This article originally appeared at https://pubmed.ncbi.nlm.nih.gov/33309538/
Authors: Randi Melissa Schuster 1 , Kevin Potter 2 , Erin Lamberth 2 , Natali Rychik 2 , Maya Hareli 3 , Sophia Allen 2 , Hannah C Broos 4 , Audrey Mustoe 2 , Jodi M Gilman 5 , Gladys Pachas 2 , A Eden Evins 5
- PMID: 33309538
- PMCID: PMC7882030
- DOI: 10.1016/j.pnpbp.2020.110205
Abstract
Objective.
Cannabis and alcohol use are correlated behaviors among youth. It is not known whether discontinuation of cannabis use is associated with changes in alcohol use. This study assessed alcohol use in youth before, during, and after four weeks of paid cannabis abstinence.
Methods.
Healthy, non-treatment seeking, cannabis users (n=160), aged 14-25 years, 84% of whom used alcohol in the last month, were enrolled for a 4-week study with a 2-4 week follow-up. Participants were randomly assigned to four weeks of either biochemically-verified cannabis abstinence achieved through a contingency management framework (CB-Abst) or monitoring with no abstinence requirement (CB-Mon). Participants were assessed at baseline and approximately 4, 6, 10, 17, 24, and 31 days after enrollment. A follow-up visit with no cannabis abstinence requirement for CB-Abst was conducted after 2-4 weeks.
Results.
Sixty percent of individuals assigned to the CB-Abst condition increased in frequency and quantity of alcohol consumption during the 4-week period of incentivized cannabis abstinence. As a whole, CB-Abst increased by a mean of 0.6 drinking days and 0.2 drinks per day in the initial week of abstinence (p’s<0.006). There was no evidence for further increases in drinking frequency or quantity during the 30-day abstinence period (p’s>0.53). There was no change in drinking frequency or quantity during the 4-week monitoring or follow-up periods among CB-Mon.
Conclusions.
On average, four weeks of incentivized (i.e., paid) cannabis abstinence among non-treatment seeking youth was associated with increased frequency and amount of alcohol use in week 1 that was sustained over 4 weeks and resolved with resumption of cannabis use. However, there was notable variability in individual-level response, with 60% increasing in alcohol use and 23% actually decreasing in alcohol use during cannabis abstinence. Findings suggest that increased alcohol use during cannabis abstinence among youth merits further study to determine whether this behavior occurs among treatment seeking youth and its clinical significance.
Keywords: Marijuana, Cannabis, Abstinence, Alcohol, Substitution, Youth, Contingency Management
INTRODUCTION
Cannabis is currently the second most widely used substance by youth in the United States,1 and rates of use may increase further as perceptions of harm2,3 and barriers to access decrease through the commercialization of legalized recreational cannabis.4–6 High rates of cannabis use are concerning because use is associated with many negative outcomes in youth, including insults to cognitive functioning,7–9 mood,10–12 long-term achievement,13–15 and driving safety.16–18 Therefore, there has been a surge of interest in the last decade to elucidate strategies for reducing teen cannabis use.19–21 It is important to know, however, whether cannabis abstinence may have unintended negative effects, such as increased use of other substances. This study aims to investigate, in an experimental design in a non-clinical sample, the effects of cannabis abstinence on subsequent changes in alcohol use, the most widely used substance by youth in the United States1 that is also associated with high rates of negative consequences particularly with early age of initiation.22–25
There is controversy regarding whether cannabis abstinence is associated with alcohol substitution,26 with studies showing increases,27–30 decreases,31,32 and no change33,34 in frequency and quantity of alcohol consumption following cessation of cannabis use. Studies reporting increased use include a report by Allsop and colleagues, who reported that two weeks of cannabis cessation in a group of 45 non-treatment seeking youth was associated with an initial mean increase of 8 standard alcohol units per week (38% increase), followed by a decrease in alcohol use at one-month follow-up that corresponded to when the majority of participants (87%) had resumed using cannabis.30 Similarly, Peters and colleagues found a 15% increase in alcohol use among non-treatment seeking young adults with two weeks of cannabis abstinence, particularly among those with a past alcohol use disorder.29 It has been postulated that increased alcohol use during early cannabis abstinence may represent efforts to mitigate cannabis withdrawal symptoms.35 It is also plausible that youth who use cannabis carry increased non-specific genetic liability for substance use,36 and thus may seek alcohol when cannabis is not available.
The literature is less robust with regard to alcohol decreases with cannabis abstinence.31,32 This may be due, in part, to the fact that studies were conducted over a much wider time frame, with measurement occurring across periods months32 or years,31 rather than during the acute period of cannabis withdrawal as has been the case with studies citing increases in alcohol use with cannabis cessation.29,30 Cannabis abstinence may impact peer affiliation, which may be one possible driver of subsequent decreases in alcohol use. Among youth, affiliation with substance using peers is one of the most robust predictors of alcohol and other drug use37–40 and is likely to contribute to, at least in part, exposure to and initiation of multiple addictive substances.38,39 Therefore, any reorganization of social networks that may accompany cannabis abstinence41 could result in decreased alcohol use due to decreased exposure and access to substances in general, as well as decreased peer pressure to use substances. Cannabis use may also increase sensitivity to the rewarding effects of other drugs,42 and this effect may be undone with cannabis abstinence, thereby resulting in decreased alcohol use.
The heterogeneity of findings regarding the impact of cannabis abstinence on alcohol use may be due in part to methodological and sampling limitations, including lack of prospective, experimental trials,28 lack of biochemical verification of cannabis abstinence, measurement of alcohol changes outside of the acute cannabis abstinence period,28,31,32 and exclusion of those with past or current heavy and/or problematic alcohol use.27,29,30,32–34 Additionally, most prior studies on changes in alcohol use with cannabis abstinence have enrolled only adults27–29,32–34 and/or treatment-seeking cannabis users.27,32,34 Effects are also important to evaluate in youth because youth are particularly susceptible to adverse consequences from cannabis use due to ongoing neuromaturational vulnerability.43,44 Further, with increased public discussion around negative health effects of early cannabis exposure, more youth may reduce or abstain from cannabis use.45 Those who do so are most likely to engage in a spontaneous quit attempt46 and/or are most inclined to seek support in settings such as school rather than formally presenting for treatment per se.47
The current study employed an experimental design with random assignment to cannabis abstinence or use as usual in an attempt to isolate the impact of one month of sustained cannabis abstinence on concurrent changes in alcohol use among a sample of non-treatment seeking cannabis using youth engaged in a paid quit attempt.
METHODS
Participants
Participants (N = 160) were non-treatment seeking medically healthy youth, aged 15-25 years (62% White; 52% male). Participants were recruited via community advertisements, onsite school screening surveys at middle and high schools in the Greater Boston area, and peer referrals. Eligibility criteria included English fluency, no history of severe developmental delays, at least weekly cannabis use (M = 4.3 days per week, SD = 2.1), cannabis use in the week prior to their baseline visit, and willingness to abstain from cannabis use for 30 days. Participants were not excluded on the basis of current or past alcohol use.
Procedures
Enrollment occurred between July 2015 and March 2020. All study procedures were approved by the Partners Healthcare Human Subjects Committee. A detailed description of procedures has been described previously (NCT03276221).48 The parent study aimed to evaluate the relationship between cannabis abstinence and changes in cognition in a sample of young adult and adolescent cannabis users, half of whom are randomized to 4 weeks of cannabis abstinence, with the expectation that cognition would improve more with cannabis abstinence compared to continued use. Prior to beginning study procedures, written informed consent was obtained for all participants ages 18 years and older, and written parental consent and participant assent were obtained for participants under the age of 18 years.
At baseline, cannabis users were randomized to four weeks of cannabis abstinence achieved through a contingency management framework (CB-Abst) or monitoring with no abstinence requirement (CB-Mon). The contingency management procedure consisted of a voucher-based program for four weeks of continuous cannabis abstinence. At baseline, participants completed an abstinence contract with a study staff member that clearly delineated the behavior to be monitored, schedule of monitoring, and contingencies.49,50 Participants randomized to CB-Abst were then instructed to refrain from using cannabis for the next month; CB-Abst were not required to abstain from any substance other than cannabis during the four-week abstinence protocol. During this month, CB-Abst earned incentives based on a two-track system for attendance and biochemically-verified continuous abstinence with static denominations for attendance and escalating denominations for abstinence to encourage achievement of longer periods of continuous abstinence. After one month, CB-Abst were no longer required to abstain from cannabis use and were paid for attendance only for the follow-up visit. The first 35 participants earned $585 for 30 days of abstinence with full attendance ($405 for abstinence and $180 for full attendance). The payment schedule was reduced by approximately 30% for the remaining participants ($315 for 30 days of abstinence and $105 for full attendance). CB-Mon were not asked to abstain from cannabis and provided urine samples for toxicology on the same schedule as those assigned to CB-Abst. CB-Mon received escalating payments for attendance, totaling $270 for full attendance. Incentives were distributed via reloadable credit cards through Clinical Trials Payer (CT Payer), a web-based platform that facilitates HIPAA and HITECH safe clinical trial payments. Payments were distributed on the day of the visit for attendance and upon receipt of the quantitative urinalysis results confirming abstinence (for CB-Abst; described below).
Randomization was stratified by sex (male vs female), age (13-16 years vs ≥ 17 years), and average frequency of cannabis use (1 day per week vs ≥ 1 day per week). CB-Abst and CB-Mon completed in-person study visits at baseline (pre-abstinence for CB-Abst) and, on average, 4, 6, 10, 17, 24 and 31 days after enrollment. A follow-up visit was also conducted two to four weeks after completion of the initial four-week protocol (M = 20.8 days after completion of four-week protocol, SD = 8.9). CB-Abst were not required to maintain cannabis abstinence after study week 4.
Assessments
Substance use.
Quantity and frequency (days and times used) of alcohol and cannabis use over the past 90 days was assessed at baseline using a modified Timeline Follow-Back interview (TLFB).51 Study staff used anchors such as key dates and events to facilitate retrospective recall. An abridged TLFB was administered at all follow-up visits to approximate amount and frequency of substance use between study visits. At baseline, an alcohol and cannabis dependence screener was administered (Alcohol Use Disorders Identification Test, AUDIT52 and Cannabis Use Disorders Identification Test-Revised, CUDIT-R,53 respectively). The AUDIT is a self-report screening questionnaire used to evaluate quantity and frequency of alcohol use (range: 0-40). A score of ≥ 8 indicates potentially dangerous or problematic drinking and a score of ≥ 13 in women and ≥ 15 in men indicates possible alcohol use disorder (AUD). The CUDIT-R is a self-report screening measure used to assess severity of cannabis use, including consumption, cannabis problems, dependence, and psychological features (range: 0-32). A score of ≥ 8 indicates potentially hazardous use and a score of ≥ 12 indicates possible cannabis use disorder (CUD). The Marijuana Motives Measure54 (MMM) is a 25-item questionnaire assessing 5 factors motivating cannabis use (conformity, expansion, enhancement, coping, and social motives).
Urine samples were collected from all participants at all visits. Samples were qualitatively screened with an immunoassay rapid dip drug test (RDDT; Medimpex United Inc.) for cannabis, cocaine, opiates, amphetamine, and methamphetamine. Samples were also shipped overnight to Dominion Diagnostics (Kingstown, Rhode Island) for quantitative assessment of 11-nor-9-carboxy-Δ9-tetrahydrocannabinol (THCCOOH) concentrations via liquid chromatography-tandem mass spectrometry and adjusted to creatinine levels. Self-reported cannabis abstinence was biochemically verified in the CB-Abst group by progressively decreasing urine concentrations of THCCOOH using a statistical model developed by Schwilke and colleagues.55
Psychiatric symptoms.
Current Axis I psychiatric disorders (including alcohol and other substance use disorders) were assessed via semi-structured interview (Structured Clinical Interview for DSM-5, SCID-556; Kiddie Schedule for Affective Disorders and Schizophrenia, K-SADS57; MINI-KID 7.0.258). Past week symptoms of anxiety and depression were assessed at baseline and all weekly visits with the Mood and Anxiety Symptom Questionnaire (MASQ),59 with higher scores indicating greater distress. Impulsivity was assessed at baseline with the Urgency, Premeditation (lack of), Perseverance (lack of), Sensation Seeking, Positive Urgency, Impulsive Behavior Scale (UPPS-P),60 with higher scores indicating higher levels of impulsivity.
Statistical methods
CB-Abst and CB-Mon were compared on baseline characteristics. We examined group differences on measures via separate linear and logistic regressions, and then computed the relative evidence for the null model versus an alternative model with group terms. Cases where evidence for the alternative model was three times as likely or more indicated significant baseline differences among groups.
We assessed differences in patterns of alcohol consumption in cannabis users randomized to abstinence (CB-Abst) compared to those who did not abstain from use (all assigned to CB-Mon). Participants who were randomized CB-Abst but were non-compliant with the abstinent protocol were included in analyses until resumption of cannabis use occurred, and time points associated with non-abstinence for these CB-Abst participants were excluded (11%). Data for primary analyses described the change over time for 1) days on which alcohol drinking occurred and 2) average number of drinks consumed per study day. We assessed differences between groups in these measures via binomial regression for days spent drinking, and a hurdle gamma regression model controlling for zero-inflation (i.e., non-drinkers) for average number of drinks consumed per day.
Change over time was modeled using a piecewise regression approach, fitting three separate linear trends: baseline to week 1, week 1 to week 4, and week 4 to the post-abstinence (for CB-Abst) follow-up. This piecewise regression approach was chosen a priori to a) simplify the number of contrasts and interactions to test (a 3 x 2 rather than a 5 x 2 design) mitigating multiple comparison issues and improving predictive power, while b) allowing us to still test for a variety of meaningful longitudinal trends (e.g., no change over time, an initial change followed by stability, no initial change followed by a gradual change over weeks 1 to 4, etc.).
Based on the piecewise regression and group terms, we specified a set of planned contrasts to fully characterize group differences (CB-Abst versus CB-Mon) in the change in drinking behavior over time (baseline to week 4). We report results for the pairwise contrasts a) testing group differences in change from baseline to week 1, the linear trend from week 1 to week 4, and change from week 4 to the follow-up visit, and b) the within-group comparisons contrasting baseline against week 1, week 1 versus week 4, and week 4 versus the follow-up visit. Additionally, we compared week 4 and the follow-up period against baseline, and to provide an overview of longitudinal trends and group differences, we report the omnibus tests assessing the main effects of group and time, as well as their interaction.
Analyses examining change in drinking frequency and quantity by randomized group were conducted controlling for age, biological sex, race (White or non-White), ethnicity (Hispanic or non-Hispanic), alcohol use disorder (AUD), CUDIT total scores, and composite scores for the UPPS-P, MMM, and the MASQ. Composite scores for the UPPS-P, MMM, and the MASQ were computed via principal components analysis (PCA) applied to the subscales for the respective inventories. We retained principal components that accounted for 10% or more of observed variance. This approach reduced the number of required covariates and addressed potential issues due to collinearity, while retaining measures that accounted for most of the observed variance among inventory scores.
We estimated models within a Bayesian framework with moderately informative priors that implement a form of regularization, the act of purposefully shrinking coefficients towards zero. Regularization has proven to be a powerful tool in the statistics and machine-learning fields,61 as it achieves a very beneficial bias-variance trade-off. By carefully biasing coefficients towards zero, we reduce the variance seen even in complex models (e.g., adjusted models with a large number of covariates), and we obtain 1) better predictive accuracy for future data (as it prevents over-fitting to noisy data), and 2) a smaller, more interpretable set of coefficients (weak, variable effects unlikely to replicate get shrunk to zero, whereas stronger, replicable effects overcome the bias). Our reliance on this regularization approach reduced the risk of false positives due to issues like multiple comparisons. To control for repeated measures and individual differences, we fit a hierarchical model with subject-level intercepts and slopes for linear trends (corresponding to the maximal random effects structure over the effects of interest62).
We chose to estimate our models within a Bayesian framework for two key reasons: 1) easier implementation and testing of significance for complex models with regularized coefficients (e.g., we do not know of any software that implements and computes frequentist p-values for a variant of the hurdle gamma regression model with regularization), and 2) more intuitive interpretations for posterior p-values compared to frequentist p-values.63 Frequentist p-values estimate the proportion of times, over a hypothetically infinite number of replications, one would see a test statistic as or more extreme than the observed statistic, assuming that a specific hypothesis is true (e.g., that a coefficient equals 0). Frequentist p-values have been challenged on multiple fronts: for example, they a) depend on unobserved data and unrealistic assumptions regarding replications,64 b) do not give the probability of a hypothesis given observed data65 so under most frameworks do not allow for post-data inference,66 and c) can underestimate uncertainty in mixed effects models due to not propagating error from lower levels of the hierarchy.67 Posterior p-values, on the other hand, give the probability for a specified hypothesis (e.g., that a coefficient equals 0) conditioned on the observed data and a set of prior beliefs.68,69 They do not require assumptions about hypothetical replications, nor do they depend on data that were not observed. They are easy to compute from the generated Monte Carlo samples even when estimating complex models, and they properly represent uncertainty across all levels of hierarchical models due to estimation of the joint probability. Finally, their interpretation matches those often mistakenly applied by researchers to frequentist p-values.70 A major criticism of posterior p-values is that they require the specification of a set of prior beliefs, which many researchers worry inject unneeded subjectivity into an analysis. However, the specification of prior beliefs provides a natural and theoretically consistent way to implement regularization,71 which has been shown to improve predictive power and control for false positives.72
We also conducted a post-hoc exploratory analysis to see what factors might predict CB-Abst participants who either a) increased frequency in drinking, or b) decreased frequency in drinking. Participants were categorized by first computing a 95% uncertainty interval around the average change in drinking frequency from baseline to week 1 for the CB-Mon group, using a non-parametric resampling approach. CB-Abst participants whose change in frequency fell above the upper boundary were categorized as having a reliable increase in drinking, while participants whose change in frequency fell below the lower boundary were categorized as having a reliable decrease in drinking. Logistic regressions were then conducted with the covariates from the primary analysis as predictors, with weakly regularizing priors to reduce the risk of false positives.
Finally, we conducted an exploratory analysis of potential binge drinking behavior. As subject-level averages were highly variable due to a large number of participants who only spent a few days drinking, we were unable to examine average drinks per drinking day as a continuous outcome in a similar manner to how we analyzed average number of drinks over total days. Therefore, to assess whether the CB-Abst group may have had inflated binge drinking, while still ensuring our analyses remained reliable and stable, we first created a binary variable categorizing participants based on whether they had an average number of drinks per drinking day in excess of 4 drinks (for females) or 5 drinks (for males). Note these participants were only deemed to have potentially engaged in binge drinking, as no details were available to whether the drinks were consumed within a two hour period. To control for potential baseline differences (i.e., the possibility that participants who engaged in potential binge drinking at baseline would be more likely to engage in binge drinking regardless of group membership), we ran a logistic regression looking at whether participants switched to potential binge drinking post-baseline based on randomized group.
Analyses were conducted using the statistical software R (version 3.6.1).73 The data sets were cleaned and processed using custom scripts and the R packages ‘REDCapR’ (version 0.9.8),74 ‘dplyr’ (version 0.8.3),75 ‘stringr’ (version 1.4.0),76 and ‘readxl’ (version 1.3.1).77 Hierarchical models were fit using the R package ‘brms’ (version 2.10.0).78 Additional analyses (e.g., group differences in participant characteristics) were conducted using the R package ‘BFpack’ (version 0.1.0).79 To ensure reporting only of robust findings, alphas less than or equal to 0.01 were considered statistically significant. Analytic scripts can be found at https://github.com/rettopnivek/ARCHES-alcohol/. Additional details regarding the priors, statistical models, outcomes, coding of covariates, and data processing can be found in the supplementary materials.
RESULTS
Participant characteristics
Table 1 summarizes baseline participant characteristics. Randomized groups did not differ across baseline characteristics (with the exception of lower scores for CB-Mon participants on the UPPS-P measure of general impulsivity). Between 97-100% of this cannabis using sample reported lifetime alcohol use. During the month prior to baseline, 84% of participants drank alcohol at least once. However, the average number of drinks consumed per day in the month preceding baseline was low, with 1.0 days for CB-Abst and 0.8 days for CB-Mon. Over the entire study period (from one month prior to baseline through the 2-4 week follow-up 2), 93% of CB-Abst and 97% of CB-Mon participants consumed alcohol at least one time.
Table 1:
Baseline sample characteristics over groups
| Measures | CB-Abst | CB-Mon |
|---|---|---|
| N | 93 | 67 |
| Age | 20 (2.3) | 19.4 (2.6) |
| Over 21; % yes (n) | 34% (32) | 25% (17) |
| Sex; % female (n) | 46% (43) | 49% (33) |
| Race; % non-White (n) | 31% (29) | 46% (31) |
| Ethnicity; % Hispanic (n) | 9% (8) | 18% (12) |
| Alcohol use | ||
| Lifetime use; % yes (n) | 100% (93) | 97% (65) |
| Age initiated | 15 (1.9) | 15.3 (1.7) |
| Days per week | 1.4 (1.2) | 1.2 (1.3) |
| Drinks per day | 1 (1.1) | 0.8 (1.1) |
| AUD; % yes (n) | 15% (14) | 25% (17) |
| AUDIT | 7.7 (5.9) | 6.3 (4.8) |
| AUDIT; % = 8 (n) | 33% (31) | 36% (24) |
| AUDIT; % = 15 (n) | 14% (13) | 9% (6) |
| Cannabis use | ||
| Age initiated | 15.2 (2) | 15.5 (2) |
| Days per week | 4.2 (2) | 4.4 (2.2) |
| CUD; % yes (n) | 44% (41) | 58% (39) |
| CUDIT-R | 13.8 (5.7) | 13.4 (4.8) |
| MMM – Social | 2.5 (0.6) | 2.3 (0.5) |
| MMM – Conformity | 2.4 (0.6) | 2.4 (0.6) |
| MMM – Non-conformity | 2.4 (0.6) | 2.5 (0.5) |
| MMM – Expansion | 2.4 (0.6) | 2.5 (0.6) |
| UPPS-P – Gen. impulsivity* | 1 (0.8) | 0.6 (0.6) |
| UPPS-P – Perseverance | 0.8 (0.7) | 0.8 (0.6) |
| UPPS-P – Neg. urgency | 0.8 (0.7) | 0.8 (0.7) |
| MASQ – General distress | 1.2 (0.8) | 0.9 (0.8) |
| MASQ – Anhedonic Depression | 0.8 (0.8) | 0.7 (0.6) |
Note. All numbers are means (standard deviations) unless otherwise noted.
Asterisks (*) indicate characteristics that significantly differed across groups based on a Bayes factor test. Composite scores for the MMM, UPPS-P, and MASQ were computed via principal components analysis, and principal components that accounted for 10% or more of observed variance are reported here. Alcohol use disorder, AUD; Alcohol Use Disorders Identification Test, AUDIT; Cannabis use disorder, CUD; Cannabis Use Disorders Identification Test-Revised, CUDIT-R; Marijuana Motives Measure, MMM; Mood and Anxiety Symptom Questionnaire, MASQ; Urgency, Premeditation (lack of), Perseverance (lack of), Sensation Seeking, Positive Urgency, Impulsive Behavior Scale, UPPS-P.
Cannabis abstinence
Ninety-three participants were randomized to CB-Abst, and 83 (89%) achieved four weeks of biochemically verified continuous cannabis abstinence. Among these, 76 (92%) returned for the two to four-week follow-up at which time 70 (92%) had resumed using cannabis.
Change in alcohol use with cannabis abstinence
Reported results are based on adjusted models controlling for age, sex, race, ethnicity, AUD, CUDIT, and composite scores for the UPPS-P, MMM, and the MASQ. Reported outcomes are the frequency (number of days spent drinking) and quantity (average number of drinks consumed per study day) of alcohol use. Table 2 summarizes omnibus tests for the main effects and interaction of group and study week, and reports effect sizes and posterior p-values for the planned contrasts that fully describe the longitudinal changes within and across groups. Figure 1 presents the change over time points by group, reporting means (black dots), model predictions (gray bands for 95% credible intervals), and observed variability (box plots). Table 3 presents the effects of covariates included in adjusted models. The data were well-characterized by the statistical models, as observed averages all fell within model predictions.
Change in outcomes over weeks by group
The figure displays the change over time points per group (CB-Abst – A and C; CB-Mon – B and D) for frequency drinking (A and B) and quantity of drinks (C and D). The light grey region denotes the 30-day abstinence period (relevant for CB-Abst). Lines track the observed average at each time point. Box plots display the 2.5%, 25%, 75%, and 97.5% quantiles for the observed data. Dark gray bands show 95% credible intervals for population-level model predictions from the hierarchical models.
Table 2:
Change in outcomes by weeks and group from adjusted models
| % days spent drinking | Avg. drinks per day in study | |||
|---|---|---|---|---|
| M (SD) | Posterior p-values | M (SD) | Posterior p-values | |
| Omnibus tests | ||||
| Main effects | ||||
| Group | p = 0.05 | p = 0.13 | ||
| Week | p = 0.99 | p = 0.65 | ||
| Interaction | ||||
| Group x week | p < 0.001 | p = 0.006 | ||
| Comparisons between time points by group | ||||
| Baseline vs. Week 1 | ||||
| CB-Abst | 8.5% (1.9) | p < 0.001 | 0.2 (0.1) | p = 0.005 |
| CB-Mon | 0.5% (1.7) | p = 0.78 | 0.0 (0.1) | p = 0.68 |
| Baseline vs. Week 4 | ||||
| CB-Abst | 7.0% (2.1) | p < 0.001 | 0.2 (0.1) | p = 0.003 |
| CB-Mon | 0.0% (1.8) | p = 0.98 | 0.1 (0.1) | p = 0.43 |
| Baseline vs. follow-up | ||||
| CB-Abst | 0.5% (1.6) | p = 0.77 | 0.0 (0.1) | p = 0.94 |
| CB-Mon | −0.3% (1.7) | p = 0.83 | 0.0 (0.1) | p = 0.58 |
| Week 1 vs. Week 4 | ||||
| CB-Abst | −1.4% (2.3) | p = 0.53 | 0.0 (0.1) | p = 0.75 |
| CB-Mon | −0.5% (2.0) | p = 0.80 | 0.0 (0.1) | p = 0.66 |
| Week 4 vs. follow-up | ||||
| CB-Abst | −6.6% (2.0) | p < 0.001 | −0.2 (0.1) | p = 0.002 |
| CB-Mon | −0.3% (1.9) | p = 0.86 | 0.0 (0.1) | p = 0.85 |
| Comparisons between groups by time point | ||||
| Baseline | ||||
| CB-Abst vs. CB-Mon | 0.8% (2.1) | p = 0.70 | 0.1 (0.1) | p = 0.37 |
| Week 1 | ||||
| CB-Abst vs. CB-Mon | 8.8% (3.0) | p = 0.003 | 0.3 (0.1) | p = 0.04 |
| Week 4 | ||||
| CB-Abst vs. CB-Mon | 7.8% (3.2) | p = 0.01 | 0.3 (0.1) | p = 0.06 |
| Follow-up | ||||
| CB-Abst vs. CB-Mon | 1.6% (2.6) | p = 0.54 | 0.1 (0.1) | p = 0.61 |
Table 3:
Covariates from adjusted model predicting average drinking behavior during the study period
| % days spent drinking | Avg. drinks per day in study | |||
|---|---|---|---|---|
| Covariate | Effect size | Posterior p-value | Effect size | Posterior p-value |
| Age | 15.2% | p < 0.001 | 0.4 | p < 0.001 |
| Sex (female vs male) | −0.2% | p = 0.83 | −0.1 | p = 0.29 |
| Race (non-White vs White) | −2.1% | p = 0.04 | −0.1 | p = 0.11 |
| Ethnicity (Hispanic vs non-Hispanic) | 0.2% | p = 0.92 | 0.0 | p = 0.83 |
| AUD | 2.3% | p = 0.07 | 0.1 | p = 0.03 |
| CUDIT-R | 0.6% | p = 0.61 | 0.0 | p = 0.84 |
| MMM – Social | −0.3% | p = 0.76 | 0.0 | p = 0.41 |
| MMM – Non-conformity | 3.1% | p = 0.02 | 0.1 | p = 0.34 |
| MMM – Conformity | 2.6% | p = 0.04 | 0.1 | p = 0.03 |
| MMM – Expansion | 0.7% | p = 0.53 | 0.1 | p = 0.33 |
| UPPS-P – General impulsivity | 2.5% | p = 0.06 | 0.1 | p = 0.07 |
| UPPS-P – Perseverance | −1.1% | p = 0.34 | −0.1 | p = 0.11 |
| UPPS-P – Negative urgency | 1.1% | p = 0.37 | 0.1 | p = 0.34 |
| MASQ – General distress | 0.8% | p = 0.36 | 0.0 | p = 0.89 |
| MASQ – Anhedonic Depression | −1.0% | p = 0.15 | 0.0 | p = 0.09 |
Note. For binary covariates (i.e., sex, race, and ethnicity), the referential group is the second listed in the parentheses.
The CB-Mon and CB-Abst groups had differing longitudinal trends in frequency and quantity of drinks consumed over the study period, with a significant group by study week interaction for both outcomes (p’s<0.007). The planned contrasts indicated that the CB-Mon group exhibited no significant changes relative to baseline over the entire study period (p’s>0.43). In contrast, the CB-Abst group saw an increase in frequency and quantity of drinks consumed by week 1. Levels remained elevated up until week 4, but declined by the 2-4 week follow-up. During the 4 weeks of abstinence, elevated levels for frequency differed significantly compared to CB-Mon, but elevated levels quantity of drinks consumed did not (for alpha = 0.01).
Planned contrasts indicate that at week 1 of the 4-week abstinence period, the CB-Abst group on average consumed drinks at a significantly higher frequency (0.6 more days), relative both to baseline levels (p < 0.001) and compared to CB-Mon (p = 0.003). The CB-Abst group also consumed a greater quantity (0.2 more drinks) relative to baseline levels (p = 0.005), but this did not differ significantly compared to CB-Mon for alpha = 0.1 (p = 0.04). At week 4 of the abstinence period, frequency and quantity of drinks consumed for CB-Abst remained elevated relative to baseline levels (p’s<0.003). Elevated levels for frequency of drinks also differed significantly compared to CB-Mon (p = 0.01), but elevated levels for quantity of drinks did not (p = 0.06). At the 2-4 week follow-up, frequency and quantity for CB-Abst significantly declined compared to levels at week 4 (p’s<0.003), and no longer significantly differed compared to baseline levels (p’s>0.77).
Potential binge drinking
We found 19 participants (20.4%) in the CB-Abst group whose average number of drinks per drinking day increased post-baseline to potential binge drinking levels. We found 13 participants (19.4%) in the CB-Mon group whose average number of drinks per drinking day increased post-baseline to potential binge drinking levels. An exploratory logistic regression found no significant difference in the proportion of participants who increased to potential binge drinking levels between groups (p = 0.89).
Variability in drinking outcomes among CB-Abst
Exploratory analyses found that by week 1, 55 CB-Abst participants (60%) had a significant increase in drinking frequency (from 1.4 to 3.0 days), whereas 21 participants (23%) had a significant decrease in drinking frequency (from 1.5 to 0.7 days), and 17% showed no change. Older participants were more likely to exhibit an increase in drinking (an increase of 2.4 years from the mean corresponded to being 1.8 times more likely to increase in frequency, p = 0.03). Females were 1.7 times less likely to increase in frequency (p = 0.04). No other predictors were significant in differentiating which participants increased versus decreased in drinking frequency by week 1 (Table 4).
Table 4:
Predictors of who increased versus decreased in alcohol frequency from baseline to week 1 in the CB-Abst group
| Predictor | Increased | Decreased | ||||
|---|---|---|---|---|---|---|
| 60% (n = 55) | +1.6 days (mean change) | 23% (n = 21) | −0.9 days (mean change) | |||
| M (SD) | Odds ratio | p-value | M (SD) | Odds ratio | p-value | |
| Age | 20.5 (2.1) | 1.8 | p = 0.03 | 19.8 (2.6) | 1.0 | p = 0.79 |
| Sex; % female (n) | 40% (22) | 0.6 | p = 0.04 | 52% (11) | 1.4 | p = 0.27 |
| Race; % non-White | 27% (15) | 0.9 | p = 0.58 | 38% (8) | 1.3 | p = 0.40 |
| Ethnicity; % Hispanic (n) | 11% (6) | 1.2 | p = 0.45 | 5% (1) | 0.9 | p = 0.57 |
| AUD | 12.7% (7) | 0.9 | p = 0.72 | 23.8% (5) | 1.3 | p = 0.30 |
| CUDIT-R | 14.2 (5.9) | 1.3 | p = 0.31 | 13.2 (5.2) | 0.9 | p = 0.53 |
| MMM – Social | 2.5 (0.7) | 0.9 | p = 0.59 | 2.6 (0.6) | 1.3 | p = 0.35 |
| MMM – Non-conformity | 2.3 (0.7) | 0.9 | p = 0.54 | 2.4 (0.6) | 1.2 | p = 0.56 |
| MMM – Conformity | 2.4 (0.6) | 0.9 | p = 0.51 | 2.5 (0.7) | 1.1 | p = 0.77 |
| MMM – Expansion | 2.4 (0.6) | 1.0 | p = 0.88 | 2.4 (0.7) | 1.0 | p = 0.99 |
| UPPS-P – General impulsivity | 0.9 (0.7) | 1.0 | p = 0.95 | 1.0 (0.9) | 1.1 | p = 0.95 |
| UPPS-P – Perseverance | 0.9 (0.7) | 1.6 | p = 0.09 | 0.6 (0.7) | 0.7 | p = 0.11 |
| UPPS-P – Negative urgency | 0.8 (0.7) | 0.9 | p = 0.61 | 0.8 (0.6) | 1.2 | p = 0.55 |
| MASQ – General distress | 1.3 (1.0) | 1.3 | p = 0.51 | 1.1 (0.6) | 0.8 | p = 0.47 |
| MASQ – Anhedonic Depression | 0.8 (0.9) | 0.9 | p = 0.74 | 0.6 (0.8) | 0.8 | p = 0.42 |
DISCUSSION
This was the first study to our knowledge to evaluate patterns of drinking behavior across one month following an experimental manipulation of cannabis abstinence among non-treatment seeking youth engaged in a paid abstinence atempt. Findings support a heterogeneous effect of cannabis abstinence on alcohol consumption. Of those who abstained from cannabis for the one-month study period, 64% reported an increase in frequency and amount of alcohol use that returned to baseline levels with resumption of cannabis. No changes in drinking patterns were observed among cannabis users who were not randomized to abstinence. Methodological differences may explain why the current findings contrast prior studies that did not find evidence for alcohol substitution, including those that restricted enrollment to only infrequent alcohol users,27,80 queried alcohol use over long retrospective recall periods,27,34 and observed changes in alcohol use with cannabis reduction, not necessarily cannabis abstinence.27
There are several possible explanations for the observed increases in alcohol uses that occurred in most participants during cannabis abstinence. First, it is possible that individuals may seek to replicate the behavioral effects of cannabis shared with alcohol, including feelings of sedation, relaxation, euphoria, relief from distress and anxiety, and disinhibition,81,82 due to the effects of both cannabinoids and sedatives in increasing dopamine release in the nucleus accumbens.83,84 It is also possible that general genetic liability36,85–87 to use substances may explain increases in alcohol use during cannabis abstinence. This would suggest that vulnerable individuals may be at risk to seek other drugs if their primary substance is no longer available or allowed. Finally, while the observed pattern of alcohol use is quite different from the pattern of cannabis use (i.e., multiple days per week), it is possible that youth engage in more frequent alcohol use in attempts to mitigate periods of more intense cannabis withdrawal symptoms, which has been retrospectively endorsed as a reason to drink by abstinent cannabis users.28–30,88 This is supported by the fact that the only increase in alcohol use occurred within the first week of cannabis abstinence, the time in which emergent withdrawal symptoms tend to be most pronounced.35,89 While alcohol use remained elevated in abstinent cannabis users beyond the time of peak of cannabis withdrawal, it may take several weeks for cannabinoids to be fully cleared from the central nervous system90 and thus for withdrawal symptoms to be fully resolved.
While findings from this experimental study support increases in alcohol use during cannabis abstinence, the clinical significance of the degree of alcohol increase is questionable. Youth who abstained from cannabis use increased in drinking frequency by less than one drinking day per week, from an average of 1.4 to 2.1 drinking days per week. Similarly, there was an increase in alcohol quantity of less than half of a standard alcoholic beverage per week, from an average of 1 drink per week prior to cannabis abstinence to 1.2 drinks per week in the first week of abstinence. Additionally, differences in drinking behavior between groups reflected an overall increase in frequency and quantity rather than an increase in binge drinking episodes, further calling into question the clinical significance of observed pattern of change in alcohol use with cannabis abstinence. Future studies should carefully consider whether the mild increase in the frequency and quantity of alcohol consumption in youth outweighs the known health, psychiatric, and cognitive benefits of full cannabis abstinence.91,92
The substantial variability between participants in drinking behavior during the cannabis abstinence period should also be noted. While, on average, the CB-Abst group showed an increase in the frequency and amount of drinks consumed during the 30-day abstinence period, 17% had no change and 23% showed a decrease. Therefore, while at the population level we would predict an average increase in drinking behavior, this does not guarantee increases at the individual level. Peters and Hughes (2010) found having an AUD significantly increased risk for alcohol substitution with cannabis abstinence.29 While AUD status was not predictive of drinking patterns during cannabis abstinence in the current study, we found that older participants and males were more likely to increase in alcohol use during cannabis abstinence. Due to restrictions in power, this study did not formally evaluate whether these covariates moderated group by time interactions. Future studies with larger sample sizes should consider a more comprehensive array of potential static and dynamic moderators in order to predict which individuals are most vulnerable to alcohol increases with cannabis abstinence.
There are some limitations of the current study. First, while cannabis use and cannabis abstinence were biochemically verified, alcohol use status was collected by weekly self-report and was not validated with objective physiological measures. However, participants’ enrollment statuses were not contingent upon their alcohol use, and thus they had no incentive to falsify self-reports. Second, this study recruited cannabis users who agreed to potentially be randomized to one month of sustained cannabis abstinence, and findings may not generalize to the population of cannabis users unwilling to attempt quitting. Further, it is not known whether findings generalize to youth seeking treatment for cannabis use, as this study largely enrolled youth who did not plan to maintain prolonged abstinence after the initial one month of experimental cannabis abstinence. In fact, we systematically excluded those actively planning to stop their cannabis use because of ethical concerns about potentially randomizing those individuals to a no-abstinence, monitoring condition.
This is one of the first studies experimentally examining prospective alcohol use during biochemically verified cannabis abstinence among youth. Findings suggest that clinicians should monitor for increased alcohol use during attempts at abstinence for cannabis among youth.
Highlights
- Four weeks of cannabis abstinence among non-treatment seeking youth was associated with increased frequency and amount of alcohol use.
- Increases in alcohol use occurred in week 1, were sustained over 4 weeks, and resolved with resumption of cannabis use.
- There was notable variability in individual-level response, with 60% increasing in alcohol use and 23% actually decreasing in alcohol use during cannabis abstinence.
Supplementary Material
1
Click here to view.(19K, docx)
Acknowledgments
Funding Source: This publication was made possible by support from NIH-NIDA (1K23DA042946, Schuster; R01 DA04204, Gilman; K24 DA030443, Evins), Harvard Medical School (Norman E. Zinberg Fellowship in Addiction Psychiatry and Livingston Fellowship; Schuster), and the Massachusetts General Hospital (Louis V. Gerstner III Research Scholar Award and Claflin Distinguished Scholar Award, Schuster).
Footnotes
Publisher’s Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Clinical Trial Registration: NCT03276221;
https://clinicaltrials.gov/ct2/show/NCT03276221?term=Randi+Schuster&rank=1
Ethical Statement
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with The Code of Ethics of the World Medical Association (Declaration of Helsinki). Informed consent and assent was obtained for experimentation with human subjects. The privacy rights of human subjects were observed at all times.
