Evaluating cannabis use risk reduction as an alternative clinical outcome for cannabis use disorder.

Abstract

Objective:

Abstinence is rarely achieved in clinical trials for Cannabis Use Disorder (CUD). Cannabis reduction is associated with functional improvement, but reduction endpoints have not been established, indicating a need to identify and validate clinically meaningful reduction endpoints for assessing treatment efficacy.

Method:

Data from a 12-week double-blind randomized placebo-controlled medication trial for cannabis cessation (NCT01675661) were analyzed. Participants (N=225) were treatment-seeking adults (M=30.6 [8.9] years old, 70.2% male, and 42.2% Non-White) with CUD who completed 12 weeks of treatment. Frequency (days of use per week) and quantity (grams per using day) were used to define high, medium, and low risk levels. Anxiety and depression were assessed using the Hospital Anxiety and Depression Scale and cannabis-related problems were assessed using the Marijuana Problems Scale. GLM for repeated measures tested associations between magnitude of risk reduction and functional outcomes from baseline (BL) to end-of-treatment (EOT).

Results:

Cannabis risk levels were sensitive to reductions in use from BL to EOT for frequency- [X²=19.35, p=.004] and quantity-based [X²=52.06, p<.001] metrics. Magnitude reduction in frequency-based risk level was associated with magnitude decrease in depression [F=2.76, p=.043, n_p²=.04], anxiety [F=3.70, p=.013, n_p²=.05], and cannabis-related problems [F=8.95, p<.001, n_p²=.12]. Magnitude reduction in quantity-based risk level was associated with magnitude decrease in anxiety [F=3.02, p=.031, n_p²=.04] and cannabis-related problems [F=3.24, p=.023, n_p²=.05].

Conclusions:

Cannabis use risk levels, as operationalized in this study, captured reductions in use during a clinical trial. Risk level reduction was associated with functional improvement suggesting that identifying risk levels and measuring change in levels over time may be a viable and clinically meaningful endpoint for determining treatment efficacy.

Introduction

The U.S. Food and Drug Administration (FDA) sets the standard for therapeutic efficacy in clinical trials – historically considering abstinence as the primary outcome for substance use disorders (SUD). Cannabis cessation trials have struggled to achieve this standard. Behavioral treatments for cannabis use disorder (CUD) demonstrate only modest efficacy (Budney, Sofis, & Borodovsky, 2019; Sherman & McRae-Clark, 2016), and despite numerous clinical trials there is no FDA approved pharmacotherapy for CUD (Kondo et al., 2020). Scientific debate on clinical trial endpoints has reached a consensus that alternative non-abstinence endpoints are needed (Donovan et al., 2012; Fitzmaurice, Lipsitz, & Weiss, 2020; Kiluk, Fitzmaurice, Strain, & Weiss, 2019; Loflin et al., 2020), and the National Institute of Drug Abuse (NIDA) recently suggested that abstinence may be too high a bar and that clinically meaningful reduction endpoints can identify efficacious treatments for CUD (Cannabis Policy Research, 2018).

Identification of clinically meaningful reduction endpoints has been hampered by challenges in measuring cannabis consumption and in establishing a consistent association between reduced consumption and functional improvement. Reduced consumption cannot stand alone and must demonstrate prognostic value recognizable by individuals, their families, and third party payers (i.e. functional improvement) (McCann, Ramey, & Skolnick, 2015). Functional improvement can be measured by examining psychiatric symptoms, substance-related problems, quality of life measures, and other markers of health (e.g., sleep quality). Alcohol use disorder (AUD) treatment research has identified consumption risk levels and demonstrated an association between risk reduction and functional improvements in mental health and alcohol-related problems (Falk et al., 2010; Witkiewitz, Hallgren, et al., 2017). Consequently, absence of heavy drinking days (HDD) has become an acceptable FDA clinical endpoint for AUD trials (Witkiewitz, Roos, et al., 2017). Similarly, cocaine use disorder treatment research has begun to identify and validate consumption risk levels. A recent pooled data analysis categorized adults with cocaine use disorder into low-frequency and high-frequency use and found an association between risk level reduction and improvement on several subscales of the addiction severity index (Roos et al., 2019).

An ideal cannabis reduction endpoint would include frequency and quantity metrics, account for relative and absolute change, and engender manifest improvement in psychosocial functioning (Kiluk et al., 2019), as has been done in the alcohol field. Previous cannabis studies have operationalized reduction outcomes as proportion of days used (Brezing et al., 2018; Lintzeris et al., 2019), grams used per day (Tomko et al., 2018), and decreasing trajectory of use (i.e. negative slope) (Hser et al., 2017), but have demonstrated an inconsistent association between reduction and improved psychosocial functioning (Brezing et al., 2018; Hser et al., 2017; Lintzeris et al., 2019). In a pharmacotherapy trial for CUD, reduction in self-reported days of use was associated with improved quality of life, but reduction in grams of cannabis used was not (Brezing et al., 2018). Another trial found reduction in days of use was associated with improvements in depression and anxiety, but not quality of life (Hser et al., 2017). A third pharmacotherapy trial reported that the medication engendered 1.3 fewer days of use per week compared to placebo; however, no treatment condition differences were observed on psychosocial outcomes (Lintzeris et al., 2019). While these studies demonstrate the potential viability of cannabis reduction endpoints in clinical trials, research is needed to clearly define reduction metrics and to determine the magnitude of reduction that is clinically significant.

Classification of cannabis consumption into high-, medium-, and low-risk use may be one viable option for evaluating CUD reduction outcomes. Analyses of data from the largest clinical trial for alcohol use disorder (Anton et al., 2006) showed that World Health Organization (WHO) alcohol consumption risk levels were clearly associated with alcohol-related consequences and mental health outcomes (Witkiewitz, Hallgren, et al., 2017). Any reduction in risk level was associated with improvements in mental health and fewer alcohol related consequences; moreover, the magnitude of improvement was dependent on the magnitude of reduction (i.e., 3-level reduction > 2-level > 1-level). A secondary data analysis of a clinical trial of quetiapine for CUD categorized cannabis use into heavy, moderate, and light use based on days per week and found a small treatment effect on reduction in use level (Mariani et al., 2021), suggesting that reduction endpoints may identify treatment efficacy where abstinence endpoints may not. To our knowledge, this is the only publication that used categorical cannabis use risk groups as a reduction outcome; however, it did not test a quantity metric or investigate the association with functional improvement. Cannabis and alcohol both activate dopaminergic reward pathways incentivizing continued drug use and leading to the development of tolerance, withdrawal, and ultimately use disorder (Robinson & Berridge, 1993; 2000). Likewise, diagnostic criteria for substance use disorders are applicable across drugs of abuse (Budney, 2006). Therefore, we anticipate that cannabis risk reduction will engender functional improvement as has been demonstrated in the alcohol field.

Standardized cannabis reduction endpoints are needed as an alternative to abstinence for assessing treatment efficacy in clinical trials. Inconsistencies in cannabis reduction research are partly due to challenges in measuring cannabis consumption. Frequency of use appears to be more consistently associated with functional outcomes, but limitations in determining which metric (e.g. days of use per week, sessions per day) remain. Moreover, cannabis is consumed by numerous methods (e.g. combustion, ingestion), and in various forms (e.g. flower, edible, tincture) making accurate quantification of THC consumption a formidable challenge. Until extensive research is able to precisely quantify THC consumption, we are limited by the data at hand. Exploratory classification systems (as in Mariani and colleagues, 2021) provide a clinically relevant and informative means for testing consumption metrics, and can be replicated and modified as new data emerges. Due to the challenges in measuring cannabis consumption, particularly in measuring quantity, an important first step is to develop and test frequency-based and quantity-based risk classification systems separately. This will provide a preliminary understanding of their comparative efficacy in capturing and validating changes in use during treatment.

The current exploratory study categorized cannabis consumption into risk levels and conducted preliminary clinical validation using data from a multi-site clinical trial for cannabis use disorder (CUD). Two classification systems were developed, a frequency-based system (days per week) and a quantity-based system (grams per using day). High, medium, and low risk use was defined for each system and tested separately. Changes in risk level from baseline to end-of-treatment were measured and associations between magnitude of reduction and functional outcomes were examined. The relative associations between frequency- and quantity-based risk level reduction and functional outcomes were also explored.

Methods

Study Design

Data were from a National Drug Abuse Treatment Clinical Trials Network pharmacotherapy trial evaluating N-Acetylcysteine (NAC) for adults with CUD (Clinicaltrials.gov: NCT01675661) (McClure et al., 2014). Participants were enrolled at six diverse sites and were randomized to 1200mg NAC bid or placebo for 12 weeks. All participants received weekly medical management cessation counseling and abstinence-based contingency management (CM). As part of the counseling platform, clinicians worked with participants on skills to address cannabis use, and encouraged and reinforced reduction in use as a positive development; psychosocial aspects of care were not strictly abstinence-focused. Method details and primary results are reported elsewhere (Gray et al., 2017); there was no treatment effect of NAC on cannabis abstinence. All procedures were approved by the Institutional Review Board at each site and were conducted in accordance with the Declaration of Helsinki.

Participants

Participants (N=225) were treatment-seeking adults age 18–50 who met DSM-IV-TR criteria for cannabis dependence, provided a positive urine cannabinoid test at screening, and completed all 12 weeks of treatment. Participants were 70.2% Male; 57.8% White, 27.6% Black, 7.1% Multiracial; and were M[SD] 30.6 [8.9] years old. Participants reported using cannabis on an average of 5.8 days per week (SD=1.6) for the four weeks prior to randomization at an average consumption rate of 2.3 grams per using day (SD=3.0).

Procedures and Assessments

Screening and Diagnostic Assessment

The Mini International Neuropsychiatric Interview Plus (Sheehan et al., 1998) was used to assess current and lifetime psychiatric disorders. SUDs were assessed using the DSM-IV-TR checklist (First, Spitzer, Gibbon, & Williams, 1996). Participants provided a urine sample to test for drugs of abuse and for pregnancy in females.

Cannabis Consumption

Participants completed the Timeline Follow-Back (TLFB) (Sobell & Sobell, 1992) to assess daily cannabis consumption for 28 days prior to screening and throughout the study. For each day, participants reported cannabis use (yes/no), and if yes, the number of bowls, joints, bongs, blunts, edibles, or other methods of administration. Prior to initial TLFB procedures, participants weighed out a “typical” amount of use in grams for each method (Mariani, Brooks, Haney, & Levin, 2011; Tomko et al., 2018). Daily total grams were calculated by multiplying number of grams per method by the number of times used per day (e.g. 0.5g per bowl × 5 bowls=2.5g).

Cannabis Consumption Risk Levels

Risk levels were operationalized using two measures of cannabis consumption: 1) days of cannabis use per week (i.e. frequency), and 2) grams of cannabis use per using day (i.e. quantity). Frequency- and quantity-based classification systems were then developed by considering existing definitions in the literature, baseline cannabis use characteristics in our sample, face validity, and clinical relevance (i.e. allowing room for reduction, including reduction to abstinence). See Table 1 for final classifications.

TABLE 1.

Risk level definitions and descriptors.

	Days per week	Grams per using day	Group Descriptor
Abstinent	0	0	Abstainer
Low Risk	1–2	<= 1	Weekend user
Medium Risk	3–5	>1 to 3	Regular user
High Risk	6–7	>3	Daily user

Note: Days per week and grams per using day were computed for each week for the 4 weeks prior to randomization and for the final 4 weeks of the study. Four-week averages were then used for group classification.

Only one recently published clinical trial has defined categorical consumption levels and tested their efficacy as reduction endpoints. Mariani and colleagues (2021) used days per week to define heavy use (5–7 days), moderate use (2–4 days), and light use (0–1 day) groups and measured changes during a 12-week medication trial. A cross-sectional study (Mooney et al., 2018) also categorized cannabis use patterns using days per week but defined them slightly differently: abstinent (0 days), low-use (<= 3 days per week), and heavy use (4+ days per week). In our sample, during the four weeks prior to randomization (i.e. baseline) 62.8% used 6–7 days per week, 29.2% used 3–5 days per week, and 8% used 1–2 days per week; none were abstinent. Considering the literature, our data, face validity, and clinical relevance, we defined our groups as follows: High Risk (6–7 days) to capture daily or near daily use, Low Risk (1–2 days) to capture weekend only use, and Medium Risk (3–5 days) to capture less than daily use, but more than weekend use.

Given the inchoate nature of the cannabis reduction literature, we felt it was imperative to develop a quantity-based classification system as a first step. Neither Mariani and colleagues, nor Mooney and colleagues attempted to define a quantity-based system so we could not draw on existing literature in operationalizing the quantity-based system. With respect to our sample, during the baseline period, approximately 38.2% used one gram or less, 36.9% used more than one and up to three grams, and 24.9% used more than three grams per using day. Considering our data, face validity, and clinical relevance, we defined risk groupings as follows: Low Risk (<=1g), Medium Risk (>1 to <=3g), and High Risk (>3g).

Risk Reduction

Risk levels were used to create a risk reduction variable designed to capture the magnitude of change from baseline (BL) to end-of-treatment (EOT). First, weekly cannabis use data were aggregated for the four weeks prior to randomization (BL) and the final four weeks of treatment (EOT). Next, participants were classified into risk levels at BL and EOT based on Table 1. Finally, “risk reduction” was computed by subtracting EOT risk level from BL risk level for each participant (e.g. if Participant A met criteria for High Risk use at BL and Low Risk use at EOT, they would be categorized into the 2-Level reduction group). Risk reduction was computed separately for frequency and quantity-based metrics as a first step in developing and testing potential standardized reduction metrics. For supplemental analysis comparing frequency and quantity metrics, a binary variable “reduced by at least one risk level” was created for each metric

Anxiety, Depression, and Cannabis-Related Problems

Anxiety and depression symptoms were assessed using the Hospital Anxiety and Depression Scale (HADS) (Zigmond & Snaith, 1983). The HADS is a 14-item self-report measure which asks participants how they have been feeling during the past week (e.g. “I feel tense or wound up”). Responses are made on a 4-point Likert scale (0-Not at all, 1-From time to time/occasionally, 2-A lot of the time, 3-Most of the time) and summed to provide past-week anxiety and depression scores each ranging from 0–21. A score of 7 on either subscale is considered the clinical cutoff for a possible anxiety or depressive disorder. The HADS subscales demonstrated good internal consistency (Cronbach’s alpha) at BL (Anx: 0.82, Dep: 0.75) and EOT (Anx: 0.84, Dep: 0.82). Cannabis-related problems were assessed using the Marijuana Problems Scale (MPS) (Stephens, Roffman, & Curtin, 2000), a 19-item self-report measure which asks participants to identify problems experienced as a results of marijuana use during the past month (e.g. “To lose a job”). Responses are made on a 3-point Likert scale (0-No problem, 1-Minor problem, 2-Serious problem) and the total score indicates the number of items endorsed as a 1 or 2. The MPS demonstrated good internal consistency (Cronbach’s alpha) at BL (0.84) and EOT (0.87). Measurements taken at screening and week 12 were used in the current analysis.

Data Analytic Plan

Primary Analyses

Pearson’s Chi-Squared test was used to determine whether BL sample distribution differed significantly from EOT distribution. General linear models (GLM) for repeated measures were used to examine the associations between magnitude of risk reduction and functional outcomes. Effects of Time (BL vs. EOT), Risk Reduction (No Change vs. 1-Level vs. 2-Level vs. 3-Level), and Time × Risk Reduction on depression, anxiety, and cannabis-related problems were tested. Model assumptions of sphericity, independent observations, and normality were all met. Post-hoc pairwise comparisons were conducted to assess all relevant within group differences. Given the limited statistical power and exploratory nature of this study, we did not correct for multiple comparisons and focused the discussion on effect sizes. Separate models were run for frequency-based and quantity-based risk metrics. In order to focus on the effects of risk reduction, and for consistency with the alcohol (Witkiewitz, Hallgreen et al., 2017) and cocaine (Roos et al., 2019) field, participants who showed an increase in risk level (frequency n=4, quantity n=3) were included in the “No Change” group. All models adjusted for treatment, age, and gender effects. Treatment was retained as a design variable despite non-significant effects in our models. Age was retained because the primary paper reported treatment subgroup effects by age with younger participants showing better outcomes (Gray et al., 2017). Likewise, extensive evidence suggests important gender differences in CUD (Calakos et al., 2017; Cooper & Craft, 2018; Sherman et al., 2017). While these variables did not achieve statistical significance in our models, they did account for a notable amount of variance and therefore were retained.

Supplemental Analyses

Using identical GLM procedures, supplemental analyses investigated whether risk reduction conferred benefit independent of abstinence by comparing those who achieved abstinence with those who reduced but did not achieve abstinence. The following risk reduction groups were compared: No Change, 1-Level reduction, 2-Level reduction, and Achieved abstinence. See Supplemental Table 1.

To preliminarily compare the relative associations among frequency (F) and quantity (Q) metrics and functional outcomes, identical GLM procedures were conducted. A new variable was computed which categorized participants into one of the following groups: Reduced by at least one F risk level, Reduced by at least one Q risk level, Reduced by at least one F and one Q risk level, Did not reduce either F or Q. Changes in depression, anxiety, and cannabis-related problem score from BL to EOT were then compared across groups.

Results

Risk Reduction from Baseline to End-of-Treatment

The frequency-based risk distribution at BL was significantly different than the distribution at EOT [X² (6, N=225)=19.35, p=.004; see Table 2] with a large effect size (Cramer’s V=.21). Similarly, the quantity-based risk distribution differed at BL as compared to EOT [X² (6, N=225)=52.06, p<.001; see Table 3] with a large effect size (Cramer’s V=.34). Results indicate a significant shift from higher to lower risk level from BL to EOT. As an example, using the frequency-based classification system results indicate that 168 participants were classified as “high-risk” at BL, while 60 were classified as “high-risk” at EOT, a sample-wide decrease of 48% for the “high-risk” classification. Using the quantity-based system, results indicate that 56 participants were classified as “high-risk” at BL, while 10 were classified as “high-risk” as EOT, a sample-wide decrease of 21% for the “high-risk” classification.

TABLE 2.

Risk Group Change from BL to EOT based on Days of Use Per Week.

	END OF TREATMENT
BASELINE	Abstinent	Low Risk	Medium Risk	High Risk	Total (BL)
Abstinent	0 (0.0%)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)
Low Risk	6 (54.5)	4 (36.4)	0 (0.0)	1 (9.1)	11 (4.9)
Medium Risk	19 (41.3)	17 (37.0)	7 (15.2)	3 (6.5)	46 (20.4)
High Risk	50 (29.8)	37 (22.0)	25 (14.9)	56 (33.3)	168 (74.7)
Total (EOT)	75 (33.0)	58 (25.8)	32 (14.2)	60 (26.7)	225 (100.0)

Note: Total (BL) reflects row total and percentage of total sample; Total (EOT) reflects column total and percentage of total sample. Results indicate a significant shift in risk group distribution from BL to EOT X² (6, N=225) = 19.35, p < .01. For example, 41.3% of participants who were in the Medium Risk group at BL moved to Abstinence group at EOT; 22.0% of participants in the High Risk group at BL moved to Low Risk group at EOT; and 36.4% of Low Risk participants at BL remained in the Low Risk at EOT.

TABLE 3.

Risk Group Change from BL to EOT based on Grams per Using Day.

	END OF TREATMENT
BASELINE	Abstinent	Low Risk	Medium Risk	High Risk	Total (BL)
Abstinent	0 (0.0%)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)
Low Risk	33 (38.8)	50 (58.8)	2 (2.4)	0 (0.0)	85 (37.8)
Medium Risk	24 (28.6)	35 (41.7)	24 (28.6)	1 (1.2)	84 (37.3)
High Risk	18 (32.1)	14 (25.0)	15 (26.8)	9 (16.1)	56 (24.9)
Total (EOT)	75 (33.0)	99 (44.0)	41 (18.2)	10 (4.4)	225 (100.0)

Note: Total (BL) reflects row total and percentage of total sample; Total (EOT) reflects column total and percentage of total sample. Results indicate a significant shift in risk group membership from BL to EOT, X² (6, N=225) = 52.06, p < .001. For example, 38.8% of participants who were in the Low Risk group at BL moved to the Abstinence group at EOT; 41.7% of participants in the Medium Risk group at BL moved to the Low Risk group at EOT; and 16.1% of High Risk participants at BL remained in the High Risk group at EOT.

Association between Risk Reduction, Psychiatric Symptoms and Cannabis-Related Problems.

Results are organized by outcome measure. Effect sizes are reported as partial eta-squared (n_p²) and interpreted as small=.01, medium=.06, and large=.14. All findings are summarized in Figure 1.

FIGURE 1. Associations between risk reduction, anxiety, depression, and cannabis-related problems.

Note: N=211. Panel A represents three models testing the frequency-based risk reduction metric (days per week) on functional outcomes. Panel B represents three models testing the quantity-based reduction metric (grams per using day) on functional outcomes. Effect sizes (n_p²) for within group comparisons are included and asterisks represent statistical significance at *p<.05, **p<.01, ***p<.001.

Depression

Frequency.

The effect of Time on depression scores from BL to EOT [F(1,204)=1.59, p=.21] was not statistically significant. There was a statistically significant effect of Risk Reduction on depression [F(3,204)=5.22, p=.002, n_p²=.07), however, a statistically significant interaction effect [Time × Risk Reduction, F(3,204)=2.78, p=.042, n_p²=.04) indicated that depression scores decreased as a function of risk reduction. Post-hoc pairwise comparisons indicated a significant decrease in depression scores from BL to EOT in the 1-Level reduction group (BL M=4.59, SE=0.48; EOT M=3.05, SE=0.50; p=.005; n_p²=.04) and the 3-Level reduction group (BL M=3.28, SE=0.47; EOT M=1.18, SE=0.48; p<.001; n_p²=.07), but no statistically significant effect in the 2-Level group (p=0.07) or the No Change group (p=.73).

Quantity.

There was no statistically significant effect of Time [F(1,204)=3.23, p=.07) or Risk Reduction [F(1,204)=2.38, p=.07) on depression score. The Time × Risk reduction interaction also did not achieve statistical significance [F(3,204)=2.15, p=.09, n_p²=.03], though depression scores decreased numerically over time as a function of risk reduction with a small to medium effect size.

Anxiety

Frequency.

A statistically significant effect of Time on anxiety score [F(1,204)=9.78, p=.002; n_p²=.05] indicated an overall reduction in anxiety across groups, and a significant effect of Risk Reduction [F(3,204)=4.00, p=.008, n_p²=.06) indicated between group differences on overall anxiety scores. A significant interaction effect [Time × Risk Reduction, F(3,204)=3.78, p=.011, n_p²=.05), indicated that anxiety scores decreased from BL to EOT as a function risk reduction. Post-hoc pairwise comparisons revealed statistically significant decreases in all reduction groups: 1-Level (BL M=6.53, SE=0.60; EOT M=5.14, SE=0.59; p=.012), 2-Level (BL M=6.89, SE=0.57; EOT M=5.22, SE=0.55; p=.001), and 3-Level (M=5.78, SE=0.58; EOT M=3.53, SE=0.57; p<.001); but not in the No Change group (p=.92). Effect sizes for the 1-Level, 2-Level, and 3-Level reduction groups (n_p² = .03, .05, .08, respectively) indicated that greater magnitude of risk reduction demonstrated an increasingly robust effect on anxiety reduction.

Quantity.

There was a statistically significant effect of Time on anxiety score [F(1,204)=11.95, p=.001, n_p²=.05], but no statistically significant effect of Risk Reduction (p=.08). A significant interaction effect [Time × Risk Reduction, F(3,204)=2.95, p=.034, n_p²=.04] indicated anxiety scores decreased over time as a function of risk reduction. Post-hoc pairwise comparisons indicated a significant decrease in anxiety scores from BL to EOT among participants in the 1-Level (BL M=6.96, SE=0.46; EOT M=5.56, SE=0.46; p=.001; n_p²=.05), 2-Level (BL M=6.94, SE=0.68; EOT M=4.68, SE=0.68; p<.001; n_p²=.06), and 3-Level reduction group (BL M=5.11, SE=0.93; EOT M=2.75, SE=0.94; p=.007; n_p²=.04), but not among participants who showed No Change (p=.40).

Cannabis-Related Problems

Frequency.

Cannabis-related problems decreased significantly from BL to EOT across groups [F(1,204)=19.25, p<.001, n_p²=.09] and risk reduction groups differed significantly on overall cannabis-related problems [F(3,204)=3.10, p=.03, n_p²=.04). A significant interaction effect [Time × Risk Reduction, F(3,204)=8.77, p<.001, n_p²=.11), indicated that cannabis-related problems decreased from BL to EOT as a function of risk reduction. Post-hoc pairwise comparison revealed significant decreases in cannabis-related problems in all groups (all p-values < .001); effect sizes were large and greater magnitude of risk reduction demonstrated an increasingly robust effect on cannabis-related problems (n_p²: No-change=.11, 1-Level=.18, 2-Level=.31, 3-Level=.36).

Quantity.

Cannabis-related problems decreased significantly from BL to EOT across groups [F(1,204)=21.39, p<.001, n_p²=.10), while overall cannabis-related problems did not differ between risk reduction groups [F(3,204)=2.41, p=.07). However, a significant interaction effect [Time × Risk Reduction, F(3,204)=3.34, p=.020, n_p²=.05), indicated cannabis-related problems decreased over time as a function of risk reduction. Post-hoc pairwise comparison indicated a significant decrease in cannabis-related problems from BL to EOT among participants in all four risk reduction groups (all p’s < .001). All effect sizes were large (n_p²=.16 to .29).

Reduction to abstinence

Supplemental analyses yielded similar results (Supplemental Figure 1). Frequency- and quantity-based risk reduction were associated with improvements in anxiety, depression, and cannabis-related problems, and for anxiety and cannabis-related problems, greater magnitude of risk reduction predicted greater magnitude of functional improvement. Findings suggest that while abstinence may engender slightly greater improvement on anxiety and cannabis-related problems, any reduction in risk level was associated with functional improvement even if abstinence was not achieved.

Comparing cannabis frequency and quantity metrics

The distribution of participants into the four comparison groups was as follows: No Change in F or Q, n=43 (19.1%); Reduced F only, 43 (19.1), Reduced Q only, 28 (12.4), Reduced both F and Q, 111 (49.3). Statistically significant interaction effects of Time × Outcome were found for anxiety and cannabis-related problems, but not depression. Post-hoc comparisons suggested significant functional improvement in the F/Q and F only reduction groups, but not the Q only or No-Change groups. Effect sizes were largest for F/Q reduction group, followed by F only group. Full results are illustrated in Supplemental Figure 2.

Discussion

The current study found that cannabis use risk levels are sensitive to reductions in cannabis use during a clinical trial and that risk reduction is associated with improvements in depression, anxiety, and cannabis-related problems. Moreover, magnitude of risk reduction was associated with magnitude of improvement. Findings provide initial validation for cannabis consumption risk levels as a clinically-meaningful reduction endpoint, though replication and further validation are needed.

The risk classification system developed in this study captured reductions in cannabis use, as a significant proportion of individuals moved from higher to lower risk levels following treatment. This risk reduction system offers one of many models that can be tested in other samples. There is a growing literature on cannabis use reduction, but standardized endpoints, which are needed to compare treatment efficacy across studies (Lee et al., 2019; Loflin et al., 2020), have not been established. As in Mariani and colleagues (Mariani et al., 2021), we defined risk levels using days per week. However, we also tested a quantity risk metric based on grams per using day. An ideal reduction endpoint would account for frequency and quantity simultaneously. For example, heavy drinking days is an accepted clinical endpoint in alcohol trials and includes frequency (number of days) and quantity (heavy drinking) in one measurement. More research is needed to identify the most sensitive measures of “high risk” cannabis use in terms of quantity (i.e., grams, sessions, mgTHC per day) and frequency (i.e. days per week or month), and on how to combine them into a reliable reduction endpoint. Once independent metrics are validated and the most sensitive measures of frequency and quantity are identified, future research can develop and test an integrated approach.

Identification of clinically-meaningful reduction endpoints requires establishing an association between reduced consumption and functional improvement. Risk level reduction engendered improvements in anxiety, depression, and cannabis-related problems. Magnitude of frequency-based risk reduction was associated with magnitude of decrease in anxiety, depression, and cannabis-related problems. Magnitude of quantity-based risk reduction was associated with magnitude of decrease in anxiety and cannabis-related problems, but not depression. Within-group comparisons of risk reduction groups (i.e., 1-level, 2-level, 3-level) generally indicated increasingly robust effect sizes relative to increasing risk level reduction. This was particularly true for cannabis-related problems. The association between reduced consumption and decreased cannabis-related problems is of particular import since decreased drug-related consequences is arguably the most clinically meaningful outcome in SUD treatment trials (Kiluk et al., 2019).

The decision to develop and test frequency and quantity metrics separately was thoughtfully considered. The field of cannabis harm reduction is in its early stages and existing literature on the effects of cannabis reduction on functional outcomes is mixed (Brezing, et al., 2018; Hser et al., 2017; Lintzeris et al., 2019). Our results suggest that reductions in frequency and quantity-based risk levels are associated with decreased anxiety and marijuana related problems, but only reduction in frequency was associated with decreased depressive symptoms. Further, supplemental analyses suggest that participants who reduced at least one F level (but not Q) showed functional improvement while those who reduced at least one Q level (but not F) did not; participants who reduced at least one F level and one Q level showed the greatest improvement. While preliminary, these data suggest that frequency, as defined in our classification system, was a more sensitive measure of the effects of cannabis reduction on functional outcomes compared to quantity. That said, our metric of quantity has not been adequately validated, and other quantity metrics may prove more representative of meaningful cannabis consumption patterns.

Perhaps the most critical challenge in risk reduction research is developing a Standard Cannabis Unit (SCU). The alcohol field uses standard drinks to measure harm reduction endpoints. Per the National Institute on Alcohol Abuse and Alcoholism, one standard drink is defined as “14 grams of pure alcohol”, which is found in “12 ounces of regular beer (typically 5% alcohol), five ounces of wine (typically 12% alcohol), and 1.5 ounces of distilled spirits (typically 40% alcohol).” Cannabis is consumed in various forms (e.g. combusted flower, edibles, vaporized oils, tinctures, etc.) and the chemical composition (i.e. THC:CBD) within and between these forms differs widely. Accurate measurement of THC consumption via each route of administration (i.e. quantity THC consumed) is critical to developing a SCU. In turn, a SCU is critical to developing a reliably sensitive combined index that includes frequency and quantity metrics (i.e. SCUs per day). This limitation does not obviate the need for studies that quantify cannabis consumption and measure the effects of quantity reduction on functional outcomes. However, we found it premature to combine frequency and quantity metrics at this time. An alternative approach to developing a risk classification system is to use latent profile analysis. Latent profile analysis would provide empirically-derived data that could be used to identify cut points. This approach warrants investigation because it considers the data more objectively than our clinically-driven approach; however, it may have limited clinical utility due the generally homogenous use characteristics of treatment-seeking samples.

An important theoretical consideration in developing reduction endpoints is how to define treatment “success” and what to do with the abstinence outcome? Abstinence could be subsumed in a reduction classification system (as in our primary analyses) or separated out as its own endpoint (as in our supplemental analyses). Together, our results suggest that whether abstinence is subsumed or retained independently, any reduction in risk level was associated with functional improvement and therefore could be considered a treatment “success.” Future studies could adopt binary endpoints (“at least 1-level” or “at least 2-level” reduction), similar to the alcohol (Witkiewitz et al., 2017) and cocaine fields (Roos et al., 2019), which more easily permits comparison of treatment efficacy across studies, while retaining a risk reduction framework. This approach may garner support from governing bodies like the FDA. Moreover, cannabis users favor reduction goals, show better engagement in reduction versus quit attempts, and show better outcomes when personal goals align with study-defined treatment outcomes (Hughes, et al., 2016; Lozano, et al., 2006).

A final consideration is treatment durability. A recent 10-year follow-up study of Project MATCH data found that psychosocial functioning at 3-years post-treatment was more predictive of healthy psychological functioning at 10-years post-treatment, irrespective of consumption level (Witkiewitz et al., 2020). Importantly, the study also found that abstinence at 3-years did not predict better psychological functioning at 10 years. These data suggest a shift from focusing exclusively on abstinence as the sole marker of success may be prudent. A recent CUD clinical trial comparing a fixed-dose psychosocial intervention with a PRN intervention followed participants for 34 months and found higher abstinence rates in the fixed-dose condition at the first follow-up (4 months), but abstinence rates converged at all subsequent follow-up assessments (Stephens et al., 2020). Given these data, combined with the overlap of withdrawal symptomology (Budney, 2006) and generally poor treatment outcomes for both drugs when defined as abstinence only (Henssler et al., 2020; Sherman & McRae-Clark, 2016), we might predict similar associations between psychological functioning and long-term outcomes among “successful” cannabis reducers, however that construct is ultimately defined.

The current study has limitations. First, the original data was not powered to test the current study effects; interpretation of results is considered preliminary and analyses should be replicated with larger data sets. Second, cannabis use data was self-report and thus subject to recall bias. Third, risk levels were operationalized based on limited existing literature, sample data distribution, and clinical expertise. This methodology yielded different sample distributions for frequency and quantity metrics, suggesting more sensitive measures are needed. Although gram estimation has shown incremental validity over and above days per week (Tomko et al., 2018), a major challenge to the field concerns the development of a “standard cannabis unit” which would provide a more accurate assessment accounting for chemical composition and route of administration (Freeman & Lorenzetti, 2020). Finally, generalizability is limited to adults with CUD who are psychiatrically stable with few comorbidities. Despite these limitations, study findings are novel and provide a framework from which to expand the field of cannabis reduction research.

Abstinence has been the primary outcome in CUD clinical trials for the past three decades. The current study presents a potential alternative for measuring treatment efficacy. Findings provide initial validation of our risk classification system by demonstrating an association between risk reduction and functional outcomes. Notably, participants who reduced but did not achieve abstinence still showed significant functional improvement, suggesting that risk level reduction may be a valid endpoint for assessing treatment efficacy. Replication and further validation with larger and diverse clinical samples, development and testing of alternative reduction classification systems, and continued investigation into the most sensitive measures of cannabis consumption are critical.

Public Health Significance:

Cannabis use risk levels are measurable, and reduction in risk level is associated with functional improvement during treatment. Given increasing access to cannabis worldwide, risk reduction approaches may be critical to reduce the public health burden of CUD.