduce concurrent use of “other illicit drugs,” whereas MI explicitly targeting both alcohol and illicit drug use was more effective than TAU in reducing “use of both substances.” In a systematic review, Li et al²⁰  similarly examined the effects of brief MI on “illicit drug use” among adolescents. In this review, the researchers included 10 included RCTs whose participants predominately reported cannabis use (80%) but also reported using multiple illicit substances, including cocaine (30%), amphetamines, and 3,4-methylenedioxy-methamphetamine (20%). Li et al²⁰  aggregated “extent of drug use outcomes” (drug use frequency, abstinence, problems related to drug use, etc) and found no evidence for an effect of MI compared to TAU (standardized effect size of 0.05 [95% confidence interval: −0.06 to 0.17]). In a recent meta-analysis, Halladay et al²¹  focused specifically on the effectiveness of BIs on cannabis use outcomes in adolescents and emerging adults. They reported no significant difference in frequency of cannabis use at 1 to 3 months postintervention between BI and passive control, with a standardized mean difference of – 0.05 (95% confidence interval: –0.14 to 0.03). Our findings, specific to cannabis use in the adolescent age group, are consistent with the Halladay et al²¹  meta-analysis, and, with them, we suggest that using MI is no more effective than using TAU to reduce days of cannabis use.

Given the apparent heterogeneity of treatment effects for alcohol and cannabis, it may be problematic to interpret intervention effects when the substance targeted is unclear, particularly when outcomes pertain to multiple substances (eg, “alcohol and other drugs” or “illicit drugs”), as has been done in previous systematic reviews.^17,20,56  If treatment effects vary by substance, estimates of effect on such composite outcomes will be determined by the relative proportion of alcohol, cannabis, and other drug use in individual studies. Because many researchers enroll adolescents with mixed use of alcohol, cannabis, and (less commonly) other drugs, we recommend that, in future systematic reviews, researchers should report substance-specific effects.⁵⁷ 

The results of our systematic review must be considered in the context of several design decisions. First, we aggregated BIs from studies across settings, which may have masked meaningful contextual factors that can modify treatment effectiveness. Second, although some consider any SU in adolescents to be problematic,¹⁰  others argue that some experimentation with alcohol and cannabis is normative in adolescence.⁵⁸  Our goal was to estimate treatment effects in participants with use that has been variously described as “risky,” “unhealthy,”⁵⁹  “hazardous,” or “harmful”¹⁶  but not yet meeting criteria for diagnosis of a SUD. Therefore, in consultation with our subject matter experts and a panel of experts in the field, we developed pragmatic inclusion criteria for problematic SU, when evaluating individual studies.

The included studies varied with regards to the primary substance of use, with researchers generally including users of alcohol, cannabis, combinations of specific drugs (eg, alcohol and cannabis), and/or unspecified substances in their studies. Some researchers reported outcomes for multiple substances but rarely explicitly identified when the intervention was targeted to a specific substance.

Finally, in the absence of information about the comparison interventions, we chose to categorize them as TAU. Consequently, the TAU category comprised a heterogeneous collection of often poorly described interventions that could be efficacious. This conservative approach may result in underestimates of treatment effects. No classification schema can be used to perfectly capture the complexity of possible interventions or be fully robust to the variable reporting in published reports. To support evidence synthesis, future researchers are encouraged to more clearly describe intervention components received by study participants, including those assigned to TAU. For example, rather than simply stating that an intervention was designed to “build motivation” or “build skills,” investigators should clearly explicate the underlying theoretical orientation and components of the intervention, to allow inferences as to whether the intervention was delivered with consistency and quality.

Because of the lack of consistently of reported outcomes across studies, the number of available comparisons was substantially reduced. The available evidence was consequently too sparse to allow a meaningful examination of factors that modify treatment effect, such as identification of key ingredients of successful interventions or examination of how intervention effects differ across demographic groups and participant factors, such as SU severity, substance of primary use, and cooccurring diagnoses. In future RCTs, researchers are encouraged to both use a consistent set of outcome measures and specify and evaluate putative mediators and moderators of treatment effect.

In the current review, we address the limitations of previous systematic reviews in several ways. We examine the specific effects of BIs on alcohol and cannabis outcomes, rather than reporting aggregate effects reflecting use of multiple substances (eg, illicit drug use). When possible, we report intervention effects in “natural units” (eg, days of use) instead of standardized mean differences to facilitate direct interpretation.⁶⁰  Furthermore, we employ network meta-analysis, which allows comparisons of >2 interventions (ie, MI, psychoeducation, and TAU) in a single, coherent analysis using both direct and indirect evidence.⁶¹  Finally, the, in current analysis, we include a rigorous SoE assessment of the body of evidence that is based on RoB and other ratings.

Compared to TAU, MI for adolescents with problematic SU is used to reduce both heavy alcohol use and overall days of alcohol use. We use these findings to lend further support to calls for wider implementation¹⁰  of MI for adolescents with problematic alcohol use in primary care settings.

Unfortunately, we found no evidence that decreased cannabis use results from brief MI. Given the ubiquity of cannabis use in adolescents, there is a vital need for additional research to identify effective interventions targeting problematic cannabis use by adolescents in primary, urgent, and emergency care settings. We use our finding (albeit with low SoE) that MI may decrease problems related to the use of alcohol and/or cannabis, such as missed school or work, to support further investigation of interventions focused on harm reduction in adolescent substance users.

We thank the members of the Key Informant and Technical Expert Panels who helped us refine the topic and develop the protocol, along with our task order officer Suchitra Iyer, PhD, at the AHRQ, Rockville, Maryland. We thank Jonah Popp, PhD; Iman Saeed, ScM; Bryant Smith, MPH, CPH; and Kelli Scott, PhD, for their careful data abstraction and verification. And we also thank our consultants Sarah Bagley, MD, MSc; Anthony Spirito, PhD; and Catherine Friedman, MD.

AHRQ

Agency for Healthcare Research and Quality

BI

brief intervention

CrI

credible interval

MI

motivational interviewing

NMD

net mean difference

OR

odds ratio

RCT

randomized controlled trial

RoB

risk of bias

SNMD

standardized net mean difference

SoE

strength of evidence

SU

substance use

SUD

substance use disorder

TAU

treatment as usual