The search strategy was designed by a health information specialist together with the research team and peer reviewed by another senior information specialist before execution using Peer Review of Electronic Search Strategies for systematic reviews [56]. The search strategy revolved around three concepts:

The strategy was initially implemented on March 18, 2020, and again on October 13, 2021, and February 13, 2023. The full, detailed search strategies for each database are presented in Multimedia Appendix 2.

We searched 7 electronic databases of published literature: CINAHL Complete, Cochrane Database of Systematic Reviews, Cochrane Central Register of Controlled Trials, Embase, MEDLINE, PubMed, and PsycINFO. No publication date filters or language restrictions were applied. A combination of free-text keywords and Medical Subject Headings was tailored to the conventions of each database for optimal electronic searching. The research team also manually screened the reference lists of the included articles and the bibliographies of existing systematic reviews [18,31,42-45] to identify additional relevant studies (snowballing). Finally, a forward citation tracking procedure (ie, searching for articles that cited the included studies) was carried out in Google Scholar.

The population, intervention, comparison, outcome, and study design process is presented in Multimedia Appendix 3. The inclusion criteria were as follows: (1) original research articles published in peer-reviewed journals; (2) use of an experimental study design (eg, RCT, cluster RCT, or pilot RCT); (3) studies evaluating the effectiveness (or efficacy) of digital interventions designed specifically to prevent, reduce, or cease CU as well as promote CU self-management or address cannabis-related harm and having CU as an outcome measure; (4) studies targeting young adults, including active and nonactive cannabis users; (5) cannabis users and nonusers not under substance use treatment used as controls in comparator, waitlist, or delayed-treatment groups offered another type of intervention (eg, pharmacotherapy or psychosocial) different from the one being investigated or participants assessed only for CU; and (6) quantitative CU outcomes (frequency and quantity) or cannabis abstinence. Given the availability of numerous CU screening and assessment tools with adequate psychometric properties and the absence of a gold standard in this regard [57], any instrument capturing aspects of CU was considered. CU outcome measures could be subjective (eg, self-reported number of CU days or joints in the previous 3 months) or objective (eg, drug screening test). CU had to be measured before the intervention (baseline) and at least once after.

Digital CU interventions were defined as web- or mobile-based interventions that included one or more activities (eg, self-directed or interactive psychoeducation or therapy, personalized feedback, peer-to-peer contact, and patient-to-expert communication) aimed at changing CU [58]. Mobile-based interventions were defined as interventions delivered via mobile phone through SMS text message, multimedia messaging service (ie, SMS text messages that include　multimedia content, such as pictures, videos, or emojis), or mobile apps, whereas web-based interventions (eg, websites and digital platforms) were defined as interventions designed to be accessed on the web (ie, the internet), mainly via computers. Interventions could include self-directed and web-based interventions with human support. We defined young adults as aged 16 to 35 years and included students and nonstudents. While young adulthood is typically defined as covering the ages of 18 to 30 years [59], we broadened the range given that the age of majority and legal age to purchase cannabis differs across countries and jurisdictions. This was also in line with the age range targeted by several digital CU interventions (college or university students or emerging adults aged 15-24 years) [31,45]. Given the language expertise of the research team members and the available resources, only English- and French-language articles were retained.

Knowledge synthesis articles, study protocols, and discussion papers or editorials were excluded, as were articles with cross-sectional, cohort, case study or report, pretest-posttest, quasi-experimental, or qualitative designs. Mixed methods designs were included only if the quantitative component was an RCT. We excluded studies if (1) use of substances other than cannabis (eg, alcohol, opioids, or stimulants) was the focus of the digital intervention (though studies that included polysubstance users were retained if CU was assessed and reported separately); (2) CU was not reported separately as an outcome or only attitudes or beliefs regarding, knowledge of, intention to reduce, or readiness or motivation to change CU was measured; and (3) the data reported were unpublished (eg, conferences and dissertations). Studies of traditional face-to-face therapy delivered via teleconference on mobile phones and computers or in a hospital-based setting and informational campaigns (eg, web-based poster presentations or pamphlets) were excluded as well. Studies with samples with a maximum age of <15 years and a minimum age of >35 years were also excluded. Finally, we excluded studies that focused exclusively on people with a mental health disorder or substance use disorder or dependence or on adolescents owing to the particular health care needs of these populations, which may differ from those of young adults [1].

Duplicates were removed from the literature search results in EndNote (version X9.3.3; Clarivate Analytics) using the Bramer method for deduplication of database search results for systematic reviews [60]. The remaining records were uploaded to Covidence (Veritas Health Innovation), a web-based systematic review management system. A reviewer guide was developed that included screening questions and a detailed description of each inclusion and exclusion criterion based on PICO (population, intervention, comparator, and outcome), and a calibration exercise was performed before each stage of the selection process to maximize consistency between reviewers. Titles and abstracts of studies flagged for possible inclusion were screened first by 2 independent reviewers (GC, BV, PA, and GR; 2 per article) against the eligibility criteria (stage 1). Articles deemed eligible for full-text review were then retrieved and screened for inclusion (stage 2). Full texts were assessed in detail against the eligibility criteria again by 2 reviewers independently. Disagreements between reviewers were resolved through consensus or by consulting a third reviewer.

In total, 2 reviewers (GC, BV, PA, GR, and GF; 2 per article) independently extracted relevant data (or informal evidence) using a data extraction form developed specifically for this review and integrated into Covidence. The form was pilot-tested on 2 randomly selected studies and refined accordingly. Data pertaining to the following domains were extracted from the included studies: (1) Study characteristics included information on the first and corresponding authors, publication year, country of origin, aims and hypotheses, study period, design (including details on randomization and blinding), follow-up times, data collection methods, and types of statistical analysis. (2) Participant characteristics included study target population, participant inclusion and exclusion criteria, sex or gender, mean age, and sample sizes at each data collection time point. (3) Intervention characteristics, for which the research team developed a matrix inspired by the template for intervention description and replication 12-item checklist [61] to extract informal evidence (ie, intervention descriptions) from the included studies under the headings name of intervention, purpose, underpinning theory of design elements, treatment approach, type of technology (ie, web or mobile) and software used, delivery format (ie, self-directed, human involvement, or both), provider characteristics (if applicable), intervention duration (ie, length of treatment and number of sessions or modules), material and procedures (ie, tools or activities offered, resources provided, and psychoeducational content), tailoring, and unplanned modifications. (4) Comparator characteristics were details of the control or comparison group or groups, including nature (passive vs active), number of groups or clusters (if applicable), type and length of the intervention (if applicable), and number of participants at each data collection time point. (5) Outcome variables, including the primary outcome variable examined in this systematic review, that is, the mean difference in CU frequency before and after the intervention and between the experimental and control or comparison groups. When possible, we examined continuous variables, including CU frequency means and SDs at the baseline and follow-up time points, and standardized regression coefficients (ie, β coefficients and associated 95% CIs). The secondary outcomes examined included other CU outcome variables (eg, quantity of cannabis used and abstinence) and cannabis-related negative consequences (or problems). Details on outcome variables (ie, definition, data time points, and missing data) and measurements (ie, instruments, measurement units, and scales) were also extracted.

In addition, data on user engagement and use of the digital intervention and study attrition rates (ie, dropouts and loss to follow-up) were extracted. When articles had missing data, we contacted the corresponding authors via email (2 attempts were made over a 2-month period) to obtain missing information. Disagreements over the extracted data were limited and resolved through discussion.

The characteristics of the included studies, study participants, interventions, and comparators were summarized in narrative and table formats. The template for intervention description and replication 12-item checklist [61] was used to summarize and organize intervention characteristics and assess to what extent the interventions were appropriately described in the included articles. As not all studies had usable data for meta-analysis purposes and because of heterogeneity, we summarized the main findings (ie, intervention effects) of the included studies in narrative and table formats for each outcome of interest in this review.

The BCTs used in the digital interventions were identified from the descriptions of the interventions (ie, experimental groups) provided in the articles as well as any supplementary material and previously published research protocols. A BCT was defined as “an observable, replicable, and irreducible component of an intervention designed to alter or redirect causal processes that regulate behavior” [48]. The target behavior in this review was the cessation or reduction of CU by young adults. BCTs were identified and coded using the BCTTv1 [48,49], a taxonomy of 93 BCTs organized into 16 hierarchical thematic clusters or categories. Applying the BCTTv1 in a systematic review allows for the comparison and synthesis of evidence across studies in a structured manner. This analysis allows for the identification of the explicit mechanisms underlying the reported behavior change induced by interventions, successful or not, and, thus, avoids making implicit assumptions about what works [62].

BCT coding was performed by 2 reviewers independently—BV coded all studies, and GC and GF coded a subset of the studies. All reviewers completed web-based training on the BCTTv1, and GF is an experienced implementation scientist who had used the BCTTv1 in prior work [63-65]. The descriptions of the interventions in the articles were read line by line and analyzed for the clear presence of BCTs using the guidelines developed by Michie et al [48]. For each article, the BCTs identified were documented and categorized using supporting textual evidence. They were coded only once per article regardless of how many times they came up in the text. Disagreements about including a BCT were resolved through discussion. If there was uncertainty about whether a BCT was present, it was coded as absent. Excel (Microsoft Corp) was used to compare the reviewers’ independent BCT coding and generate an overall descriptive synthesis of the BCTs identified. The BCTs were summarized by study and BCT cluster.

Meta-analyses were conducted to estimate the size of the effect of the digital interventions for young adult CU on outcomes of interest at the posttreatment and follow-up assessments compared with control or alternative intervention conditions. The outcome variables considered were (1) CU frequency and other CU outcome variables (eg, quantity of cannabis used and abstinence) at baseline and the posttreatment time point or follow-up measured using standardized instruments of self-reported CU (eg, the timeline followback [TLFB] method) [66] and (2) cannabis-related negative consequences measured using standardized instruments (eg, the Marijuana Problems Scale) [67].

Under our systematic review protocol, ≥2 studies were needed for a meta-analysis. On the basis of previous systematic reviews and meta-analyses in the field of digital CU interventions [31,42-45], we expected between-study heterogeneity regarding outcome assessment. To minimize heterogeneity, we chose to pool studies with similar outcomes of interest based on four criteria: (1) definition of outcome (eg, CU frequency, quantity consumed, and abstinence), (2) type of outcome variable (eg, days of CU in the previous 90 days, days high per week in the previous 30 days, and number of CU events in the previous month) and measure (ie, instruments or scales), (3) use of validated instruments, and (4) posttreatment or follow-up time points (eg, 2 weeks or 1 month after the baseline or 3, 6, and 12 months after the baseline).

Only articles that reported sufficient statistics to compute a valid effect size with 95% CIs were included in the meta-analyses. In the case of articles that were not independent (ie, more than one published article reporting data from the same clinical trial), only 1 was included, and it was represented only once in the meta-analysis for a given outcome variable regardless of whether the data used to compute the effect size were extracted from the original paper or a secondary analysis paper. We made sure that the independence of the studies included in the meta-analysis of each outcome was respected. In the case of studies that had more than one comparator, we used the effect size for each comparison between the intervention and control groups.

Meta-analyses were conducted only for mean differences based on the change from baseline in CU frequency at 3 months after the baseline as measured using the number of self-reported days of use in the previous month. As the true value of the estimated effect size for outcome variables might vary across different trials and samples, we used a random-effects model given that the studies retained did not have identical target populations. The random-effects model incorporates between-study variation in the study weights and estimated effect size [68]. In addition, statistical heterogeneity across studies was assessed using I², which measures the proportion of heterogeneity to the total observed dispersion; 25% was considered low, 50% was considered moderate, and 75% was considered high [69]. Because only 3 studies were included in the meta-analysis [70-72], publication bias could not be assessed. All analyses were completed using Stata (version 18; StataCorp) [73].

Multimedia Appendix 5 [70-72,77-92] describes the general characteristics of the 19 included studies. The studies were published between 2010 and 2023, with 58% (11/19) published in 2018 or later. A total of 53% (10/19) of the studies were conducted in the United States [77-86], 11% (2/19) were conducted in Canada [87,88], 11% (2/19) were conducted in Australia [71,89], 11% (2/19) were conducted in Germany [72,90], 11% (2/19) were conducted in Switzerland [70,91], and 5% (1/19) were conducted in Sweden [92]. A total of 79% (15/19) were RCTs [70-72,77,79,81-83,86-92], and 21% (4/19) were pilot RCTs [78,80,84,85].

The studies enrolled a total of 6710 participants—3229 (48.1%) in the experimental groups, 3358 (50%) in the control groups, and the remaining 123 (1.8%) from 1 study [82] where participant allocation to the intervention condition was not reported. Baseline sample sizes ranged from 49 [81] to 1292 [72] (mean 352.89, SD 289.50), as shown in Multimedia Appendix 5. Participant mean ages ranged from 18.03 (SD 0.31) [79] to 35.3 (SD 12.6) years [88], and the proportion of participants who identified as female ranged from 24.7% [91] to 84.1% [80].

Of the 19 included studies, 10 (53%) targeted adults aged ≥18 years, of which 7 (70%) studies focused on adults who had engaged in past-month CU [70,71,80,84,85,90,91], 2 (20%) studies included adults who wished to reduce or cease CU [72,89], and 1 (10%) study focused on noncollege adults with a moderate risk associated with CU [88]. Sinadinovic et al [92] targeted young adults aged ≥16 years who had used cannabis at least once a week in the previous 6 months. The remaining 8 studies targeted college or university students (aged ≥17 y) specifically, of which 7 (88%) studies focused solely on students who reported using cannabis [78,79,81-83,86,87] and 1 (12%) study focused solely on students who did not report past-month CU (ie, abstainers) [77].

The 19 included studies assessed nine different digital interventions: (1) 5 (26%) evaluated Marijuana eCHECKUP TO GO (e-TOKE), a commercially available electronic intervention used at colleges throughout the United States and Canada [77,78,81-83]; (2) 2 (11%) examined the internationally known CANreduce program [70,91]; (3) 2 (11%) evaluated the German Quit the Shit program [72,90]; (4) 2 (11%) assessed a social media–delivered, physical activity–focused cannabis intervention [84,85]; (5) 1 (5%) investigated the Swedish Cannabishjälpen intervention [92]; (6) 1 (5%) evaluated the Australian Grassessment: Evaluate Your Use of Cannabis website program [89]; (7) 1 (5%) assessed the Canadian Ma réussite, mon choix intervention [87]; (8) 1 (5%) examined the Australian Reduce Your Use: How to Break the Cannabis Habit program [71]; and (9) 4 (21%) each evaluated a unique no-name intervention described as a personalized feedback intervention (PFI) [79,80,86,88]. Detailed information regarding the characteristics of all interventions as reported in each included study is provided in Multimedia Appendix 6 [70-72,77-113] and summarized in the following paragraphs.

In several studies (8/19, 42%), the interventions were designed to support cannabis users in reducing or ceasing their consumption [70,72,80,87,89-92]. In 37% (7/19) of the studies, the interventions aimed at reducing both CU and cannabis-related consequences [79,81-85,88]. Other interventions focused on helping college students think carefully about the decision to use cannabis [77,78] and on reducing either cannabis-related problems among undergraduate students [86] or symptoms associated with CU disorder in young adults [71].

In 26% (5/19) of the studies, theory was used to inform intervention design along with a clear rationale for theory use. Of these 5 articles, only 1 (20%) [87] reported using a single theory of behavior change, the theory of planned behavior [114]. A total of 21% (4/19) of the studies selected only constructs of theories (or models) for their intervention design. Of these 4 studies, 2 (50%) evaluated the same intervention [72,90], which focused on principles of self-regulation and self-control theory [93]; 1 (25%) [70] used the concept of adherence-focused guidance enhancement based on the supportive accountability model of guidance [94]; and 1 (25%) [71] reported that intervention design was guided by the concept of self-behavioral management.

The strategies (or approaches) used in the delivery of the digital interventions were discussed in greater detail in 84% (16/19) of the articles [70-72,79-81,83-92]. Many of these articles (9/19, 47%) reported using a combination of approaches based on CBT or motivational interviewing (MI) [70,71,79,83-85,90-92]. PFIs were also often mentioned as an approach to inform intervention delivery [7,71,79,86-88].

More than half (13/19, 68%) of all the digital interventions were asynchronous and based on a self-guided approach without support from a counselor or therapist. The study by Côté et al [87] evaluated the efficacy of a web-based tailored intervention focused on reinforcing a positive attitude toward and a sense of control over cannabis abstinence through psychoeducational messages delivered by a credible character in short video clips and personalized reinforcement messages. Lee et al [79] evaluated a brief, web-based personalized feedback selective intervention based on the PFI approach pioneered by Marlatt et al [95] for alcohol use prevention and on the MI approach described by Miller and Rollnick [96]. Similarly, Rooke et al [71] combined principles of MI and CBT to develop a web-based intervention delivered via web modules, which were informed by previous automated feedback interventions targeting substance use. The study by Copeland et al [89] assessed the short-term effectiveness of Grassessment: Evaluate Your Use of Cannabis, a brief web-based, self-complete intervention based on motivational enhancement therapy that included personalized feedback messages and psychoeducational material. In the studies by Buckner et al [80], Cunningham et al [88], and Walukevich-Dienst et al [86], experimental groups received a brief web-based PFI available via a computer. A total of 16% (3/19) of the studies [77,78,82] applied a program called the Marijuana eCHECKUP TO GO (e-TOKE) for Universities and Colleges, which was presented as a web-based, norm-correcting, brief preventive and intervention education program designed to prompt self-reflection on consequences and consideration of decreasing CU among students. Riggs et al [83] developed and evaluated an adapted version of e-TOKE that provided participants with university-specific personalized feedback and normative information based on protective behavioral strategies for CU [97]. Similarly, Goodness and Palfai [81] tested the efficacy of eCHECKUP TO GO-cannabis, a modified version of e-TOKE combining personalized feedback, norm correction, and a harm and frequency reduction strategy where a “booster” session was provided at 3 months to allow participants to receive repeated exposure to the intervention.

In the remaining 32% (6/19) of the studies, which examined 4 different interventions, the presence of a therapist guide was reported. The intervention evaluated by Sinadinovic et al [92] combined principles of psychoeducation, MI, and CBT organized into 13 web-based modules and a calendar involving therapist guidance, recommendations, and personal feedback. In total, 33% (2/6) of these studies evaluated a social media–delivered intervention with e-coaches that combined principles of MI and CBT and a harm reduction approach for risky CU [84,85]. Schaub et al [91] evaluated the efficacy of CANreduce, a web-based self-help intervention based on both MI and CBT approaches, using automated motivational and feedback emails, chat with a counselor, and web-based psychoeducational modules. Similarly, Baumgartner et al [70] investigated the effectiveness of CANreduce 2.0, a modified version of CANreduce, using semiautomated motivational and adherence-focused guidance-based email feedback with or without a personal online coach. The studies by Tossman et al [72] and Jonas et al [90] used a solution-focused approach and MI to evaluate the effectiveness of the German Quit the Shit web-based program that involves weekly feedback provided by counselors.

In addition to using different intervention strategies or approaches, the interventions were diverse in terms of the duration and frequency of the program (eg, web-based activities, sessions, or modules). Of the 12 articles that provided details in this regard, 2 (17%) on the same intervention described it as a brief 20- to 45-minute web-based program [77,78], 2 (17%) on 2 different interventions reported including 1 or 2 modules per week for a duration of 6 weeks [71,92], and 7 (58%) on 4 different interventions described them as being available over a longer period ranging from 6 weeks to 3 months [70,72,79,84,85,87,90,91].

A total of 42% (8/19) of the studies [72,77-80,85,87,92] used a passive comparator only, namely, a waitlist control group (Multimedia Appendix 5). A total of 26% (5/19) of the studies used an active comparator only where participants were provided with minimal general health feedback regarding recommended guidelines for sleep, exercise, and nutrition [81,82]; strategies for healthy stress management [83]; educational materials about risky CU [88]; or access to a website containing information about cannabis [71]. In another 21% (4/19) of the studies, which used an active comparator, participants received the same digital intervention minus a specific component: a personal web-based coach [70], extended personalized feedback [89], web-based chat counseling [91], or information on risks associated with CU [86]. A total of 21% (4/19) of the studies had more than one control group [70,84,90,91].

The methodological characteristics and major findings of the included studies (N=19) are presented in Multimedia Appendix 7 [67,70-72,77-92,115-120] and summarized in the following sections for each outcome of interest in this review (ie, CU and cannabis-related consequences). Of the 19 studies, 11 (58%) were reported as efficacy trials [7,77,79,81-83,86-88,91,92], and 8 (42%) were reported as effectiveness trials [70-72,78,84,85,89,90].

Across all the included studies (19/19, 100%), participant attrition rates ranged from 1.6% at 1 month after the baseline [77,78] to 75.1% at the 3-month follow-up [70]. A total of 37% (7/19) of the studies assessed and reported results regarding user engagement [71,78,84,85,90-92] using different types of metrics. In one article on the Marijuana eCHECKUP TO GO (e-TOKE) web-based program [78], the authors briefly reported that participation was confirmed for 98.1% (158/161) of participants in the intervention group. In 11% (2/19) of the studies, which were on a similar social media–delivered intervention [84,85], user engagement was quantified by tallying the number of comments or posts and reactions (eg, likes and hearts) left by participants. In both studies [84,85], the intervention group, which involved a CU-related Facebook page, displayed greater interactions than the control groups, which involved a Facebook page unrelated to CU. One article [84] reported that 80% of participants in the intervention group posted at least once (range 0-60) and 50% posted at least weekly. In the other study [85], the results showed that intervention participants engaged (ie, posting or commenting or clicking reactions) on average 47.9 times each over 8 weeks. In total, 11% (2/19) of the studies [90,91] on 2 different web-based intervention programs, both consisting of web documentation accompanied by chat-based counseling, measured user engagement either by average duration or average number of chat sessions. Finally, 16% (3/19) of the studies [71,91,92], which involved 3 different web-based intervention programs, characterized user engagement by the mean number of web modules completed per participant. Overall, the mean number of web modules completed reported in these articles was quite similar: 3.9 out of 13 [92] and 3.2 [91] and 3.5 [71] out of 6.

As presented in Multimedia Appendix 7, the included studies differed in terms of how they assessed CU, although all used at least one self-reported measure of frequency. Most studies (16/19, 84%) measured frequency by days of use, including days of use in the preceding week [91] or 2 [80], days of use in the previous 30 [70-72,78,84-86,88-90] or 90 days [79,81,82], and days high per week [83]. Other self-reported measures of CU frequency included (1) number of CU events in the previous month [87,90], (2) cannabis initiation or use in the previous month (ie, yes or no) [77], and (3) days without CU in the previous 7 days [92]. In addition to measuring CU frequency, 42% (8/19) of the studies also assessed CU via self-reported measures of quantity used, including estimated grams consumed in the previous week [92] or 30 days [72,85,90] and the number of standard-sized joints consumed in the previous 7 days [91] or the previous month [70,71,89].

Of the 19 articles included, 10 (53%) [70-72,80,84-86,89,90,92] reported using a validated instrument to measure CU frequency or quantity, including the TLFB instrument [66] (n=9, 90% of the studies) and the Marijuana Use Form (n=1, 10% of the studies); 1 (10%) [79] reported using CU-related questions from an adaptation of the Global Appraisal of Individual Needs–Initial instrument [115]; and 30% (3/10) [81,82,91] reported using a questionnaire accompanied by a calendar or a diary of consumption. The 19 studies also differed with regard to their follow-up time measurements for assessing CU, ranging from 2 weeks after the baseline [80] to 12 months after randomization [90], although 12 (63%) of the studies included a 3-month follow-up assessment [70-72,79,81,82,84,85,88,90-92].

Of all studies assessing and reporting change in CU frequency from baseline to follow-up assessments (19/19, 100%), 47% (9/19) found statistically significant differences between the experimental and control groups [70-72,80,81,83,85,87,91]. Importantly, 67% (6/9) of these studies showed that participants in the experimental groups exhibited greater decreases in CU frequency 3 months following the baseline assessment compared with participants in the control groups [70-72,81,85,91], 22% (2/9) of the studies showed greater decreases in CU frequency at 6 weeks after the baseline assessment [71,83], 22% (2/9) of the studies showed greater decreases in CU frequency at 6 months following the baseline assessment [81,85], 11% (1/9) of the studies showed greater decreases in CU frequency at 2 weeks after the baseline [80], and 11% (1/9) of the studies showed greater decreases in CU frequency at 2 months after treatment [87].

In the study by Baumgartner et al [70], a reduction in CU days was observed in all groups, but the authors reported that the difference was statistically significant only between the intervention group with the service team and the control group (the reduction in the intervention group with social presence was not significant). In the study by Bonar et al [85], the only statistically significant difference between the intervention and control groups at the 3- and 6-month follow-ups involved total days of cannabis vaping in the previous 30 days. Finally, in the study by Buckner et al [80], the intervention group had less CU than the control group 2 weeks after the baseline; however, this was statistically significant only for participants with moderate or high levels of social anxiety.

A total of 53% (10/19) of the studies also assessed cannabis-related negative consequences [78-84,86,88,92]. Of these 10 articles, 8 (80%) reported using a validated self-report instrument: 4 (50%) [81,82,86,88] used the 19-item Marijuana Problems Scale [67], 2 (25%) [78,79] used the 18-item Rutgers Marijuana Problem Index [121,122], and 2 (25%) [80,84] used the Brief Marijuana Consequences Questionnaire [116]. Only 10% (1/10) of the studies [92] used a screening tool, the Cannabis Abuse Screening Test [117,118]. None of these 10 studies demonstrated a statistically significant difference between the intervention and control groups. Of note, Walukevich-Dienst et al [86] found that women (but not men) who received an web-based PFI with additional information on CU risks reported significantly fewer cannabis-related problems than did women in the control group at 1 month after the intervention (B=−1.941; P=.01).