Illicit and prescription unhealthy drug use is associated with significant global morbidity, mortality, and economic and social costs.1 Unhealthy drug use is any hazardous use of non-alcohol substances beyond legal and medical guidelines,2 which may signify a substance use disorder (SUD).1 Globally, about 5% of adults use illicit drugs and 0.4% fit criteria for an SUD.1 However, only about 10% fitting an SUD criteria actually receive treatment.3,4 This 90% disparity is known as the treatment gap. The treatment gap led to the development of screening and brief intervention (SBI) programs in outpatient medical settings to identify and treat or provide referrals to non-treatment-seeking patients. SBI consists of a brief individual session lasting about 25 min, administered before or after patient visits, and is indicated for low to moderate unhealthy use.5,6

Many outpatient medical care facilities across the globe have dedicated resources7,8,9 to implementing SBI for early detection and early intervention.5,6 However, delivery of SBI in outpatient medical care is predicated on the efficacy of the brief intervention (BI) component. Findings from BI trials have been mixed, with supportive findings for unhealthy alcohol use,10,11,12,13 but currently lack evidence for unhealthy drug use.11,14,15,16 The US Preventive Services Task Force found some evidence supporting BI for unhealthy drug use in any setting.17 However, much of the evidence came from treatment-seeking populations, meaning the efficacy of BI in outpatient medical care settings remains unknown.17 Moreover, BI’s efficacy may vary in different outpatient care settings. Thus, a systematic review and meta-analysis are necessary to determine BI efficacy for unhealthy drug use in outpatient medical care.11,12,13,14,18,19,20,21

The present research determined (1) the efficacy of BI for unhealthy drug use, (2) expected length of effects, and (3) subgroup findings by setting. Findings supported continued BI delivery or re-evaluation of patient needs for unhealthy drug use in specific outpatient medical settings. Findings may help determine care delivery and resource allocation.


The present methods are concise, as a more detailed methods protocol has been previously published.22 The methods reported in this systematic review and meta-analysis are aligned with the preferred reporting items for systematic reviews and meta-analyses guidelines for systematic review protocols (PRISMA-P)23 and is registered in PROSPERO (CRD42020157733).

Criteria for Considering Studies

Studies were limited to RCTs and cluster randomized trials (CRTs) screening for unhealthy drug use and comparing BI with usual care (UC). The population includes people presenting in outpatient medical care facilities for physical health concerns, who are screened for, and report unhealthy drug use. Unhealthy alcohol use is not included. We included BI studies ranging from 1 to 5 sessions, with each session lasting from 5 to 60 min either in-person, by phone, mailed, or web-based. UC controls could include minimal screening/assessment, sham interventions, information pamphlets about unhealthy drug use, or advice below the BI threshold. Following previous study methods, outpatient medical care was meant to capture ambulatory outpatient care — emergency and primary care. Inpatient and specialty clinics were excluded. Outpatient medical care is defined as immediately accessible care for broad health concerns, including but not limited to hospital/community primary care clinics, emergency departments, family practice, and women’s health clinics.6,10


Primary outcomes were (a) endpoint drug use frequency and (b) endpoint drug use severity at intermediate follow-up (4–8 months). Secondary outcomes included frequency and severity at short-term (< 4 months) and long-term (> 8 months) follow-up. Frequency was measured as the number of days in a specified timeframe the primary problem substance was used. Severity was measured as a composite score which accounts for use frequency, craving, withdrawal, and consequences in physical, psychological, social, and vocational domains.

Search Strategy

We conducted a systematic search in CENTRAL, EMBASE, MEDLINE, and PsycINFO until January 13, 2021 (eTable 1 in the Appendix). Searches were supplemented with trial registries (e.g., and through relevant existing article reference lists.

Study Selection and Data Extraction

Two reviewers independently screened, selected (moderate interrater reliability, κ = 0.66), and extracted data from studies and came to a consensus. Papers determined to be related to primary studies were collated and reported as secondary studies.

Assessment of Risk of Bias in Included Studies

Two review authors independently assessed the risk of bias for each study using the Cochrane risk of bias tool version 2 (RoB2).24 RoB2 findings were entered into the GRADE calculation to aid in certainty of evidence determination, rated as very low, low, moderate, or high. GRADE accounts for risk of bias, consistency across studies, directness in answering the research question, precision of effects, publication bias, and effect size.

Data Synthesis

Data were analyzed in Review Manager 5 software (RevMan5). All included studies provided count or continuous data at variable assessment timeframes that were assessed with the standardized mean difference (SMD) and a 95% confidence interval (CI). The SMD is interpreted as small effect (0.2), moderate effect (0.5), and large effect (0.8).25 Studies with more than two relevant BI treatment arms were combined.24,26 We prioritized intention to treat (ITT) data for all outcomes, either mixed-effect models or last observation carried forward (LOCF). However, LOCF data was excluded in a sensitivity analysis. All analyses were conducted using random-effects pairwise meta-analyses.24,27 Statistical heterogeneity was assessed by visually inspecting CIs for individual trials visually represented in a forest plot and with the I2 and τ2 statistics.28

Sensitivity analyses were conducted with primary analyses to assess the effects of excluding studies with controls receiving more than UC, studies using LOCF, and studies with a high risk of bias. Prespecified subgroup analyses were conducted to examine differences between subsets of the population of interest in patient age groups, baseline severity groups, and primary drug treated. Because unanticipated diversity in population settings emerged, a post hoc subgroup analysis compared different outpatient settings. This study used published aggregate data with no informed consent, and the Kyoto University institutional review board exempted this study from review.


Figure 1 shows the PRISMA search strategy and study inclusion. In all, 27 primary studies were included and described in the narrative review.27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53 eTable 2 in the Appendix describes individual study characteristics. All studies were individually randomized except for 1, which was cluster randomized among 195 general practitioners.48 There were 14 single-site studies29,30,31,32,33,34,41,49,50,51,52,53,54,55 and 13 multi-site studies.35,36,37,38,39,40,42,43,44,45,46,47,48

Figure 1.
figure 1

PRISMA flowchart of study inclusion for systematic review and meta-analysis. Search strategy includes initial screening to final inclusion of studies in meta-analysis.

Study Populations

The majority (21 studies) were conducted in the USA.29,30,31,32,33,35,37,38,40,41,42,44,46,47,49,50,51,52,53,54,55 Additionally, 2 were conducted in South Africa,34,45 1 in Thailand,36 1 in Chile,43 and 1 in France.48 There was 1 multi-national study conducted in Australia, the USA, India, and Brazil.39 Most populations (12 studies) comprised adults aged 18 and older.32,35,37,40,41,44,45,46,50,51,54,55 There were 5 studies that excluded older adults with cutoff ages ranging from 55 to 64 years old.30,33,38,43,49 Additionally, 1 study included graduate-school-aged adults only.52 There were 2 studies that included emerging adult samples aged 18–2434 and college-aged adults.53 The inclusion of adolescents varied with 7 studies including adolescents.29,31,36,42,47,48 Other subsamples included 2 studies of women only,47,49 2 studies of students,52,53 and 1 study of HIV-positive patients.35

Outpatient Medical Care Facilities

Most (10 studies) were conducted in emergency departments,29,30,33,38,41,45,46,49,50,55 5 in hospital primary care centers,36,37,44,51,54 and 5 in community health clinics.31,34,39,42,47 Other settings included student health centers,52,53 a general practice,48 and a women’s health clinic.40 Additionally, 1 study combined emergency and women’s health populations,32 1 study combined emergency departments with police stations,43 and another was in an HIV primary care clinic.35

Unhealthy Drug Use Screening

For primary problem substance, 14 studies accepted any non-alcohol-related drugs,30,34,35,36,37,38,40,41,44,45,46,49,51,54 7 studies included cannabis only,29,31,42,47,48,52,53 3 studies limited a mix of specific drugs,32,43,55 and 2 studies were opioid specific.33,50 Most (10 studies) used the ASSIST to screen for unhealthy drug use. Other screening methods included the DAST-10,32,41,44 CRAFFT,31,42,47 a single question about any unhealthy drug use,34,35,46 and less common substance or language-specific instruments.29,33,48,49,50,55 In addition, 2 studies used a single question for inclusion and the ASSIST to exclude severe use.52,53

Brief Intervention Experimental Conditions

Most studies used motivational interviewing.29,31,32,33,34,35,37,38,39,40,41,42,43,45,46,47,48 There were 4 studies that used motivational feedback,36,49 and web-based motivational feedback.30,40,52,53 Other modalities included motivational enhancement,44,55 brief advice,51 and brief negotiated interviews.50 Some used BI combinations in separate arms.30,54 Some (12 studies) added booster sessions by telephone consultation/counseling, though booster attendance was generally low.29,30,32,35,38,41,44,49,51,52,54,55 BI was administered by a variety of professionals, from physicians and psychologists to interventionists and research assistants of varying education levels.

Control Conditions

Only 6 studies used UC controls with zero intervention components, plus screening and assessment for research purposes.31,38,39,47,48,52 Next, 5 studies also provided UC, but gave attention-control sham interventions focused on non-drug-related health issues.37,46,49,51,53 Alternatively, 13 studies had a UC control group that also received unhealthy drug use information pamphlets listing associated health risks and referral lists.29,32,33,34,40,41,42,43,44,45,50,55 Finally, 3 studies were considered more active controls but did not meet BI criteria, with 1 study providing simple advice,36 1 study using follow-up control boosters,30 and 1 study providing brief SUD education.35

Baseline Frequency and Severity

All 27 studies provided baseline data for the outcomes reporting no differences between treatment and control baseline means. There were 8 studies that restricted severity to low and moderate levels by imposing a lower and upper limit to exclude severe cases.36,37,39,42,43,51,52,53

Meta-analytic Findings

Of the 27 studies included in the narrative review, 7 studies were excluded from the meta-analysis due to incompatible outcomes (e.g., abstinence, time to overdose)32,33,47,48 or mixed drug and alcohol outcomes.31,34,45 For the meta-analysis, 20 studies with 9182 randomized patients were included at three timepoints.29,30,35,36,37,38,39,40,41,42,43,44,46,49,50,51,52,53,54,55 eFigures 1 and 2 in the Appendix show all included studies and timepoints for each outcome. RoB2 results are displayed in eTable 3 in the Appendix and calculated in the GRADE certainty of evidence in eTable 4.

Primary Outcomes — BI Efficacy for Unhealthy Drug Use in Outpatient Medical Care at Intermediate Follow-up (4–8 Months)

Intermediate Follow-up Frequency

Figure 2 shows meta-analytic findings for the primary outcomes. With 3378 patients from 9 studies, there was insufficient evidence to support efficacy of BI for drug use frequency at intermediate follow-up (SMD = −0.07, 95% CI = −0.17, 0.02, p = 0.12, I2 = 37%). These studies demonstrated moderate heterogeneity. The certainty of evidence was rated high.

Figure 2.
figure 2

Drug use frequency and severity primary outcomes at intermediate follow-up (4–8 months). SD, standard deviation; SMD, standardized mean difference; CI, confidence interval; df, degrees of freedom; BI, brief intervention; UC, usual care; ASSIST, the alcohol, smoking and substance involvement screening test

Intermediate Follow-up Severity

With 3312 patients from 7 studies, there was insufficient evidence to support efficacy of BI in reduction of drug use severity at intermediate follow-up (SMD = −0.27, 95% CI = −0.78, 0.24, p = 0.30, I2 = 98%). Severity demonstrated extremely high heterogeneity across studies, which appears to be due the highly divergent findings from Woodruff.46 The certainty of evidence was rated low and downgraded by two levels due to very serious inconsistency of effects between studies.

Sensitivity Analyses

All sensitivity analyses found equivalent results to primary analyses. Excluding studies with more than UC controls (i.e., information pamphlet, advice, booster, education below the BI threshold) revealed insufficient evidence for intermediate follow-up frequency (SMD = −0.21, 95% CI = −0.51, 0.09, p = 0.17, I2 = 0%) and severity (SMD = −0.64, 95% CI = −1.65, 0.37, p = 0.22, I2 = 98%). Excluding studies using last observation carried forward imputation methods for the primary outcomes still found insufficient evidence for both frequency (SMD = −0.08, 95% CI = −0.18, 0.02, p = 0.08, I2 = 45%) and severity (SMD = −0.36, 95% CI = −0.98, 0.26, p = 0.25, I2 = 99%). Finally, excluding studies with a high risk of bias in both the primary outcomes still showed insufficient evidence for both frequency (SMD = −0.08, 95% CI = −0.18, 0.02, p = 0.14, I2 = 45%) and severity (SMD = −0.35, 95% CI = −0.97, 0.26, p = 0.26, I2 = 99%).

Secondary Outcomes

Secondary outcomes represent drug use frequency and severity at short-term efficacy (<4 months) and long-term efficacy (9–12 months). There was insufficient evidence to support efficacy in all secondary outcomes at any timepoint.

Short- and Long-term Frequency

There was insufficient evidence to support efficacy of outpatient medical care BI in reducing drug use frequency at both short-term and long-term follow-ups (eFigure 1). The short-term frequency follow-up was the most comprehensive assessment with 4486 patients from 15 studies and moderate heterogeneity (SMD = −0.06, 95% CI = −0.14, 0.3, p = 0.17, I2 = 40%, low certainty of evidence). Long-term follow-up assessment also revealed non-significant findings among 2431 patients in 6 studies with substantial heterogeneity (SMD = −0.06, 95% CI = −0.19, 0.8, p = 0.4, I2 = 57%, moderate certainty of evidence).

Short- and Long-term Severity

There was insufficient evidence to support efficacy of outpatient medical care BI in reducing drug use severity at both short-term and long-term follow-ups (eFigure 2). Short-term severity used 3714 patients from 9 studies and moderate heterogeneity (SMD = −0.01, 95% CI = −0.10, 0.08, p = 0.84, I2 = 38%, high certainty of evidence). Long-term severity assessments revealed non-significant findings among 2142 patients in 3 studies with heterogeneity that may not be important (SMD = −0.01, 95% CI = −0.09, 0.08, p = 0.89, I2 = 0%, high certainty of evidence).

Subgroup Analyses

Comparing outpatient setting subgroups (Fig. 3) revealed significant differences between 5 subgroups in drug use frequency at intermediate follow-up (χ2[4] = 11.7, p = 0.02, I2 = 65.8%). The emergency department subgroup demonstrated a statistical difference favoring BI for lower drug use with small effects (SMD = −0.15, 95% CI = −0.25, −0.04, p < 0.01, n = 1397). Hospital and community primary care clinics (SMD = 0.07, 95% CI = −0.03, 0.18, p = 0.18, n = 1400) and student health (SMD = −0.21, 95% CI = −0.38, 0.09, p = 0.07, n = 168) subgroups failed to demonstrate significant treatment differences. Additionally, women’s health and HIV primary care subgroups each consisted of one study and the individual studies failed to show significance. There were no other subgroup differences between BI and UC in either drug use frequency or severity in the primary outcomes (eTable 5 in the Appendix).

Figure 3.
figure 3

Setting subgroup analysis for drug use frequency at intermediate follow-up. Note. SD, standard deviation; SMD, standardized mean difference; CI, confidence interval; df, degrees of freedom; BI, brief intervention; UC, usual care.


The present investigation aimed to determine (1) efficacy of BI for unhealthy drug use in outpatient medical care, (2) expected length of effects, and (3) subgroup findings by setting. Findings were taken from 20 studies with a total of 9182 randomized patients. The certainty of evidence ranged from low to high, depending on the outcome as rated through the GRADE approach. There is insufficient evidence to support the efficacy of BI for unhealthy drug use across all various outpatient medical care settings. No significant effects were found comparing BI to UC control. The lack of support includes both drug use frequency and severity. There is insufficient evidence to support efficacy at any timepoint: short-term, intermediate, or long-term follow-up. However, support was found favoring BI for lower drug use frequency in emergency department settings, but not for hospital/community primary care clinics, student health, women’s health, or HIV primary care settings. The result for BI in emergency departments was statistically significant, but the effect size was small (SMD = −0.15), and clinical meaningfulness is unknown. This estimate is based on smaller sample sizes with wide-ranging confidence intervals, leaving the true effect still debatable.

The present findings support a reassessment of SBI care delivery for unhealthy drug use.17 Findings are important for use of healthcare resources and for providing the best available evidence-based treatment for unhealthy drug use in medical settings. The call for more RCTs in BI for unhealthy drug use was made a decade ago.15,20 The present research represents the most comprehensive evaluation of outpatient medical care BI for unhealthy drug use to date. While there is a lack of evidence for all outpatient medical settings, clinical meaningfulness remains possible for emergency departments and more large trials are needed in these settings.

Despite insufficient evidence among all included studies, something may be working in emergency departments. It is important to reiterate that subgroup effects are small and clinical meaningfulness requires further investigation. Compare the efficacy of BI for reducing alcohol use in outpatient medical care (weighted MD = −38.4)6 with the present study’s findings in all outpatient settings (SMD = −0.17, nonsignificant) and emergency department subgroups (SMD = −0.15, significant). Unhealthy alcohol use BI is moderately efficacious, reducing alcohol ranging between 23 and 54 less grams per week at 1-year follow-up.6 For general outpatient settings, the present study found the estimated drug use could range from 2 fewer days to 0.2 more days per month compared to UC. While this comparison was nonsignificant, there still may be some benefits across all outpatient settings, given the resources. When comparing subgroups, an estimated difference in drug use could range from 0.5 to 3 fewer days per month compared to UC, but only in emergency departments. In primary care settings, the BI group demonstrated greater average drug use than UC (−0.3 to 2.1 days used/past month).

The present study followed previous methods in defining outpatient medical care in order to compare findings.6 Though not directly tested, we submit that there may be different motivational factors at play in emergency departments. BI is based on patient motivation, and different settings may be better adept at targeting effective motivational anchors. For instance, emergency departments may see a larger proportion of patients who present with high acuity and urgency due to drug use consequences, either primary (i.e., overdose) or secondary (i.e., falls, injuries). Such precipitating incidents may serve as more effective motivation to reduce drug use. There are a number of factors involved, but perhaps those presenting in other outpatient medical settings do not have readily identifiable reasons to change. Remember, these are patients who are not seeking SUD treatment. Re-evaluation and further investigation of BI for unhealthy drug use might improve with consideration of differential motivation factors associated with presenting physiological problems.

The present research adds a considerable amount of information to the literature based on BI. This is the first meta-analysis to specifically investigate BI for unhealthy drug use in outpatient medical care. While evidence of BI for reducing alcohol intake is established,6,10 the present study results show that BI for unhealthy drug use in hospital and community primary care, women’s health clinics, and HIV clinics still lacks evidence of efficacy. Facilities continuing to provide BI for unhealthy drug use as part of existing SBI protocols could consider alternative evidence-based treatments for unhealthy drug use and development of more in-house drug treatments serving more patients. Alternative to BI, facilities should continue providing screening and referrals to specialty SUD treatment for all who screen positive for unhealthy drug use in high severity levels. However, more attention would need to be given to referral follow-ups and case management because SBI program referrals lack efficacy for increasing SUD treatment utilization.56


This research has some limitations of note. We were unable to assess publication bias in the primary analyses because there were too few studies. However, publication bias was unlikely because most published research found null results. The use of UC with zero intervention is more of an ideal than a reality. Most trials included some small active component. While most were only information pamphlets and referral lists, this could have affected the results. However, we did our best to correct for this through a sensitivity analysis that excluded active SUD components. A meaningful portion of the risk of bias assessment was unknown. This was primarily due to a lack of protocol reporting. The actual bias associated with selective reporting of outcome measurements and analyses may be higher, making the certainty of evidence lower. When the information is unknown, this can erroneously benefit trialists. We did our best to be as accurate as possible despite a significant amount of unknown information. One limitation of note is in the lack of treatment fidelity and delivery investigation. Trials were delivered by a variety of providers ranging in education level. All studies used some form of motivational interviewing-based BI, but these were delivered for various and mixed problem substances. The present study focused on patient factors and settings. More investigation is needed for treatment factors of unhealthy drug use BI in outpatient medical care. Finally, trials for student health, women’s health, and HIV subpopulations consisted of small, low-powered studies. The evidence is not conclusive among these settings and more work is needed.


The present study found a lack of support for unhealthy drug use BI when all settings were included but found small significant effects in emergency departments. In a public health perspective, small and even nonsignificant effects may be clinically meaningful. If a substantial portion of the population receives BI, then small effects could translate to improved outcomes across other health domains. Alternatively, small effects may not translate to meaningful treatment gains in individual facilities. While there may be subpopulations with better outcomes, more work is needed to understand why it works for some and not for others. Once a patient is screened and unhealthy use is discovered, they must receive an evidence-based treatment. Presently, BI for unhealthy drug use lacks evidence as the best choice for patients not seeking SUD treatment when all outpatient medical care settings are evaluated together. BI does appear to be more efficacious than UC in emergency departments. However, the effects were small and clinical meaningfulness is yet to be determined. Future trials should prioritize studies in these populations with sufficient sample sizes.