Background

Screening, Brief Intervention, and Referral to Treatment (SBIRT) is a comprehensive, integrated public health approach to the delivery of early intervention and treatment services for individuals experiencing substance use-related harms, as well as those who are at risk of experiencing such harms [1], who are not seeking, or unlikely to seek treatment. The SBIRT model is based on public health principles and procedures, and is designed to reduce the burden of injury, disease, and disability associated with the nonmedical use of psychoactive substances.

The protocol typically begins with a screening procedure that involves asking questions to evaluate whether the individual has experienced, or is at-risk of experiencing, substance use-related harms. Brief interventions (BIs) are typically delivered to those individuals at low to moderate risk of harms. Individuals identified as experiencing significant harm and/or having more serious signs of substance dependence warranting formal diagnosis may be referred to treatment services that are outside the scope of BIs.

Our interest in evaluating BIs was with respect to nonmedical use of psychoactive substances excluding alcohol, caffeine, or nicotine. For the purpose of this review, nonmedical psychoactive substances are drugs prohibited by international law [2], which include, but are not limited to: amphetamine-type stimulants, cannabis, cocaine, heroin, and MDMA (3,4-methylenedioxymethamphetamine); the nonmedical use of pharmaceuticals such as benzodiazepines, opioids, or dextromethorphan; and the use of substances such as solvents or inhalants (for example, gasoline, acetone, etcetera) when they are used for their intoxicating effects.

Screening

SBIRT is intended to be implemented in healthcare settings or other community service settings frequented by the general population. In order to determine the likelihood that an individual is experiencing, or is at-risk of experiencing substance use-related harms, screening needs to be universal and opportunistic. By this, we mean that individuals are screened upon entering a program or organization (for example, hospital, primary care clinic, prison, or school program) as part of a standard intake procedure or process.

Screening may be conducted in a number of different ways. For example, intake staff may use psychometrically validated questionnaires or instruments that have been developed to accurately categorize users into low, moderate, or high risk categories. Psychometrically validated instruments have been developed for some types of substances, such as alcohol (Alcohol Use Disorders Identification Test, AUDIT [3]) or cannabis (the Cannabis Use Disorders Identification Test, CUDIT [4]). General drug screening instruments also exists (for example, Drug Abuse Screening Test, DAST [5]). However, screening instruments that reliably categorize users of other substances into low or moderate risk groups have not been developed (for example, heroin and cocaine). For those substances, screening may simply take the form of self-reported use or biological markers indicating use (for example, hair, urine, oral fluid, or blood). In the absence of validated instruments or biological markers, others may rely on even less rigorous screening methods such as the subjective judgment of the individual conducting the assessment. Regardless of the screening method employed, those deemed at risk of harms are typically provided a BI or referred to treatment.

Brief interventions

In addition to the variability in screening procedures used, there is much variation in how BIs are defined and delivered. In general, BIs are in-person, time-limited efforts to provide information or advice, increase motivation to avoid substance use, or to teach behavior change skills with the aim of reducing substance use and the likelihood of experiencing negative consequences. This variation includes the number of conversations or meetings that take place during intervention delivery, as well as the amount of time spent conducting the BI. The systematic review conducted by Kaner et al. [6] defined ‘brief’ to mean four or fewer sessions and, in the context of BIs for alcohol in primary healthcare settings, are typically delivered within the normal consultation period of 5 to 30 minutes. In a review of interventions targeting alcohol, Bien et al. [7] suggested that successful BIs typically focus on the following elements, collectively referred to using the acronym FRAMES: Feedback on behavior and consequences, Responsibility to change, Advice, Menu of options to bring about change, Empathy, and Self-efficacy for change.

There is substantial scientific evidence of the benefits of the SBIRT model in primary health-care settings as a means of preventing and/or reducing the serious long-term harms associated with excessive alcohol use [8–10]. There is also accumulating evidence suggesting that BIs may be effective in reducing the nonmedical use of psychoactive substances, such as cannabis [11–15], ecstasy [16], cocaine [12, 17, 18], benzodiazepines [19], and opioids [3, 17, 20] among both youth and adults.

Although systematic reviews assessing the efficacy of BIs in reducing harms associated with risky alcohol have been conducted [6, 21], there have been no published systematic reviews or meta-analyses that assess the effectiveness of BIs, among opportunistically screened populations, as part of the SBIRT model in reducing illicit drug use [22].

Our objective was to determine the effectiveness of BIs as part of the SBIRT model, compared with no BI or provision of information only, for reducing the nonmedical use of psychoactive substances among opportunistically screened populations identified as being at risk of harms and further, to determine whether any factors moderate the effect, using randomized evidence.

Methods

We published our methods as a protocol before conducting the review [23] and registered the review with PROSPERO (Registration number CRD42012002414 [http://www.crd.york.ac.uk/PROSPERO/display_record.asp?ID=CRD42012002414]). This review is reported according to the PRISMA statement [24] and was conducted according to AMSTAR tool items for additional quality control [25] [see Additional file 1].

Search strategy

We searched the following electronic databases: Ovid MEDLINE™ In-Process & Other Non-indexed Citations and Ovid MEDLINE™ (1946 to April 2012), Embase Classic + Embase (1947 to 6 April 2012), The Cochrane Library (searched 8 April 2012), Cumulative Index to Nursing and Allied Health Literature (CINAHL™) (searched 18 April 2012), PsycINFO™ (1806 to week 1 April 2012), Education Resources Information Center (ERIC) (searched 13 May 2012) and the CORK Database (searched 28 May 2012). All electronic search strategies were peer reviewed using the PRESS tool prior to implementation [26]; search strategies are presented in Additional file 2. We did not restrict the searches based on language, year of publication, or publication status.

For gray literature sources, numerous websites of relevant organizations, including those listed in ‘Grey Matters: a practical tool for evidence-based searching’ [27], were searched between 16 May and 22 May 2012 and are listed in Additional file 2. We scanned bibliographies of included articles and relevant systematic reviews. We searched clinicaltrials.gov and the WHO International Clinical Trials Registry Platform for ongoing studies.

Selection criteria and process

We selected studies according to the following criteria:

Inclusion criteria:

  • Study written in English or French.

  • Randomized controlled trials (RCTs) or cluster RCTs.

  • BIs administered to adolescents (12 to 18 years of age or equivalent by level of schooling), young adults (19 to 24 years of age), or adults (25 years and older) screened at risk of harms related to psychoactive substance use.

  • Participants were identified by opportunistic screening regardless of setting (that is, the participants in the study were from a screen-detected population and not a population seeking treatment for substance abuse). We included studies with any screening procedure.

  • Intervention was four sessions or less, included at least one of the FRAMES elements, and was delivered as a one-to-one verbal intervention to the individual.

  • Intervention was compared with no/delayed intervention or provision of information only.

Exclusion criteria:

  • Studies assessing alcohol, nicotine, or caffeine only.

  • Group interventions or text-only online interventions.

  • Studies addressing the effectiveness of the referral to treatment component of the SBIRT model only.

We uploaded the literature search results to systematic review software (Distiller SR©) for the study selection process. Our search was limited to systematic reviews published after 2009 because a recently conducted scoping review [22] did not locate earlier reviews on this topic. For all levels of study selection, we developed and pilot tested screening questions [see Additional file 3]. All titles and abstracts of records at Level 1 were screened once; those deemed not relevant were verified by a second person. Full-text reports of potentially relevant studies were assessed at Levels 2 and 3 by two independent reviewers; more than one level was used due to the complexity of applying the selection criteria to this literature. Disagreements were resolved by consensus or by a third reviewer. One reviewer tracked author responses regarding eligibility and identified multiple (companion) reports of the same study at Level 4.

Data extraction and process

We extracted study and publication details, study design characteristics, inclusion and exclusion criteria, participant characteristics, details regarding the screening methods and personnel, details regarding the intervention and comparison groups and personnel, outcomes, and other additional information from included studies [see Additional file 4].

We identified the following as outcomes of interest:

Primary outcomes

  • Substance use

  • Frequency of use

  • Quantity of use

  • Use-related harms or negative consequences of use

  • Changes in behavior likely to result in the reduction of negative substance use-related consequences (positive behavior change)

  • Decision to attend treatment

Secondary outcome s

  • Use of different substances (including alcohol, caffeine, nicotine) from that for which the client received the intervention

  • Intention to reduce substance use

  • Other health measures

Adverse outcomes

  • Any other reported adverse outcomes

These outcomes were extracted regardless of study follow-up time. We performed data calculations where needed (for example, change from baseline). For change-from-baseline calculations, we assumed a correlation coefficient of r = 0.25.

We developed and reviewed a data collection form in Distiller SR. One team member extracted information, and a second person verified all information.

Risk of bias assessment

All RCTs were assessed using the Cochrane Risk of Bias tool (RoB tool) [28]. Other sources of potential bias that were assessed included fidelity (performance bias), recruitment bias for cluster trials [29], single versus multicenter studies [30], and study sponsorship bias. Some bias items were assessed at the study level (for example, randomization), while others were assessed at the outcome level (for example, selective reporting). We contacted corresponding authors of included studies regarding their consent procedures to inform the assessment of participant blinding. We assessed each study for the risk of bias for a given outcome and then determined a summary assessment across all studies for that outcome. Summary assessments were categorized into low, medium, and high risk of bias and incorporated into grading the quality of evidence.

One team member extracted risk of bias information and a second person verified all information.

Evidence synthesis

Study characteristics were summarized narratively in the text and presented in tables. In order to assess whether meta-analyses of the data were possible, we assessed the quantity and methodological and clinical homogeneity of studies. We conducted narrative syntheses as meta-analysis was either not appropriate or possible. Where possible, we calculated and presented dichotomous outcomes as risk ratios (RR) and continuous outcomes as mean differences (MD), both with 95% confidence intervals. We contacted corresponding authors regarding inadequately reported data. Other analyses, such as subgroup analyses and funnel plot assessment, were pre-planned but not carried out due to few included studies.

The quality of evidence for all outcomes was evaluated using the GRADE methodology [31]. The quality of evidence was assessed across the domains of risk of bias, consistency, directness, precision, and publication bias. Each outcome was given a final adjudication of high, moderate, low, or very low.

Protocol modifications

The pre-planned outcome ‘any standard/accepted biological markers of substance use’ represents a method for measuring use and is captured within other listed outcomes. Similarly, the pre-planned adverse outcome ‘self- or other reported use or increased use of different substances’ is captured as a secondary outcome. We included ‘composite’ outcomes where reported in studies as they captured relevant measures (for example, the Alcohol, Smoking and Substance Involvement Screening Test (ASSIST)). For feasibility we used a verification process for data extraction and risk of bias assessment rather than dual extraction. When grading the evidence, we felt ‘medium’ risk was more representative than ‘unclear’ risk for interpreting the risk of bias assessment across domains and studies for an outcome [32].

Results

We located 9,631 bibliographic and 17 gray literature records. A flow diagram of the study selection process is shown in Additional file 5. Of the 8,836 records that remained after removing duplicates, 7,940 were excluded during title and abstract screening. Of the 896 full-text reports reviewed, 886 reports were excluded during two rounds of full text screening. Over 50% (n = 454) of the studies at this stage were excluded because the population was not opportunistically screened. A further 35% (n = 307) of the studies were excluded because they did not meet our study design criteria (that is, RCT, or cluster RCT). Remaining studies were excluded for a variety of reasons that occurred with a frequency of 5% or less, including full-text report was unavailable, intervention did not meet the definition of BI, etcetera. Of the ten remaining potentially eligible reports, eight reports detailing the results of five unique RCTs were included. Sixteen ongoing or completed trials were located [see Additional file 6]. A list of excluded studies is reported in Additional file 7.

Other study designs

Given only five RCTs were located, we went back to determine if any excluded reports that employed other experimental designs (for example, non-RCTs, controlled before-after (CBA) or interrupted time series (ITS)) were relevant. Our search revealed 38 studies, but none of those studies met remaining selection criteria. These searches were mainly targeting RCT and non-RCT designs, for feasibility of resources; however, given the few studies meeting criteria of the 8,836 records screened, we felt it unlikely that additional studies would have been missed by our search.

General characteristics of included trials

Of the five included studies, three were single-site RCTs [17, 33, 34], one was a multisite RCT [12, 35, 36], and one was a multisite, cluster RCT [37, 38]. The single-site studies were conducted in the United States [17, 34] and the cluster trial in Germany [37, 38]. The other multisite study was conducted in Australia, Brazil, India, and the United States [12, 35, 36]. All studies were published after 2005 (Table 1). For the remainder of the review only the main report for each included study will be cited.

Table 1 Study and participant descriptive characteristics of included studies
Full size table

Participants and setting

One of the included studies assessed the effectiveness of BIs in homeless youth (13 to 18 years old) [33], one in youth and young adults (14 to 21 years old) [34], one in young adults and adults (16 to 62) [35], and two in adults only [17, 37]. Four studies took place in a healthcare setting (primary care or hospital) [17, 34, 35, 37] while one took place at a drop-in center [33].

Screening

There was considerable diversity in the screening instruments employed in the included studies (Table 2). Only one study [35] used a screening instrument whose psychometric properties have been well established. The other studies either used screening instruments of unknown/questionable validity or adapted versions of instruments whose psychometric properties had been published. Zahradnik et al. [37] used a combination of two screening instruments followed with a diagnostic interview to determine who would receive the intervention [37], while Bernstein et al. [17] employed a group of unspecified ‘substance abuse screening questions’ in conjunction with an instrument whose validity and reliability had been established for use in a clinical or research setting only. The two remaining studies used investigator developed, self-report screening instruments whose psychometric properties are unknown [33, 34].

Table 2 Description of screening procedure and instruments of included trials
Full size table

Humeniuk et al. [35] was the only study to provide BIs to those participants screened at a moderate risk level only and then refer to treatment those screened at high risk (as determined by the ASSIST). All other included studies assigned participants to control or intervention groups if they scored more than a specific threshold but did not specify any upper level threshold.

Brief intervention

There was also a considerable degree of heterogeneity in the characteristics of the BIs (see Table 3 for details regarding the BIs and how they were administered). Bernstein et al. [34] screened for and administered an intervention targeting a single, specific substance (cannabis). Bernstein et al. [17], Baer et al. [33], and Zahradnik et al. [37] screened for and administered interventions targeting a set of or a group of drugs: cocaine and/or heroin [17]; alcohol, cannabis, and other drugs [33]; or prescription drugs [37], respectively. Humeniuk et al. [35] screened for multiple drugs then targeted the BI at the substance that screening indicated was most problematic or the substance of most concern to the participant [35].

Table 3 Characteristics of brief interventions (BIs) and control groups
Full size table

In addition to targeting different substances, interventions comprised diversity in the number and length of sessions. Humeniuk et al. [35] simply assessed the effect of a single verbal intervention and accompanying written information [35]. The BIs in Bernstein et al. [17] and Bernstein et al. [34] consisted of an initial verbal intervention, take-home written information, and a follow-up telephone call. Among the last two trials, Baer et al. [33] consisted of four sessions, and Zahradnik et al. [37] consisted of two sessions with a feedback letter mailed eight weeks later. Despite these variations, there was consistency in the treatment approach that informed the BIs in that all BIs either explicitly or implicitly incorporated a motivational interviewing approach.

All trials reported using techniques/strategies to ensure adherence to the planned BI. Though it was implied by all the studies that these strategies ensured that all interventions were administered as planned, only one of the included studies reported this explicitly [34].

Comparison groups

All three comparison groups of interest were encountered among included studies (Table 3). Three studies provided participants with written information about the risks of drug use [34, 37] or list of local treatment options [17]. For the remaining two studies (no BI) Baer et al. [33] provided the care or service sought by the individual, and Humeniuk et al. [35] provided the care or service plus delayed intervention.

Risk of bias assessments

Additional file 8 outlines the risk of bias assessments by domain for included studies. Supports for judgments are provided in Additional file 9. Overall, studies were deemed at medium or high risk of bias for outcomes.

Most studies reported an adequately randomized method, while two studies reported using a concealed method to implement randomization [17, 35]. We assessed ‘blinding of participants and personnel’ across all outcomes as any systematic changes would have affected all outcomes. In all studies, it was not possible to blind personnel to what they were delivering to participants. In two studies, participants were aware of the study intent and what groups they could be allocated to; this information was unclear or likely to have occurred in remaining studies. Outcomes collected through objective means (biochemical analysis or use of database records) are at low-risk for assessor bias. Loss-to-follow-up (total amount or differential amount between groups), handling of missing data, and unknown/unreported reasons were important issues regarding attrition, and almost all studies were at unclear or high risk of bias. Studies were also at unclear or high risk for selective reporting bias. One study was at low risk of performance bias regarding fidelity of the intervention (other bias), while remaining studies were at unclear or high risk. The one cluster trial was at unclear risk of recruitment bias. Study sponsorship bias was not an issue in these studies.

Effects of brief interventions

The effects of BIs in the included studies are described and analyzed based on comparison group: BI versus no intervention and BI versus written information only. Included in the former group is the Humeniuk et al. [35] study in which the delayed intervention was administered after outcome data were collected at follow-up. Results of included studies are presented in Tables 4 and 5. We were unable to fore-analyze unit of analysis errors in studies due to insufficient information, so we report the point estimate without confidence intervals.

Table 4 Evidence table for brief intervention (BI) versus no BI in participants screened for at-risk substance use
Full size table
Table 5 Evidence table for brief intervention (BI) versus written information in participants screened for at-risk substance use
Full size table

Brief intervention versus no intervention

Two studies evaluated the BI with no or delayed intervention. Baer et al. [33] assessed a four-session BI targeting alcohol, cannabis, and other drug use among adolescents at a faith-based drop-in center. Humeniuk et al. [35] assessed a single-session intervention plus written materials targeting cannabis, cocaine, amphetamine-type stimulants, or opioids among those 16 years and older at healthcare settings across four countries.

Few outcomes were reported and by only one study each. Groups were not statistically significant for change in frequency of use measures [33], a composite score measure [35], and use of drop-in or other agency services from baseline to one and/or three months of follow-up [33] (Table 4). In Humeniuk et al. [35] general health outcomes were only assessed among participants in the intervention group and are therefore not included in Table 4. Remaining outcomes of interest were not assessed in either study.

Brief intervention versus written information only

Three studies evaluated the BI compared with the provision of written information only about the risks of drug use or local treatment services/options. Bernstein et al. [34] assessed one session plus written information and a telephone call targeting adolescents and young adults cannabis use at a pediatric emergency hospital department. Zahradnik et al. [37] assessed one session plus telephone call and a feedback letter targeting those 18 and older who regularly use prescription drugs with addiction potential or met DSM-IV criteria for dependence in a hospital setting. Finally, Bernstein et al. [17] assessed one session plus written information and a telephone call targeting those 18 years and older reporting heroin or cocaine use.

Most outcomes of interest were reported on but mainly by one study each. Some outcomes were assessed with multiple measures. Where authors reported two or more substances together, data for the individual substances are provided in Additional file 10. A few studies addressed abstinence as a measure of substance use for differing substances and with varied follow-up times: Two studies were not statistically significant at 3-months [34, 37], one study not significant at 6 months [17], and two studies reported mixed results at 12-months [34, 37] (Table 5). With few studies and important clinical and methodological heterogeneity between those studies, we did not meta-analyze. Remaining outcome measures (other substance use measures, frequency of use, quantity of use, use-related harms, positive behavior change, other health measures) were mostly not statistically significant [17, 34, 37] (Table 5, Additional file 10). A few measures (quantity of use, decision to attend treatment, and use of different substances) were poorly reported (Table 5). ‘Intention to reduce use’ was not reported in any study.

Discussion

Few studies have assessed the effectiveness of BIs among opportunistically screened populations as part of the SBIRT model for reducing the nonmedical use of psychoactive substances. When comparing the intervention with no intervention or to written information only, most outcomes were not statistically significant. However, the overall quality of the evidence per outcome is limited or very uncertain [see Additional file 11]. Due to the few included studies, results are imprecise and largely could not be assessed for consistency. The literature has important methodologic limitations leading to medium or high risks of bias for outcomes. The body of evidence, therefore, is limited given the few included studies with mainly small sample sizes and the heterogeneity in study characteristics, including the measurement of outcomes.

Practice implications

Insufficient evidence exists to make conclusions as to whether BIs are effective or ineffective at reducing the use of or harms associated with the nonmedical use of psychoactive substances other than alcohol, nicotine, or caffeine when these interventions are administered to nontreatment-seeking, screen-detected populations.

Research implications

We are aware of 16 ongoing studies, some with large sample sizes, which are potentially relevant to this review. Since the current evidence base is inconclusive, updating this review when the results of the ongoing studies become available will be important.

Given the variation observed among characteristics of the included studies, future research in the area would benefit from modifications in scope and study design. Firstly, we propose a focus on primary care settings, to supplement the evidence base of four of our included studies, before evaluating other community service settings. Secondly, a standard, validated screening instrument with an acceptable sensitivity and specificity profile is an important first step in determining the effectiveness of the SBIRT model among nontreatment-seeking populations. The instrument should be designed to take a relatively short time to complete, as a pragmatic consideration for implementing the SBIRT model. Thirdly, interventions that are more clearly reported and with sufficient detail to be replicated in other trials would also help develop this body of evidence. Fidelity of the delivered intervention should be collected and reported on. Finally, agreement on a core set of defined outcomes, their measurements, and lengths of follow-up will be essential to ensuring relevance to practice and to allow meta-analysis of studies.

There are additional items that researchers could implement to increase internal validity. The consent process should be designed such that participants are unaware of the intent of the study or to the groups for which allocation is possible. Researchers should consider using a sham intervention as a control group that would address another aspect of wellbeing (for example, nutrition) to help blind participants to group allocation. In addition, if known, researchers should indicate reasons for client drop-out.

Incomplete reporting (for example, informed consent procedures and intervention details) was a general barrier in attempting to assess studies against inclusion criteria as well as to assess for risks of bias. Davidson et al. [59] provides detailed reporting guidance.

Future research incorporating these modifications will enable meaningful statements on SBIRT effectiveness.

Consult the protocol modifications section for possible limitations; our work has been conducted according to AMSTAR standards.

Conclusions

Insufficient evidence exists as to whether BIs as part of SBIRT are effective or ineffective at reducing the use of or harms associated with the use of nonmedical use of psychoactive substances other than alcohol, nicotine, or caffeine when these interventions are administered to nontreatment-seeking, screen-detected populations. Given the evidence base is inconclusive, emerging evidence from existing ongoing studies may help to stabilize conclusions about the effectiveness of BI.