Impact of a computerized intervention for high distress intolerance on cannabis use outcomes: A randomized controlled trial

1. Introduction

Regular cannabis use among adults has increased in prevalence over the past two decades in the United States; for instance, national survey data suggest that daily/near-daily prevalence has increased from 1.9% in 2002 to 3.5% in 2014 (Compton et al., 2016). Although most cannabis users perceive regular use to pose minimal risk (Okaneku et al., 2015), past-year cannabis use disorder (CUD) is associated with significant disability (Hasin et al., 2016) and nationally representative data suggest past-year CUD prevalence has increased from 1.5% in 2001–2002 to 2.9% in 2012–2013 (Hasin et al., 2015; Hasin & Grant, 2016). Unfortunately, there are no FDA-approved medications for CUD, extant psychotherapeutic approaches have modest efficacy (e.g., abstinence rates ~20%; Gates et al., 2016), and many cannabis users with CUD do not seek treatment (Stinson et al., 2006), all of which creates an urgent and growing unmet public health need. Delivery of brief, harm reduction-focused interventions to high-risk cannabis users is one approach that has demonstrated efficacy in reducing cannabis use and/or use-related problems (Parmar & Sarkar, 2017), but these interventions do not generally target mechanisms implicated in theoretical models of the development and maintenance of substance use disorders (SUDs) (e.g., Baker et al., 2004). Treatment programs need brief interventions that target treatment-malleable risk factors that have been empirically and theoretically linked directly to the exacerbation and maintenance of CUD in current cannabis users. Such interventions could have a substantial public health impact given their personalization for specific subgroups of at-risk users (Shoham & Insel, 2011) and dissemination potential (Kazdin & Blase, 2011).

A well-supported model of the development of SUDs posits that escape from negative affect is the motivational core of addictive behavior (Baker et al., 2004). Repeated withdrawal-use cycles sensitize individuals to internal cues signaling increasing negative affect (i.e., affective components of withdrawal), which become generalized to negative affect arising from nonpharmacological sources, linking aversive emotional states to substance use behavior via associative learning and negatively reinforced conditioning. Consistent with this model, negative reinforcement motives for cannabis use are stronger predictors of CUD onset, severity, and maintenance than other motives (Moitra et al., 2015; van der Pol et al., 2013; van der Pol et al., 2015). Relatedly, relief craving, defined as craving for a substance specifically to relieve negative emotional states (see Glöckner-Rist et al., 2013, Verheul et al., 1999), may play an important role in motivating cannabis use during acute stress states among individuals with CUD (e.g., McRae-Clark et al., 2011). Finally, neuroscience-informed approaches to understanding the development of dependence also emphasize the centrality of internal affective cues, and suggest that individual differences in the propensity to negatively appraise and escape from these cues (e.g., distress intolerance [DI]) may be an important risk factor for SUDs (Verdejo-Garcia et al., 2012).

DI is a vulnerability factor (Kraemer et al., 1997) reflective of the learned tendency to negatively appraise and respond avoidantly to aversive emotional states, and research has theorized it to be a vulnerability factor important to SUDs (Leyro et al., 2010), including CUD. Self-report measures asses an individuals’ perceived capacity to tolerate emotional distress in general, whereas behavioral measures typically operationalize DI as the quit latency on a distressing laboratory task (McHugh & Otto, 2011). Self-reported DI has been positively linked with CUD severity in two community samples of regular users (Bujarski et al., 2012; Farris et al., 2016) and greater cannabis use during a quit attempt in disordered users (Hasan et al., 2015). Further, the self-reported inability to refrain from cannabis use while emotionally distressed has been associated with greater cannabis use during CUD treatment (Gullo et al., 2017). Of particular relevance to negative reinforcement models of the development of SUDs (Baker et al., 2004), high self-reported DI has been associated with stress-induced cannabis craving (Buckner et al., 2019) and coping motives for cannabis use in regular users (Bujarski et al., 2012; Farris et al., 2016), suggesting that self-reported DI may be an important intervention target for reducing coping-motived cannabis use and associated problems.

A DI-targeted intervention may have utility as a standalone or adjunctive treatment, and may potentially be most impactful for specific subgroups of cannabis users (e.g., those with high DI). Only two studies known to the authors have explicitly tested a DI-targeted intervention in a RCT (Bornovalova et al., 2012; Macatee & Cougle, 2015). Bornovalova and colleagues (2012) reported that a 6-week, DI-targeted group psychosocial treatment delivered in the context of ongoing residential addiction treatment resulted in greater improvement in behavioral DI at post-treatment than supportive counseling or treatment-as-usual control conditions in substance dependent patients with elevated pre-treatment DI (13.5% with cannabis dependence). However, the authors did not assess self-reported DI nor were substance use treatment outcomes reported, precluding stronger conclusions about DI’s theorized causal role in SUD maintenance. Macatee and Cougle (2015) showed that a two-session, DI-targeted computerized intervention resulted in greater decreases in self-reported DI at post-treatment relative to a waitlist control in a sample of young adults with elevated pre-treatment self-reported DI. The study also assessed behavioral DI, which revealed greater decreases in persistence time in the waitlist control compared to intervention group, conceptually consistent with Bornovalova and colleagues’ (2012) finding. Further, reductions in self-reported and behavioral DI mediated the intervention’s effect on decreased subjective urges to avoid an in-vivo, aversive stimulus elicited in the lab (i.e., visualization of a negative, intrusive thought), suggesting the possibility that the intervention may affect other behaviors maintained through negative reinforcement (e.g., stress-elicited cannabis craving, coping motives for cannabis use). However, this study only recruited individuals with high self-reported DI and did not target distress intolerant cannabis users specifically. To replicate and extend the promising findings that Macatee and Cougle (2015) reported and to continue to develop a DI intervention, studies must use a credible control condition, include follow-up time points, recruit distress intolerant cannabis users, and assess the intervention’s impact on cannabis use outcomes.

The current study sought to fill these important gaps in the literature. This study evaluated the impact of a two-session, DI-targeted computerized intervention on cannabis use outcomes in a sample of regular cannabis users with high self-reported DI. Although the pilot RCT showed that the intervention reduced DI as expected, the pilot used a waitlist control condition (Macatee & Cougle, 2015). Thus, we used a control treatment condition to rule-out expectancy effects. Further, the pilot RCT did not conduct follow-up assessments, which meant that the study could not evaluate the maintenance of DI reductions over time. Therefore, this study assessed self-reported DI at post-treatment as well as one- and four-month follow-up assessments. To test the intervention’s impact on a proximal, CUD-related outcome, we assessed acute stress–elicited relief cannabis craving pre and post-treatment. To test the intervention’s effect on more distal, CUD-related outcomes, we assessed cannabis use frequency, coping motives for cannabis use, and cannabis use–related problems pre-treatment, post-treatment, and both one- and four-month follow-ups. In addition, we assessed CUD criteria pre-treatment and at four-month follow-up. This design allowed us to evaluate the intervention’s target engagement (i.e., self-reported and behavioral DI), impact on proximal (i.e., acute stress–elicited relief cannabis craving) and distal cannabis use outcomes (cannabis use frequency, coping motives for cannabis use, cannabis use–related problems, CUD criteria), as well as the theorized causal role of DI in the maintenance of CUD.

Based on the pilot RCT findings (Macatee & Cougle, 2015), we hypothesized that (1) the computerized, DI-targeted intervention group would demonstrate greater decreases in self-reported and behavioral DI from pre- to post-treatment relative to the control group, and that patients would maintain self-reported DI reductions at both follow-up assessments. Based on extant research showing that self-reported DI is positively associated with stress-elicited cannabis craving, cannabis use frequency after a quit attempt, coping motives for cannabis use, CUD severity, and cannabis use–related problems (Buckner et al., 2019; Bujarski et al., 2012; Hasan et al., 2015; Farris et al., 2016), we hypothesized that (2) the DI-targeted intervention group would demonstrate greater decreases in acute stress–elicited relief cannabis craving from pre- to post-treatment relative to the control group, as well as greater decreases in cannabis use frequency, coping motives for cannabis use, CUD severity, and cannabis use–related problems from pre-treatment through the four-month follow-up. Finally, we hypothesized that (3) the intervention effect on cannabis use outcomes would be mediated by reductions in self-reported DI.

2. Materials and methods

2.1. Participants

We recruited sixty participants from a large university in the southeastern United States and the surrounding community. We used online postings to Craigslist, Facebook, and Reddit as well as printed fliers to advertise a computerized stress management treatment for regular cannabis users who struggle with effectively managing negative emotions. Inclusion criteria consisted of: (1) English fluency, (2) 18–30 years old, (3) Distress Intolerance Index (DII; McHugh & Otto, 2012) score >= 20 based on the median in treatment-seeking samples (Macateeet al., 2018a) and the cut-off used in the pilot RCT (Macatee & Cougle, 2015), and (4) average self-reported cannabis use of at least 2–3 times per week over the past year. We excluded participants from the study if they endorsed (1) current suicidal ideation, (2) history of psychotic symptoms, (3) history of a bipolar-spectrum disorder in the absence of stabilization on psychotropic medication for >= three months, (4) current cognitive behavioral therapy, or (5) a change in psychotropic medication regimen in the past month. We compensated participants $90 total for completing the screening, pre-treatment, treatment sessions, and post-treatment portions of the study; alternatively, participants could earn course credit for completing these portions of the study if they were current psychology students. Of the randomized participants, 31 received $90 for completing the assessments, and 29 received course credit. We compensated all participants $20 total for completing the one- and four-month follow-up assessments. Of the randomized sample, 91.7% had a current CUD diagnosis, and of these participants 20.0%, 27.3%, and 52.7% had CUD severity specifiers of mild, moderate, and severe, respectively.

2.2. Measures

2.2.1. DI outcomes

2.2.1.1. Distress Intolerance Index (DII; McHugh & Otto, 2012).

The DII is a 10-item self-report measure that assesses individual differences in the perceived capacity to tolerate distress. Respondents rate the extent to which they agree each item reflects their beliefs about distress on a zero (“not at all”) to four (“very much”) scale; the summed total score can range from zero to 40. The DII has demonstrated adequate retest and internal reliability (Cakir, 2016; McHugh & Otto, 2011) as well as convergent validity with behavioral measures of DI (McHugh & Otto, 2011). We administered the DII during the screening, pre-treatment, post-treatment, and both follow-up appointments; the internal consistency across pre-treatment, post-treatment, and follow-up appointments was good to excellent (α=.82-.92).

2.2.1.2. Mirror-Tracing Persistence Task (MTPT; Quinn et al., 1996).

The MTPT is a computerized behavioral assessment of DI. Research staff tell the participant to use the computer mouse to trace a red dot along the contours of three increasingly difficult stars. To make the task difficult and distressing, the computer mouse controls are reversed (e.g., moving the mouse right moves the red dot left, moving the mouse down moves the red dot up) and, upon participant error, the assessment plays a loud, aversive sound and the red dot resets to the beginning of the star. Following the three stars, the program presents a final star that is tailored to the participant’s skill level. Staff tell the participant that they can quit the final star at any time by pressing the spacebar, but that the number of times they are entered into a lottery for a $20 gift card is proportional to their performance on the final star. Unbeknownst to the participants, the final star ends automatically after seven minutes. We use latency to final star termination as the behavioral index of DI; predictive validity (e.g., substance use treatment dropout; Daughters et al., 2005), convergent validity with other behavioral and self-report measures of DI (McHugh & Otto, 2011), and retest reliability (Bornovalova et al., 2008) of MTPT quit latency have been established in prior work. In the current study, we administered the MTPT pre- and post-treatment. To ensure that the task was distressing, we measured subjective stress before and after the first three stars using the mean of five negative emotion words (anxiety, frustration, irritability, difficulty concentrating, physical discomfort) scored on a 0–100 VAS scale. Internal consistency of the subjective stress scale was acceptable to excellent in the combined sample at both pre- and post-treatment (α=.76–91).

2.2.2. Cannabis use outcomes

2.2.2.1. Marijuana Craving Questionnaire – Short Form (MCQ; Heishman et al., 2009).

The MCQ is a self-report measure that assesses four forms of current cannabis craving (i.e., compulsivity, emotionality, expectancy, purposefulness) with 12 items rated on a one (“strongly disagree”) to seven (“strongly agree”) Likert-type scale. The MCQ has demonstrated adequate internal consistency (Heishman et al., 2009) and sensitivity to acute stress inductions (McRae-Clark et al., 2011) as well as CUD treatment (Trigo et al., 2018) in prior research. We averaged the three “emotionality” subscale items to form the relief craving measure used in the current study. We administered the MCQ four times at both pre- and post-treatment assessments. Specifically, participants completed the MCQ before and after a cue-exposure task that itself was repeated before and after a stress induction (described below). The cue exposure task involved randomized presentation of threatening, cannabis-related, and neutral image blocks that we asked the participant to view; we included this task to test secondary neurophysiological outcomes that will be explored in separate papers (see Macatee et al., 2019 for examination of baseline data). Before and after viewing all of the images, participants completed the MCQ to control for the influence of cannabis cue exposure on craving. The internal consistency of the emotionality subscale across all administrations at pre- and post-treatment was acceptable to good (α=.75–.87).

2.2.2.2. Structured Clinical Interview for DSM-V – Research Version (SCID-5-RV; First et al., 2015).

During the screening interview, we administered bipolar-spectrum and psychotic disorder modules to assess exclusion criteria, and we assessed lifetime and current (i.e., past three months) CUD criteria using the substance use disorder module. We assessed current CUD criteria again at the four-month follow-up. We used total number of current CUD criteria to operationalize CUD severity at the screening interview and four-month follow-up. A previous study used CUD criteria count to operationalize severity (Farris et al., 2016), and research has shown criteria counts for alcohol use disorder to be sensitive to treatment (Kiluk et al., 2018). A doctoral-level graduate student and a post-baccalaureate research assistant blind to treatment condition assignment administered interviews. Previously published data on reliability coding of a subsample of screening interview CUD criteria counts revealed excellent inter-rater reliability (r=.98, p<.001; Macatee et al., 2019).

2.2.2.3. Marijuana Problems Scale (MPS; Stephens et al., 2000).

The MPS is a self-report measure asking respondents to rate the extent to which they have experienced each of 19 cannabis use–related problems in the past month using a zero to two scale (0 = no problem, 1 = minor problem, 2 = serious problem); items are summed to form a total score. The MPS has demonstrated good internal consistency (Buckner et al., 2007) and sensitivity to CUD treatment (Budney et al., 2015) in prior work. We administered the MPS at the pre-treatment, post-treatment, one-month follow-up, and four-month follow-up assessments; the internal consistency across administrations was good (α=.83–.87).

2.2.2.4. Marijuana Motives Measure (MMM; Simons et al., 1998).

The MMM is a 25-item self-report measure that assesses five forms of cannabis use motives using a one (“almost never/never”) to five (“almost always/always”) Likert-type scale. The coping subscale consists of four items that assess emotional relief motives for cannabis use; this subscale has demonstrated good internal consistency (Simons et al., 1998) and sensitivity to CUD treatment (Banes et al., 2014) in prior work. We administered the MMM at pre-treatment to assess coping motives for cannabis use in general and post-treatment to assess motives for cannabis use in the past three weeks, as well as at both one- and four-month follow-up assessments to assess motives for cannabis use in the past month; the internal consistency across all administrations was good (α=.82–.89).

2.2.2.5. Timeline Follow-Back (TLFB; Robinson et al., 2014).

We used a self-report version of the TLFB to assess percentage of cannabis use days in the past four weeks at the pre-treatment, one-month follow-up, and four-month follow-up assessments, whereas we assessed percentage of use days during the treatment period (i.e., from treatment session 1 to post-treatment) at the post-treatment assessment. Self-reported number of cannabis use days on the TLFB has demonstrated excellent reliability (Robinson et al., 2014), covariation with urinalysis-confirmed cannabis use (Hjorthoj et al., 2012), and sensitivity to CUD treatment (Budney et al., 2015) in prior work.

2.2.3. Cannabis use and treatment descriptives

2.2.3.1. Cannabis use history and motivation to change use.

The Marijuana Ladder (Slavet et al., 2006) is a visual analogue comprising 10 rungs with statements corresponding to the stages of behavior change for cannabis use. It has demonstrated good concurrent and predictive validity with regard to treatment engagement and cannabis use (Slavet et al., 2006). We used the Marijuana Smoking History Questionnaire (Bonn-Miller & Zvolensky, 2009) to assess age at initial cannabis use, age at regular cannabis use onset, and years of regular cannabis use. Because we did not recruit participants based on cannabis use reduction intentions or their long-term cannabis use history, we used these measures to ensure treatment groups were matched on motivation to reduce use and long-term cannabis use history.

2.2.3.2. Credibility/Expectancy Questionnaire (CEQ; Borkovec & Mathews, 1988).

The CEQ is a self-report measure of the logical credibility and expected efficacy of a presented treatment. We assessed treatment credibility with three items rated on a one to nine scale, and we assessed expected improvement in stress management using a discrete zero to 100% scale with steps of 10%. Consistent with Borkovec and Matthews (1998), we scored credibility by summing all three items and we examined the expectancy item on its own. The CEQ has demonstrated good psychometric properties and has predicted treatment outcome (Borkovec & Matthews, 1998). We administered the CEQ at the end of treatment session one in both conditions; we included it to ensure treatments were matched on credibility/expectancy and, if not, to be used as a covariate in analyses. The credibility scale demonstrated good internal consistency (α=.88) in the current study.

2.3. Interventions

2.3.1. Distress Tolerance Intervention (DTI).

We developed this computerized intervention to closely model similar computerized interventions for related risk factors (i.e., anxiety sensitivity; Schmidt et al., 2014) and target the core DI construct as revealed by recent factor analytic findings (McHugh & Otto, 2012) that demonstrated negative affective responses to distress and strong action tendencies to immediately reduce distress to be most central to the risk factor. Thus, the intervention consisted of two components presented in two, one-hour computerized sessions: psychoeducation and imaginal emotional exposure (see Macatee & Cougle, 2015 for greater detail). In the current study, largely in response to participant feedback from the pilot RCT, we modified the DTI from the version used in Macatee and Cougle (2015) in the following ways: (1) We used idiographic distress scripts in both session one and two; (2) Participants used an index card to construct a customized implementation intention to encourage practice of emotional tolerance in personally relevant contexts over the following week; and (3) Imaginal emotional exposure used real-time recording of skin conductance to index habituation, with feedback presented to the participant regarding their latency to skin conductance habituation and trajectory of subjective distress during imaginal exposure (see supplemental material for more detail).

2.3.2. Healthy Video Control (HVC).

The HVC condition is a psychoeducational, computerized video series based on protocols that other researchers have used as a placebo intervention (e.g., Schmidt et al., 2014). Each of the two, one-hour sessions consist of four 15-minute videos that discuss healthy habits and self-care (e.g., sleep, nutrition, hygiene, exercise). Quizzes are presented at the end of each session to ensure comprehension of the psychoeducational material. We presented participants with the rationale that healthy habits and self-care are beneficial for stress management. Prior work has perceived this intervention as credible for negative affect reduction (e.g., Smith et al., 2018).

2.4. Laboratory stress induction (Mannheim Multimodal Stress Test [MMST]; Reinhardt et al., 2012).

The MMST is a five-minute computerized task that involves simultaneous exposure to (1) increasingly difficult mental arithmetic, (2) continuous presentation of negatively valenced, highly arousing images, (3) continuous loud white noise, and (4) loss of monetary reward for each mistake during the mental arithmetic. Significant subjective, autonomic, and cortisol reactivity to the MMST has been reported in healthy and clinical samples (Cackowski et al., 2014; Reinhardt et al., 2012). In the current study, we measured subjective stress before and after the MMST using the mean of five negative emotion words (anxiety, frustration, irritability, difficulty concentrating, physical discomfort) scored on the same 0–100 VAS scale utilized in the MTPT. We used the total number of correct responses on the mental arithmetic during the MMST as a behavioral index of stress reactivity, with a smaller total number of correct responses indicating greater stress reactivity. Internal consistency of the subjective stress scale was good in the combined sample at both pre- and post-treatment (α=.83–.91). We used the MMST as a stress induction at pre and post-treatment to test the impact of intervention group on acute stress potentiation of negative reinforcement cannabis craving.

3. Procedure

The Institutional Review Board approved all components of this study, and we retrospectively registered the study with ClinicalTrials.gov (NCT04173078). Recruitment began in June 2016 and we completed all follow-up assessments by October 2017. We obtained informed consent from each participant prior to their completion of the online screening questionnaire. Due to concerns over the impact of acute/residual intoxication on task-related neural activity during the pre/post-treatment assessments and learning/memory during treatment sessions, we asked all participants to abstain from nonprescribed substance use for at least 24 hours prior to the pre- and post-treatment assessments as well as both treatment sessions. Further, we asked participants to abstain from caffeine/nicotine on the day of each visit. We confirmed adherence verbally prior to each visit and, in the event of a reported violation, we rescheduled the visit without penalty. Figure 1 provides a diagram of study flow; we present reasons for missing data in the supplemental material.

Figure 1. CONSORT diagram.

This figure depicts the study flow from pre-treatment through the four-month assessment in the Distress Tolerance Intervention and Healthy Video Control groups.

DTI=Distress Tolerance Intervention, HVC=Healthy Video Control, MMST=Mannheim Multimodal Stress Test, MTPT=Mirror-Tracing Persistence Task, CUD=Cannabis Use Disorder.

3.1. Screening process

We directed interested participants who consented to an online screening questionnaire where we assessed inclusion criteria. We assessed inclusion criteria with questions about the participant’s age, perceived distress intolerance, and average cannabis use frequency in the past year. We contacted participants meeting inclusion criteria to schedule a clinical interview conducted over the phone or in-person to assess current/lifetime CUD severity and exclusion criteria. We assessed exclusion criteria with the psychotic and bipolar-spectrum SCID-V modules, questions about history of cognitive behavioral therapy and psychotropic medication use, and the current suicidal ideation item on the Beck Depression Inventory-II (BDI-II; Dozois et al., 1998). We scheduled interested participants who did not meet any exclusion criteria for a pre-treatment laboratory assessment.

3.2. Pre-treatment assessment

Upon arrival to the lab, research staff asked all participants to think of four recent stressful events in which the participant found it difficult to calm down and that continue to elicit distress when the participant thinks about them. Participants rated each event using a 0 (“Wasn’t a stressful event”) to 100 (“Was the most stressful event in my life”) scale and then wrote about the two highest rated events in detail, focusing on the circumstances of the event and the bodily sensations, emotions, and cognitions experienced during the event. We used these event descriptions to generate second-person, present-tense scripts for the imaginal exposure portions of the DTI, but all participants completed the writing assignments to control for therapeutic effects of expressive writing. After completing the writing assignments, participants completed self-report questionnaires while undergoing preparation for electroencephalography (EEG) recording. Participants then completed a brief resting task followed by the MTPT. Next, participants completed a go/no-go and cue exposure task (i.e., viewing neutral, threat, and cannabis images) before and after the MMST; we included these tasks as secondary outcomes; they will be the focus of separate manuscripts (see Macatee et al., 2018b, 2019 for analyses utilizing baseline EEG data). We assessed current craving for cannabis before and after the cue exposure task at both pre and post-MMST (see Figure S1 in supplement for diagram of task flow). At the end of the visit, we randomized participants to either the DTI (n=30) or HVC condition (n=30) using block randomization (https://www.randomizer.org). We did not inform participants of their condition assignment. We scheduled intervention session one approximately one week after the pre-treatment assessment.

3.3. Intervention sessions

During both treatment sessions one and two, a research assistant unblinded to condition assignment brought the participant to a private room where they completed the computerized treatment session. We scheduled treatment sessions one and two approximately one week apart. Upon completing treatment session two, we scheduled the post-treatment session approximately one week later.

3.4. Post-treatment assessment

The post-treatment assessment was nearly identical to the pre-treatment assessment, but participants did not complete the writing assignments at post-treatment. Three participants were not able to visit the lab for the post-treatment assessment and so MMST, craving, and MTPT data are missing for these participants; however, all three participants completed the self-report questionnaires online.

3.5. One- and four-month follow-up assessments

Online self-report questionnaire batteries at both follow-up assessments were identical to those completed at pre/post-treatment assessments. We emailed a link to the online questionnaire battery to the participants one and four months following the post-treatment assessment for the one- and four-month follow-up assessments, respectively. At the four-month follow-up only, an interviewer blind to condition assignment administered a CUD interview over the phone or in-person. Upon completion of the four-month follow-up, research staff debriefed participants and unblinded them to their condition assignment.

4. Data analytic plan

We determined the sample size based on the four-month follow-up cannabis use outcomes for which the intervention was expected to have the smallest effect size. Based on the effect sizes observed on alcohol use behavior at four-month follow-up in a RCT testing a brief intervention targeting a DI-related construct (i.e., anxiety sensitivity; Conrod et al., 2006), we expected a small-to-medium effect of intervention condition (effect size f = .20) on cannabis use outcomes at the four-month follow-up. We determined with a power analysis program (G-power; Faul et al., 2007) that 26 participants per condition would be required to test hypotheses with .80 power and α=.05 in a repeated measures framework. To allow for an attrition rate of 15%, which is consistent with the pilot study’s attrition rate (Macatee & Cougle, 2015), we increased the sample size to 30 participants per condition.

We used an intent-to-treat (ITT) approach, in which we included data from all randomized participants in the analyses, to test hypotheses. We employed a mixed model framework using maximum likelihood because this approach provides unbiased ITT estimates by including all available data and estimating missing data based on observed values. We used a model building approach to determine the best fitting model for each outcome. Models including random effects did not converge and therefore all the best-fitting models included fixed effects only. For each outcome, we compared models with differing covariance matrices for the within-subject residuals and we chose the covariance matrix resulting in the lowest Bayesian Information Criterion value. We modeled CUD criteria count and MPS scores with a negative binomial distribution due to the positive skew and high concentration of low values, which fit the data better than a normal distribution. Because we computed use frequency as a percentage, we modeled this outcome with a binomial distribution to ensure predicted values were always between zero and one. We modeled all other outcomes with a normal distribution, including MTPT persistence time after natural log-transforming the quit latency data (Macatee & Cougle, 2015).

To test the impact of condition on all outcomes except negative reinforcement craving, we entered condition (DTI or HVC), time (pre-tx, post-tx, one-month fu, or four-month fu), and their interaction as fixed effects, with time a within-subject factor and condition a between-subjects factor. We evaluated the significance of the condition*time interaction to test the hypothesized impact of condition on outcomes. To test the impact of condition on acute stress modulation of negative reinforcement craving, we entered condition, time, stress (pre-MMST, post-MMST), cue (pre-cue exposure, post-cue exposure), and their interactions as fixed effects. We evaluated the significance of the condition*time*stress and condition*time*stress*cue interactions to test the hypothesized impact of condition on stress potentiation of relief craving, both overall and as a function of cue exposure. We considered fixed effects significant at α=.05. We examined significant main effects and interactions with pairwise contrasts adjusted for multiple comparisons using the Bonferroni correction. We computed planned contrasts for each outcome to examine change from pre-treatment to post-treatment as well as from pre-treatment to both follow-ups in each treatment condition to evaluate initial treatment-elicited change and its maintenance, respectively (see Table 2). We conducted analyses using SPSS (Version 26 for PC).

Table 2.

Outcomes and effect sizes with 95% CIs over time by intervention group

	Pre-treatment				Post-treatment				FU1				FU2
	DTI		HVC		DTI		HVC		DTI		HVC		DTI		HVC
	M	SD	M	SD	M	SD	M	SD	M	SD	M	SD	M	SD	M	SD
DII	28.8	5.02	29.20	4.78	23.1	6.75	25.7	7.13	19.3	9.60	21.80	7.94	17.00	7.80	20.36	8.24
MTPT	3.59	1.31	3.73	1.71	2.74	1.57	2.39	1.71	--	--	--	--	--	--	--	--
Stress Modulation of Relief Craving	0.35	0.59	0.23	0.68	0.08	0.56	0.35	0.63	--	--	--	--	--	--	--	--
Cannabis Use Frequency	0.75	0.22	0.69	0.20	0.61	0.21	0.65	0.23	0.66	0.30	0.49	0.34	0.63	0.40	0.65	0.37
Cannabis Use Coping Motives	4.12	0.73	4.10	0.76	3.53	1.01	3.77	0.93	2.95	1.10	3.21	0.86	2.44	1.20	3.05	1.17
Cannabis Use-Related Problems	8.93	5.76	10.43	5.24	7.79	5.89	8.96	5.21	7.04	5.80	7.76	5.10	3.90	4.20	5.36	4.48
Cannabis Use Disorder Criteria	4.80	2.52	5.93	2.82	--	--	--	--	--	--	--	--	2.13	2.10	3.76	3.05
Effect sizes
	DTI			HVC
	Pre-to-Post	Pre-to-FU1	Pre-to-FU2	Pre-to-Post	Pre-to-FU1	Pre-to-FU2
DII	.95 [.56, 1.34]***	1.29 [.78,1.81]***	1.83 [1.04,2.61]***	.62 [.21, 1.03]**	1.22 [.63,1.81]***	1.55 [.64,2.47]***
MTPT	.53 [.17,. 89]**	--	--	.83 [.41, 1.25]***	--	--
Stress Modulation of Relief Craving	.22 [−.35, .79]	--	--	−.08 [−.59, .42]	--	--
Cannabis Use Frequency	.55 [.24, .86]***	.45 [.11, .79]*	.51 [.11, .91]*	.14 [−.24, .52]	.68 [.17, 1.19]**	.30 [−.34, .94]
Cannabis Use Coping Motives	.64 [.30, .98]***	1.17 [.71,1.62]***	1.73 [.95, 2.52]***	.39 [.07, .71]*	.94 [.35, 1.53]***	1.15 [.51,1.79]***
Cannabis Use-Related Problems	.20 [−.01, .41]	.38 [.12, .64]**	.68 [.23, 1.13]***	.31 [.05, .57]*	.45 [.15, .75]**	.66 [.15, 1.17]***
Cannabis Use Disorder Criteria	--	--	1.16 [.60, 1.72]***	--	--	.74 [.19, 1.28]**

Note. DII=Distress Intolerance Index. MTPT=Mirror-Tracing Persistence Task. HVC=Healthy Video Control. DTI=Distress Tolerance Intervention. FU1=One-month follow-up. FU2=Four-month follow-up. Pre=Pre-treatment. Post=Post-treatment.

^*p<.05

^**p<.01

^***p<.001.

We computed effect sizes (_pη²) for the fixed effects and their 90% CIs using the noncentral F distribution with an online SPSS script; we used 90% CIs because an F test is always a one-sided test and thus a 90% CI ensures zero will be excluded when the test is statistically significant (Smithson, 2001). We computed effect sizes (Cohen’s d) for the planned contrasts based on the means and standard errors derived from the mixed model analyses; we also cpmputed approximate 95% CIs for each effect size (Kline, 2004).

5. Results

5.1. Descriptives

Descriptives for pre-treatment demographics, DI, cannabis use, stress induction reactivity, and treatment credibility/expectancy are presented in Table 1. There were no significant differences between treatment groups on any demographic, DI, or cannabis use variable at the pre-treatment assessment (ps > .10). Pre-MMST subjective negative affect was significantly lower in the DTI compared to HVC group, t(58)= −3.25, p=.002, but, importantly for the stress-elicited craving analyses, negative affect reactivity to the MMST was not significantly different between groups (p=.25). However, individuals in the DTI group perceived the treatment rationale as more credible, t(50.03)=4.46, p<.001, and expected the treatment to produce greater improvement in their stress management abilities, t(57)=3.79, p<.001, than those in the HVC group. Thus, for all significant condition*time interactions, we included the main effects of credibility/expectancy and their interactions with time in separate follow-up analyses to ensure condition*time interactions were not attributable to differences in treatment credibility and/or expectancy. Planned contrasts examining each outcome’s initial pre- to post-treatment change and maintenance over the follow-up period are presented in Table 2.

Table 1.

Pre-treatment descriptives (N=60)

	DTI (n=30)	HVC (n=30)	t or χ2	p
	M (SD)	M(SD)
Demographics
Age	20.37 (2.95)	20.23 (1.63)	0.22	0.83
Gender			0.00	1.00
Male	9 (30.00%)	9 (30.00%)
Female	20 (66.67%)	20 (66.67%)
Other	1 (3.33%)	1 (3.33%)
Sexual Orientation			1.52	0.68
Heterosexual	24 (80.00%)	21 (70.00%)
Bisexual	5 (16.67%)	5 (16.67%)
Homosexual	0 (0.00%)	1 (3.33%)
Other	1 (3.33%)	2 (6.67%)
Ethnicity			3.79	0.29
White	16 (53.33%)	19 (63.33%)
Non-White Hispanic	9 (30.00%)	7 (23.33%)
Black	5 (16.67%)	2 (6.67%)
Other	0 (0.00%)	2 (6.67%)
Level of Education			1.20	0.55
High school diploma	9 (30.00%)	6 (20.00%)
Some college	18 (60.00%)	22 (73.33%)
Bachelor’s degree	3 (10.00%)	2 (6.67%)
Distress Intolerance
Distress intolerance index	28.80 (5.02)	29.20 (4.78)	−0.32	0.75
Mirror-tracing Persistence Task	3.59 (1.31)	3.73 (1.71)	−0.36	0.72
Lab Stressor
Pre-MMST negative affect	25.01 (16.27)	39.68 (18.59)	−3.25	0.00
Post-MMST negative affect	59.96 (24.40)	68.48 (22.33)	−1.41	0.16
MMST-elicited negative affect reactivity	34.96 (19.43)	28.80 (21.36)	1.17	0.25
Total # of correct responses	35.27 (11.57)	33 (13.01)	0.71	0.48
Cannabis Use
Age at first cannabis use	15.03 (2.17)	15.30 (1.78)	−0.52	0.61
Age at onset of regular cannabis use	17.17 (1.56)	17.23 (1.63)	−0.16	0.87
Total years of regular cannabis use	3.77 (2.67)	3.12 (1.45)	1.17	0.25
Percent Use Days in Past Month	0.75 (0.22)	0.69 (0.20)	1.11	0.27
Coping motives	4.12 (0.73)	4.10 (0.76)	0.09	0.93
Intention to change cannabis use	5.53 (2.51)	5.73 (2.30)	−0.32	0.75
Cannabis use-related problems (past month)	8.93 (5.76)	10.43 (5.24)	−1.06	0.30
CUD criteria count (lifetime)	6.63 (2.59)	6.93 (2.75)	−0.43	0.67
CUD criteria count (past 3 months)	4.80 (2.52)	5.93 (2.82)	−1.64	0.11
Relief Craving
Pre-MMST craving (pre-picture viewing)	4.76 (1.43)	4.56 (1.31)	0.57	0.57
Pre-MMST craving (post-picture viewing)	4.90 (1.51)	4.85 (1.33)	0.12	0.90
Post-MMST craving (pre-picture viewing)	5.33 (1.45)	5.03 (1.52)	0.78	0.44
Post-MMST craving (post-picture viewing)	5.02 (1.64)	4.71 (1.44)	0.78	0.44
Stress-elicited craving reactivity	0.35 (0.59)	0.23 (0.68)	0.72	0.48
Treatment Attitudes
Expectancy	0.50 (0.22)	0.27 (0.25)	3.79	<.001
Credibility	18.67 (4.53)	12.17 (6.46)	4.46	<.001

Note. DTI=Distress Tolerance Intervention. HVC=Healthy Video Control. MMST=Mannheim Multimodal Stress Test. CUD= Cannabis Use Disorder.

5.2. Effect of intervention on DI (Hypothesis 1)

For self-reported DI, the main effect of time, F(3,190)=27.51, p<.001, _pη² =.30 90% CI [.21, .38], was significant, which showed reductions in DI through the one-month follow-up that participants maintained at the four-month follow-up in both conditions (see supplemental material for analyses). Contrary to the hypotheses, the condition*time, F(3,190)=0.82, p=.49, _pη² =.01 90% CI [.00, .04], interaction was not significant, indicating comparable reduction in self-reported DI across the study period in both intervention conditions (see Figure 2).

Figure 2. Impact of interventions on perceived and behavioral indices of distress intolerance.

This figure depicts the impact of the Distress Tolerance Intervention and Healthy Video Control intervention on the Distress Intolerance Index score as well as the natural-log transformed quit latency on the Mirror-Tracing Persistence Task.

For behavioral DI, subjective stress significantly increased from pre- to post-MTPT at pre- and post-treatment, Fs > 67.24, ps < .001, indicating that participants experienced the task as distressing; the significant time*stress interaction, F(1,218)=4.73, p=.03, revealed smaller subjective stress reactivity to the MTPT at post-treatment, ΔStress M=20.13, SE=2.79, relative to pre-treatment, ΔStress M=27.74, SE=2.48, and this effect did not vary by treatment group, F(1,218)=1.20, p=.27. With regard to MTPT quit latency, the main effect of time, F(1,109)=30.18, p<.001, _pη² =.22 90% CI [.11, .32], was significant, which revealed a decrease in persistence time from pre- to post-treatment, t(109)=5.49, p<.001. Contrary to the hypotheses, the condition*time, F(1,109)=1.46, p=.23, _pη² =.01 90% CI [.00, .07], interaction was not significant, indicating comparable decreases in persistence time across the study period in both intervention conditions (see Figure 2).¹

5.3. Effect of intervention on cannabis outcomes (Hypothesis 2)

5.3.1. Stress induction manipulation check

Subjective stress significantly increased from pre- to post-MMST at both pre- and post-treatment, Fs > 110.47, p<.001, suggesting that the MMST was effective as a stress induction; further, nonsignificant time*stress, F(1,216)=3.16, p=.08, and condition*time*stress, F(1,216)=0.91, p=.34, interactions suggest that stress reactivity to the MMST did not differ over time or by group. However, we found a significantly greater improvement in behavioral performance during the MMST in the DTI compared to the HVC group, F(1,108)=4.35, p=.04, pη² =.04 90% CI [.00, .11]. Because we did not hypothesize treatment effects on behavioral performance during the MMST, these analyses are included in the supplemental material.

5.3.2. Acute stress modulation of relief craving

As expected, the main effect of stress was significant, F(1,427)=7.06, p=.008, _pη² =.02 90% CI [.00, .04], which revealed that relief craving increased from pre- to post-stressor, t(427)= −2.66, p=.008. The main effect of time was also significant, F(1,427)=52.42, p<.001, _pη² =.11, 90% CI [.07, .16], indicating decreased overall relief craving from pre- to post-treatment; the main effect of cue was not significant, F(1,427)=0.60, p=.44, _pη² =.00 90% CI [.00, .01], as was the cue*time interaction, F(1,427)=2.86, p=.09, _pη² =.01 90% CI [.00, .03], suggesting that viewing cannabis/threat images did not significantly affect relief craving at pre- or post-treatment. The cue*stress, F(1,427)=2.86, p=.09, _pη² =.01 90% CI [.00, .03], and time*stress, F(1,427)=0.12, p=.73, _pη² =.00 90% CI [.00, .01], interactions were not significant, but the time*stress*cue interaction, F(1,427)=5.78, p=.02, _pη² =.01 90% CI [.00, .04], was significant. Probing of acute stress modulation of relief craving reported before vs. after cue exposure revealed a significant stress-elicited increase in pre-cue-exposure relief craving at pre-treatment, t(427)= –3.38, p=.001, that became nonsignificant at post-treatment, t(427)= −0.84, p=.40, indicating a decrease in stress potentiation of pre-cue-exposure relief craving from pre- to post-treatment. In contrast, acute stress modulation of post-cue-exposure relief craving was nonsignificant at pre-treatment, t(427)= −0.06, p=.95, and post-treatment, t(427)= −1.59, p=.11. Contrary to our hypotheses (see Figure 3), the condition*time*stress, F(1,427)=0.58, p=.45, _pη² =.00 90% CI [.00, .01], and condition*time*stress*cue, F(1,427)=0.37, p=.54, _pη² =.00 90% CI [.00, .01], interactions were both nonsignificant. The condition*time interaction, F(1,427)=2.80, p=.095, _pη² = .01 90% CI [.00, .03], was also nonsignificant, though the marginal effect suggested a larger overall decrease in relief craving in the DTI, t(427)= −6.30, p<.001, vs. HVC, t(427)= −3.94, p<.001, condition.

Figure 3. Impact of interventions on cannabis use outcomes.

This figure depicts the impact of the Distress Tolerance Intervention and Healthy Video Control intervention on acute stress potentiation of relief cannabis craving, cannabis use frequency, cannabis use coping motives, cannabis use-related problems, and Cannabis Use Disorder criteria.

5.3.3. Cannabis use frequency

Consistent with our hypotheses, the condition*time, F(3,186)=3.58, p=.02, _pη² =.06 90% CI [.01, .10], interaction was significant (see Figure 3); in separate follow-up analyses including the effects of treatment credibility and expectancy, the condition*time interaction remained significant, Fs>3.45, ps<.019. Planned contrasts (adj. α = .017) revealed a significant decrease in use frequency during the treatment period compared to pre-treatment in the DTI, t(186)=3.71, p<.001, d=.55 95% CI [.24, .86], but not HVC group, t(186)=0.89, p=.37, d=.14 95% CI [−.24, .52]. There was an average 12.2% (95% CI[5.7%, 18.7%]) reduction in proportion of cannabis use days during the treatment period relative to pre-treatment in the DTI group, whereas the average reduction was 3% (95% CI [−3.6%, 9.7%]) in the HVC group.² Further, in the DTI group past-month use frequency was still numerically lower relative to pre-treatment at both the one-month, t(186)=2.13, p=.034, d=.45 95% CI [.11–.79], average 10.4% reduction in proportion cannabis use days (95% CI [.01%, 20%]), average 3.02 reduction in use days/month (95% CI [0.23, 5.80]), and four-month follow-up, t(186)=1.97, p=.050, d=.51 95% CI [.11–.91], average 13.1% reduction in proportion cannabis use days (95% CI [00.00%, 26.10%]), average 3.80 reduction in use days/month (95% CI [0.00, 7.57]). However, it is important to note that these latter two comparisons did not survive Bonferroni correction.

In contrast, in the control group past-month use frequency was significantly lower at the one-month follow-up relative to pre-treatment, t(186)=2.96, p=.004, d=.68 95% CI [.17, 1.19], average 16.5% reduction in proportion cannabis use days (95% CI [5.5%, 27.6%]), average 4.79 reduction in use days/month (95% CI [1.60, 8.00]), but this difference was no longer significant at the four-month follow-up, t(186)=0.94, p=.35, d=.30 95% CI [−.34, .94], average 7.5% reduction in proportion cannabis use days (95% CI [−8.3%, 23.3%]), average 2.18 reduction in use days/month (95% CI[−2.41, 6.76]).

5.3.4. Cannabis use coping motives

The main effect of time, F(3,190)=20.19, p<.001, _pη² =.24 90% CI [.15, .31], was significant, which showed reductions in coping motives through the one-month follow-up that were maintained at the four-month follow-up in both conditions (see supplemental material for analyses). Contrary to hypotheses, the condition*time, F(3,190)=0.90, p=.44, _pη² =.01 90% CI [.00, .04], interaction was nonsignificant, indicating comparable reduction in coping motives across the study period in both intervention conditions (see Figure 3).

5.3.5. Cannabis use–related problems

The main effect of time, F(3,190)=9.69, p<.001, _pη² =.13 90% CI [.06, .20], was significant, which showed reductions in cannabis use–related problems from pre-treatment to post-treatment that were maintained at one-month follow-up and further decreased at four-month follow-up (see supplemental material for analyses). Contrary to hypotheses, the condition*time interaction was not significant, F(3,190)=0.15, p=.93, _pη² =.00 90% CI [.00, .00], indicating comparable reduction in cannabis use–related problems across the study period in both intervention conditions (see Figure 3).

5.3.6. Cannabis use disorder criteria

The main effect of time, F(1,97)=24.13, p<.001, _pη² =.20 90% CI [.09, .31], was significant, which revealed a significant decrease in CUD criteria count from pre-treatment to four-month follow-up, t(97)=5.91, p<.001. Contrary to our hypotheses, the condition*time, F(1,97)=2.00, p=.16, _pη² =.02 90% CI [.00, .09], interaction was not significant, indicating a comparable reduction in CUD criteria count from pre-treatment to four-month follow-up in both intervention conditions (see Figure 3).

5.4. ΔDI Mediation of Intervention Effect on Cannabis Use Outcomes (Hypothesis 3)

Because the condition*time interaction was nonsignificant for self-reported DI, we did not conduct formal mediation analyses. However, we assessed associations between change in self-reported DI and cannabis use outcomes in the combined sample to test its theorized role in the maintenance of CUD. Partially consistent with our predictions, pre-treatment to one-month follow-up reductions in self-reported DI were significantly associated with greater decreases in coping motives for cannabis use, F(2,121)=8.77, p<.001, _pη² =.13, 90% CI [.04, .21], and CUD criteria, F(1,80)=11.97, p=.001, _pη² =.13, 90% CI [.04, .24], from pre-treatment through follow-up; effects for cannabis use frequency, F(2,116)=2.07, p=.13, _pη² =.04, 90% CI [.00, .09], and cannabis use–related problems, F(2,121)=2.08, p=.13, _pη² =.03, 90% CI [.00, .09], were nonsignificant (see supplemental material for greater detail).

6. Discussion

Overall, our results were inconsistent with our hypotheses. Pertaining to DI target engagement, self-reported DI significantly decreased in the DTI condition from pre-treatment through follow-up, replicating and extending the pilot RCT results (Macatee & Cougle, 2015), but in contrast to the pilot RCT we also observed comparable reductions in the control condition. Further, as in the pilot RCT, behavioral DI worsened from pre- to post-treatment, but in contrast to the pilot RCT the degree of worsening was not significantly different across conditions. Coping motives for cannabis use, cannabis use–related problems, and CUD criteria significantly decreased from pre-treatment through follow-up in the DTI group as expected, but we also observed comparable reductions in the control condition. We found evidence for a reduction in acute stress potentiation of relief craving assessed before but not after cue exposure in both conditions. Cannabis use frequency was the only outcome to demonstrate a significant intervention effect such that proportion of cannabis use days significantly decreased during the treatment period in the DTI but not HVC group; however, maintenance of use frequency reductions at both follow-ups in the DTI group were only significant using an unadjusted alpha. Finally, because of the absence of differential change in DI by group, we did not conduct formal mediation analyses, but exploratory analyses revealed that reductions in self-reported DI covaried with reductions in coping motives and CUD severity in the combined sample.

Reduction in cannabis use frequency during the treatment period in the DTI but not HVC group was the only hypothesized intervention effect to reach statistical significance, revealing a ~9.2% greater average reduction in proportion of cannabis use days during the treatment period in the DTI compared to HVC group. Although treatment trials often use complete abstinence as a clinical endpoint for CUD, a growing body of literature suggests that reductions in cannabis use frequency are clinically meaningful (e.g., associated with change in quality of life; Brezing et al., 2018). Use frequency reductions at both follow-ups in the DTI group were of comparable effect size but only significant using an unadjusted alpha, though these analyses were likely underpowered given the decreased sample size available at the follow-ups compared to the post-treatment assessment. Conversion of contrast estimates to number of cannabis use days in the past month revealed ~1.63 fewer cannabis use days/month at four-month follow-up relative to pre-intervention in the DTI compared to HVC group. The mean group difference at four-month follow-up compares favorably to a meta-analysis of brief interventions in cannabis users that found an average but nonsignificant reduction of ~0.55 cannabis use days/month at one- to three-month follow-ups (Halladay et al., 2019). It is also noteworthy that the RCTs included in Halladay and colleagues’ (2019) meta-analysis predominantly recruited cannabis users with lower use frequency/CUD symptoms compared to the current study’s frequent user sample, 91.7% of whom had a current CUD diagnosis. Additionally, in contrast to the majority of the interventions included in the meta-analysis, the DTI did not directly address cannabis use behavior but instead aimed to reduce the functional utility of use (i.e., negative affect reduction) to indirectly impact cannabis use outcomes (e.g., see Conrod et al., 2006 for a similar intervention targeting a related construct in alcohol users). Nevertheless, the ultimate clinical significance of these reductions in cannabis use is unclear. Further, reduction in cannabis use days/month favored the control group at the one-month follow-up (i.e., ~1.77 fewer cannabis use days/month in HVC vs. DTI group), and confidence intervals were wide at both follow-up assessments. Thus, the DTI’s relatively greater impact on cannabis use frequency should be interpreted cautiously pending replication.

The absence of significant differential intervention effects on DI and the majority of cannabis use outcomes was due to the unexpected efficacy of the control condition rather than a failure of the DTI. Indeed, we observed a large effect size reduction in self-reported DI from pre- to post-treatment in the DTI group, comparable to the effect size that the pilot study found in the DTI group (d=1.07 95% CI [.47, 1.68]; Macatee & Cougle, 2015). We observed an unexpected increase in behavioral DI (i.e., decreased quit latencies on the MTPT) from pre- to post-treatment in the DTI and HVC groups, which also occurred in the pilot study DTI group (Macatee & Cougle, 2015) but not in Bornovalova and colleagues’ (2012) RCT. Although speculative, behavioral DI may have increased in both groups in the current study due to the raffle reward being insufficiently motivating compared to contingencies that directly link payment to task performance (e.g., Bornovalova et al., 2012). It is important to note that we modified the DTI used in the current study from the version used in the pilot RCT in response to participant feedback, making it difficult to directly compare each version’s impact on DI. However, the current study’s DTI was only modified through additions that would theoretically improve the intervention, making it unlikely that the changes negatively impacted its efficacy.

There are several possible reasons for the unexpected improvements in self-reported DI and cannabis use outcomes in the HVC group. First, though the HVC intervention demonstrated significantly lower credibility and improvement expectancy compared to the DTI, both scores were greater than the scale minimums,³ suggesting that a placebo effect could still have plausibly contributed to the observed reductions in DI and cannabis outcomes. Second, it may be that other nonspecific treatment effects, such as repeated contact with kind and attentive research personnel in the context of a stress management treatment trial, were responsible for the HVC intervention’s efficacy. Third, participants in both conditions wrote in detail about two recent stressful events at the beginning of the pre-treatment assessment, which may have had some therapeutic benefits (Baikie & Wilhelm, 2005). Fourth, it is also possible that the HVC intervention was a more stringent control condition than we anticipated. For instance, the HVC intervention contains modules focused on the importance of exercise and sleep hygiene for improving stress management, two areas shown to be longitudinally associated with change in perceived DI (Kechter & Leventhal, 2019) and related constructs (LeBouthillier et al., 2015) as well as various cannabis use outcomes (e.g., craving, use frequency; Babson et al., 2015; Buchowski et al., 2011). Related to this possibility, Sabourin and colleagues (2016) also found a health education control intervention to be unexpectedly efficacious at reducing anxiety sensitivity, a construct related to DI (Allan et al., 2015), which the authors speculated may be due to increasing health behavior (e.g., exercise). Further research on attitude/behavioral change in response to the HVC intervention is needed to understand its unexpected efficacy found in the current study.

Exploratory analyses testing the associations between change in self-reported DI and change in cannabis use outcomes were partially consistent with extant research. Specifically, greater pre-treatment to one-month follow-up reductions in self-reported DI were associated with greater pre-treatment to follow-up reductions in CUD criteria and coping motives, replicating and extending previous cross-sectional findings (Bujarski et al., 2012; Farris et al., 2016) by demonstrating the relations between change in these constructs over time. However, associations with reductions in cannabis use frequency and use-related problems were nonsignificant. The nonsignificant association between change in self-reported DI and cannabis use frequency suggests that these processes may be orthogonal, which is consistent with some previous cross-sectional studies (e.g., Potter et al., 2011) but inconsistent with a prospective study in cannabis users attempting to quit (Hasan et al., 2015). We did not expect the absence of a significant relationship with change in use-related problems, though this may be due to the relatively small degree of change in this construct compared to the other outcomes. Indeed, there was a significant main effect of self-reported DI reductions on use-related problems such that smaller reductions in self-reported DI were associated with greater cannabis use–related problems across all assessment points (see supplemental material). These exploratory analyses provide support for the theorized relevance of self-reported DI to coping motives for cannabis use and CUD severity by demonstrating correlated change over time in these constructs. However, because these analyses were correlational and the intervention effect on self-reported DI was nonsignificant, we cannot make causal interpretations and further research is needed.

The current study had important limitations. First, due to missed control items and participant attrition at the follow-up assessments, particularly at the four-month follow-up, the resulting decreased sample size reduced statistical power, likely limiting our ability to detect significant, small-to-medium condition*time effects. However, it is also important to note that we tested multiple outcomes, and of these tests only one achieved statistical significance, raising the possibility that the sole significant intervention effect on cannabis use frequency may be spurious. Additionally, we did not conduct an a priori power analysis for the hypothesized condition*time*stress three-way interaction predicting relief craving; a post-hoc Monte Carlo simulation suggested that an N of 60 would have power of .35 to detect a medium effect size three-way interaction (_pη² =.06). Given the decreased sample size at the follow-up assessments, a large number of tested outcomes, and likely underpowered cannabis craving analysis, these results should be considered preliminary and suggestive pending replication. Second, given the comparable reductions in self-reported DI in both treatment conditions, the mechanism(s) driving the superior DTI effects on cannabis use frequency is not clear. Exploratory analyses revealed that the superior DTI effect on cannabis use frequency was not attributable to the DTI’s greater credibility or improvement expectancies, nor improved behavioral performance during the MMST (see supplemental material). Thus, future research should determine the mechanism through which the DTI impacted cannabis use frequency (e.g., via changes in mindfulness of distress/craving; Hsu et al., 2013). Third, we designed this study to evaluate the impact of a DI-targeted intervention in cannabis users with elevated DI, and so the current study’s results cannot inform the DTI’s impact on cannabis use outcomes in other cannabis user populations. Fourth, most of the participants were current students, with approximately half of the sample completing the pre-treatment through post-treatment assessments for course credit; thus, generalizability of findings to other cannabis user populations remains to be tested. Fifth, the requirement for abstinence from nicotine/caffeine the day of each lab visit may have elicited withdrawal symptoms that could have influenced results, though only 18.3% of the sample used nicotine once per week or more (see Macatee et al., 2019). Relatedly, the requirement for 24 hours of abstinence from cannabis may have produced withdrawal symptoms in some participants, which could have impacted craving and stress reactivity. Sixth, we assessed cannabis craving at only pre- and post-treatment, and so we do not know the durability of the craving reduction found in both groups. Relatedly, cue exposure did not significantly increase relief craving, which may be due to intermingling of cannabis, threatening, and neutral cues during the paradigm and/or a ceiling effect for relief craving measured after the stressor. Seventh, there was variability in the time frames of reference across some of the outcome measures at each assessment point (e.g., coping motives assessed “in general” at pre-treatment and “past month” at follow-ups), making strict interpretation of changes over time difficult.

The current study’s results suggest some possible clinical utility for the DTI. Compared to a control condition, the DTI produced greater reductions in cannabis use frequency during the intervention period in a sample of frequent cannabis users in which 91.7% had a current CUD diagnosis. The DTI’s brevity (i.e., ~ 2.5 hours total), portability, and absence of therapist involvement suggests that it may be a low-cost intervention with high potential for dissemination, especially since cannabis use frequency reduction occurred in a relatively severe sample despite its lack of an explicit focus on reducing cannabis use. However, participants did not maintain reductions in cannabis use frequency at the follow-up assessments after correction for multiple comparisons, and the large number of tested, nonsignificant outcomes warrants caution in interpretation of the DTI’s sole significant effect on cannabis use frequency. Although the current study’s results do not provide strong support for the DTI as a standalone intervention for reducing DI and associated cannabis use outcomes, future studies with larger sample sizes could explore its utility as an adjunctive intervention in the context of a larger SUD treatment package (e.g., see Bornovalova et al., 2012).

Highlights.

• High distress intolerance (DI) is associated with Cannabis Use Disorder (CUD) risk

• A brief computerized intervention for high DI was compared to a control condition

• Both interventions improved self-reported DI and cannabis use outcomes

• The DI intervention had a greater effect on cannabis use frequency

• The DI intervention may be a cost-effective adjunctive intervention