Ecological Momentary Assessment of Cannabis Use and Affect Among Adolescents Following Psychiatric Discharge

Abstract

Objective:

Affect regulation is central to multiple theoretical models that explain cannabis use (CU) behavior. However, much of the research has been conducted with adults, leaving unanswered questions about the nature of associations among adolescents, especially those with affective disorders. Using clinical interviews and ecological momentary assessment (EMA), we assessed rates of adolescent CU and momentary associations with affect following psychiatric discharge among youth hospitalized for suicidal thoughts and behaviors.

Method:

Participants were 13- to 18-year-olds (N = 62; 64.5% female) recruited from an inpatient psychiatric hospital who reported having ever used cannabis. Participants completed clinical interviews during hospitalization. EMA was conducted for 21 days upon discharge.

Results:

Concurrent use of other drugs was associated with greater odds of CU (odds ratio = 27.63). Momentary CU was associated with higher levels of positive affect and lower levels of anger/irritability, but not with negative affect. The effect of momentary CU on positive affect was greater among youth with a diagnosis of posttraumatic stress disorder (PTSD) or generalized anxiety disorder (GAD).

Conclusions:

Findings suggest that adolescents may use cannabis to enhance positive emotion, especially those with PTSD/GAD. Results highlight the importance of tailored interventions that focus on providing alternative and adaptive methods to enhance positive affect.

In the united states, cannabis is one of the most widely used psychoactive drugs (Substance Abuse and Mental Health Services Administration [SAMHSA], 2020), with rates of cannabis use (CU) continuing to rise over the past few decades (National Academies of Sciences, Engineering, and Medicine, 2017). Among adolescents (ages 12–17), 15.4% have reported lifetime use and 12.5% have reported use in the past year (SAMHSA, 2020). CU is often initiated in adolescence and declines with the onset of adulthood (Copeland et al., 2013); however, heavy and persistent use is associated with a host of negative outcomes (Meier et al., 2012, 2016; Sideli et al., 2021). Individuals with co-occurring symptoms of depression/anxiety are more likely to become problematic users by age 24 (Swift et al., 2008), and states with medical cannabis laws have shown evidence of higher rates of self-medication with CU among individuals with mood or anxiety disorders (Sarvet et al., 2018). Despite these potential harms, Americans increasingly perceive cannabis to be harmless (Carliner et al., 2017; Keyes et al., 2016), and places in which cannabis is legalized have demonstrated increases in adolescent perception of the drug's availability and reductions in perceived potential harm (Schuermeyer et al., 2014).

Negative reinforcement models suggest that substance use often arises as a means of relieving unpleasant physical and/ or emotional states (Baker et al., 2004; Koob & Le Moal, 2001). A recent review of CU in daily life supported mixed relations between affect and CU among both community and clinical samples, with the strongest support for CU in response to, and as relief for, negative affect (NA; Wycoff et al., 2018). In contrast, positive reinforcement models suggest that cannabis may be used to enhance positive affect (PA). Sznitman et al. (2022) found that higher PA, but not NA, occurred when college students reported moments of CU. Likewise, Treloar Padovano and Miranda (2018) found that CU was associated with increases in positive subjective effects, especially for adolescents as compared with emerging adults. Findings reinforce the salience of adolescence as a high-risk period for CU given the increased hedonic sensitivity to reward and dampened self-control during this developmental period. Thus, although many ecological momentary assessment (EMA) studies focus on affective and other proximal predictors of CU, evidence suggests that positive reinforcement may serve as a strong mechanism by which cannabis is used as means to enhance PA. In fact, studies suggest that positive reinforcement may be an important mechanism driving early use experiences, whereas negative reinforcement plays a larger role as individuals progress toward use disorder (Bresin & Mekawi, 2019; Mason et al., 2021; Sznitman et al., 2022).

Few studies have been conducted among adolescent populations, and it remains unclear if observed processes among adults are consistent in younger individuals. May et al. (2020) found support for positive, but not negative, reinforcement based on functional imaging in a sample of adolescents with alcohol or CU disorder. Their results support the notion that the peak in rewards responsivity typically observed during adolescence may serve to increase susceptibility to substance use as a means of reward seeking. Furthermore, controlled-laboratory work has demonstrated affective changes following administration of cannabis that were impacted by expectancies and placebo, suggesting that expectancies related to tension reduction may be more related to PA, whereas impairment expectancies may be related to NA (Metrik et al., 2011). Nevertheless, negative life events, prior trauma exposure, and maladaptive coping are all associated with increased consumption, with stress-coping–related use most evident among chronic users (Hyman & Sinha, 2009).

Thus, consistent with an affective-motivational model of addiction, CU may arise in part to relieve or attenuate unpleasant symptoms or as an enhancement and reward-seeking behavior, and this effect may be particularly salient for individuals with affective psychopathology (Baker et al., 2004). Indeed, longitudinal research suggests that individuals with major depressive disorder (MDD) may be at an increased risk for initiating CU (Feingold et al., 2015). CU, as well as other drug and alcohol use, is particularly prevalent among adolescents suffering from mental health disorders. For example, among psychiatric patients between ages 2 and 17 with comorbid substance use and axis I/II disorders, CU disorder was the most prevalent and represented 80% of all substance use disorder cases (Wu et al., 2011). Furthermore, although the days and weeks following discharge from inpatient psychiatric hospitalization are characterized by a marked period of risk among all populations (Chung et al., 2017), CU among hospitalized adolescents may further enhance risk. In a study of young adults, adolescent mood disorder and adolescent CU were independently associated with a sevenfold increase in risk for suicide attempt (Clarke et al., 2014). Moreover, individuals who have been diagnosed for the first time with an affective disorder are more than 18 times more likely to have a high suicide risk assessment at follow-up in the first week after discharge (Madsen et al., 2021). Consistent with a confluence of research underscoring the co-occurrence of CU with mood and anxiety disorder symptoms (Bolton et al., 2009; Chen et al., 2002; Cheung et al., 2010), findings highlight a crucial need for a clearer understanding of the interplay between affective processes and CU during times of heightened risk. This is especially true among youth with mood and anxiety disorders, as they may use cannabis to self-regulate negative emotions and/or to seek positive emotions.

We sought to examine CU among adolescents ages 13–18 and its association with in vivo affective experiences, especially among high-risk youth with mood and anxiety disorders. We drew data from a larger, prospective study of affect regulation among psychiatrically hospitalized adolescents during the 3 weeks following hospital discharge. All youth in this study had been hospitalized for suicidal thoughts or behaviors, representing a sample of individuals who may be at a heightened risk for affective dysregulation as well as substance use. Analyses focused on a subset of youth who reported having ever used cannabis. First, we examined rates of self-reported CU during this high-risk period using EMA. Next, we assessed the momentary associations between CU and experienced affect among youth during the weeks following psychiatric discharge, followed by an examination of whether psychiatric diagnosis moderated the relationship between CU and affect. Across goals, we aim to inform clinical decisions regarding youth experiencing a psychiatric crisis who may be most susceptible to problematic CU during a high-risk period.

Method

Study procedures

The current study presents secondary analyses drawn from a larger study (Nugent et al., 2022) examining affective predictors of suicidal ideation and behavior among psychiatrically hospitalized youth. Adolescents hospitalized for a suicide-related crisis were recruited from a psychiatric hospital in Rhode Island. Following admission, adolescents and their caregivers were invited to participate, and informed consent and assent procedures were completed in person. Self-report surveys and interviews were conducted with adolescents and parents during hospitalization and again 3 weeks and 6 months following hospital discharge. Daily assessments using EMA (described below) deployed on a smartphone were collected during the 3-week period following discharge. Both adolescents and parents were compensated for their time. Study procedures were approved by the hospital institutional review boards (protocol #633181). At the time of this research, recreational CU was not legal for any age group in the state of Rhode Island.

Participants

Inclusion criteria were hospitalization because of suicidal thoughts or behaviors; past-month suicidal thoughts or behaviors verified by clinical interview; age 13 to 18; ability to speak, read, and understand English sufficiently well to complete study procedures; consent of a parent or legal guardian; adolescent assent; and comfort with the use of smartphone technology. Participants were excluded if they had current psychotic symptoms, developmental delay, or pervasive developmental disorder. Adolescents in this secondary analysis included those who reported ever having used cannabis at baseline (n = 66), completed baseline procedures, and returned the study-provided mobile device with valid EMA data (n = 172, 91.5% of participants who completed baseline procedures, N = 188). Of the 16 who completed baseline procedures but did not provide EMA data, 8 participants did not return to follow-up, 2 participants lost their device, 2 participants were either ineligible or withdrew after baseline, and 4 had unknown reasons. See Nugent et al. (2022) for a full description of the study and procedures. The final, analytic sample for the present study was n = 62 cannabis ever users among participants with valid EMA data.

Measures

Demographic characteristics. Participants self-reported demographic characteristics, including sex assigned at birth (female/male), race (White, Black/African American, American Indian/Alaska Native, Asian, Native Hawaiian/other Pacific Islander, or more than one race), ethnicity (Hispanic/ Latinx, not Hispanic/Latinx), and age at enrollment (13–18 years).

Adolescent Alcohol and Drug Involvement Scale (AADIS). Lifetime CU was assessed via the AADIS before hospital discharge (Moberg & Hahn, 1991). The AADIS has demonstrated adequate internal consistency (α = .85) and concurrent validity (Moberg & Hahn, 1991). Participants self-reported how frequently they used cannabis on an 8-point scale ranging from 0 (never used) to 7 (several times a day).

Schedule for Affective Disorders and Schizophrenia for School Aged Children, Present and Lifetime (K-SADSPL). Participant lifetime and current psychopathology was assessed during hospitalization using the K-SADS-PL (Kaufman et al., 1997). Relevant to the current study, participants completed the K-SADS-PL modules assessing MDD, generalized anxiety disorder (GAD), and CU disorder (CUD) according to diagnostic criteria from the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5; American Psychiatric Association, 2013). The interview was administered by clinicians and trained research assistants and audiotaped for interrater reliability. For the present analyses, we focused on current, rather than lifetime, diagnoses.

Clinician-Administered PTSD Scale Child/Adolescent Version (CAPS-CA). Participant posttraumatic stress disorder (PTSD) was diagnosed via the CAPS-CA (Pynoos et al., 2015) during hospitalization. The CAPS is a structured interview that assesses for the severity of PTSD symptoms in the past month on a 5-point scale (0 = absent to 4 = extreme). A symptom counted toward a diagnosis of PTSD if it had a frequency and severity rating equal to or greater than 2, as reported to the interviewer by the participant. A PTSD diagnosis was given if the number of positive symptoms met or exceeded DSM-5 criteria and the participant reported distress caused by these problems. The interview was administered by clinicians and trained research assistants and audiotaped for interrater reliability.

Ecological momentary assessment. Participants were prompted to complete three to five daily random assessments during nonschool waking hours (schedules were tailored to each participant's daily schedule but included before [morning] and after school assessments) for 3 weeks following hospital discharge. Timestamps from each survey were converted to continuous numeric values between 0 and 21 to represent the passage of time. CU was assessed via a series of questions. First, youth were asked to respond (yes/no) to the statement: “Since I completed the last questionnaire, I have consumed or used alcohol or drugs.” Those who reported yes received a follow-up question allowing them to check boxes for the following: alcohol, marijuana, meth, cigarettes, other. Those who reported having used cannabis (marijuana) were coded as 1. Those who reported not using drugs or only using a substance other than cannabis were coded as 0. A second dummy-coded variable was computed to represent the use of any other substances besides cannabis. If participants reported having used alcohol, cigarettes, meth, or other, they were coded as 1. Those who reported having not used any other substances (or reported having used cannabis) were coded as 0. Affect was assessed via selected items drawn from the Positive and Negative Affect Schedule (PANAS-X; Watson & Clark, 1994). The following items were assessed: happy, excited, confident, sad, guilty, worried, lonely, hopeless, shame, and irritable. We also included two items relating to anger (angry at self, angry at others) to capture an additional aspect of negative emotions that was not captured on the PANASX. Item response options comprised a 5-point scale ranging from 1 (very slightly or not at all) to 5 (extremely). Finally, participants responded to the following screening question about their risk for suicide: “Since you last completed a questionnaire, have you thought about hurting or killing yourself?” Participants who responded yes were coded as 1 for thoughts of self-harm and those who responded no were coded as 0. Finally, we created a person-level variable representing compliance based on the number of prompts each participant responded to over the total number of prompts they received over the course of assessment.

Analyses

Cannabis use among adolescents following psychiatric hospitalization. Descriptive statistics were calculated to present baseline rates of CU, as well as to describe data reported across follow-up.

Ecological associations among cannabis use and experienced affect. To examine in vivo CU among youth, we selected a subset of n = 62 who reported having ever used cannabis before. We conducted multilevel logistic regression to estimate the odds of reporting CU while taking dependency of the repeated measures within person into account. In these models, Level 1 represented the within-person (momentary) level and Level 2 represented the between-person level. Model testing proceeded via a series of steps, beginning with a null model to examine the variation in the log-odds of CU across people and calculation of the intra-class correlation coefficient (ICC). Next, predictors of CU were included to determine if CU varied as a function of age, sex, time since discharge, and whether the individual reported other substance use or thoughts of self-harm. Given that other substance use and thoughts of self-harm were both time-varying, each variable was person mean-centered and disaggregated to reflect moments when an individual reported other substance use or thoughts of self-harm (within-person effect) and individuals who reported substance use or thoughts of self-harm more often than others (between-person effect). Finally, we examined whether higher odds of CU were associated with current MDD, GAD, PTSD, or CUD (assessed via separate models).

We then conducted a series of multilevel structural equation models, which accounted for the dependency of repeated measures within person and fits a structural equation model at both the momentary (within) and between-person levels. We used multilevel confirmatory factor analysis (ML-CFA) to evaluate the factor structure of affect items in vivo (Roesch et al., 2010; Sadikaj et al., 2021). Given that we hypothesized that three domains of emotion were theoretically viable among the items we included (i.e., PA, NA, and anger/irritability), we fit a three-factor model but tested alternate one- and two-factor models to determine whether PA, NA, and anger/irritability represented distinct domains. In the two-factor model, we included anger/irritability items as part of NA. In the one-factor model, all items loaded on a single, general factor.

Next, we tested whether EMA-reported CU was associated with affect factor scores at the between- and within-person levels. Because the wording of the CU item was “since your last assessment” and the wording of affect items was “right now,” this association was temporally aligned in the survey such that reported CU preceded current affect. Mean affect scores were computed for each survey and models were adjusted for the covariances between affect scores at the within and between levels. Latent mean centering was used to disaggregate time-varying CU, other substance use, and self-harm into within- and between-level effects (Asparouhov & Muthén, 2018; Curran & Bauer, 2011). Models also included age and sex as time-invariant covariates at the between level and time since discharge as a time-varying covariate at the within level. Last, to test whether observed associations were moderated by mood and anxiety disorders, we entered current MDD, GAD, PTSD, and CUD as between-level predictors of affect factors (assessed as direct effects via separate models), followed by models examining interactions of each diagnosis with CU at the within and between level. Thus, our interaction models tested whether individuals with a psychiatric diagnosis were more likely to report higher/lower affect following moments of CU (i.e., a cross-level interaction) and whether individuals with a psychiatric diagnosis who used cannabis more in general reported higher/lower affect (i.e., a Level 2 interaction).

Analyses were conducted using Mplus Version 8.6, and missing data were handled using full information maximum likelihood estimation with robust estimation (Muthén & Muthén, 1998–2017). Models examining CU outcomes used a logit link. Model fit was evaluated with the model chi-square statistic, as well as several goodness-of-fit criteria including the root mean square error of approximation (with values <.05 indicating good fit), and the comparative fit index (with values >.95 indicating good fit; Browne & Cudeck, 1992; Hu & Bentler, 1999), the Akaike information criterion, and the Bayesian information criterion, in which lower values indicate better fit (Bates et al., 2015).

Results

Participants in the analytic sample of youth who reported ever having used cannabis (n = 62) were between ages 13 and 18 (M_age = 15.7 years, SD = 1.3 years). The majority reported female sex assigned at birth (64.5%), White race (66.7%), and non-Hispanic ethnicity (87.1%). See Table 1 for a summary of all demographics within the full (i.e., all participants with EMA data; n = 172) and analytic samples (i.e., participants with EMA data who reported having used cannabis at baseline; n = 62). More than three quarters (77.4%) of ever users were currently diagnosed with MDD, 40.3% with GAD, 40.4% with PTSD, and 22.6% with CUD.

Table 1.

Participant characteristics

Variable	EMA sample (n = 172)	Analytic sample of ever users (n = 62)
Sex
Female	114 (68.26%)	40 (64.52%)
Male	53 (31.74%)	22 (35.48%)
Race
White	113 (72.44%)	40 (66.67%)
Black or African American	14 (8.97%)	9 (15.00%)
Asian	2 (1.28%)	0 (0.00%)
American Indian/Alaska Native	2 (1.28%)	2 (3.33%)
Native Hawaiian or other Pacific Islander	0 (0.00%)	0 (0.00%)
Ethnicity
Hispanic or Latino	24 (15.00%)	8 (12.90%)
Not Hispanic or Latino	136 (85.00%)	54 (87.10%)
Age, in years
M (SD)	15.14 (1.44)	15.7 (1.3)
Baseline diagnoses
Cannabis use disorder	15 (8.72%)	14 (22.58%)
Posttraumatic stress disorder	42 (26.25%)	23 (40.35%)
Major depressive disorder	132 (76.74%)	48 (77.42%)
Generalized anxiety disorder	62 (36.05%)	25 (40.32%)
Baseline cannabis use
Never used	102 (60.71%)	0 (0.00%)
Tried but quit	22 (13.10%)	20 (32.26%)
Several times a year	9 (5.36%)	9 (14.52%)
Several times a month	11 (6.55%)	11 (17.74%)
Weekends only	4 (2.38%)	3 (4.84%)
Several times a week	13 (7.74%)	12 (19.35%)
Daily	2 (1.19%)	2 (3.23%)
Several times a day	5 (2.98%)	5 (8.06%)

Note: Ns may vary because of missing data.

EMA data comprised a total of 2,056 assessments among ever users. On average, ever users completed 2.84 assessments per day (SD = 1.71), averaging to 31.15 (SD = 24.85) assessments per person with an overall compliance rate of 34% (compliance was operationalized as the number of prompts responded to over the total number of prompts received; participants with known tech or app issues [i.e., those who reported not receiving prompts] were excluded from the overall compliance rate but were included in the analyses). Compliance among ever users was lower than the overall compliance among all participants (41%, n = 172); consequently, compliance was included as a covariate for all subsequent analyses. Across all ever users and days, the middle 50% of EMA responses occurred after school hours, between 3:21 P.M. and 7:52 P.M., and 7.7% occurred before noon.

Cannabis use among adolescents

At baseline, 60.7% of all youth reported having never tried cannabis and 8.7% met criteria for CUD. A total of 11.9% of all youth were experimental users (i.e., several times a month or several times a year), and 14.3% were regular users (i.e., weekends, several times a week, daily, or several times a day). Among ever users 22.6% met criteria for CUD, and among regular users 58.3% met criteria for CUD. During the 3 weeks following hospital discharge, 21.5% (n = 37) of all youth, regardless of whether they had previously used, reported at least one instance of use via EMA. Among the analytic sample of youth who reported prior use, 47.0% (n = 31) reported at least one instance of use via EMA, with a mean number of uses of 6.8 (SD = 11.1). Regarding other substances, 15.2% of youth in the analytic sample reported at least one instance of alcohol use, 10.6% reported cigarette use, and 3.0% reported other drug use. Together, nearly one quarter (21.2%) of ever users reported using any other substance other than cannabis (i.e., alcohol, cigarettes, or other drugs).

Predictors of in vivo cannabis use among ever users

The ICC for in vivo CU across the 3 weeks of assessment was large (66%, adjusted for multilevel logistic models; Wu et al., 2012), indicating that observations of CU were largely explained by between-person variability. When we added demographic and contextual covariates to the model, four youth were excluded from analyses because of missing data on either age or sex. Results are presented in Table 2 and indicate that youth who reported other drug use in the moment (a within-person effect) were more likely to report having used cannabis (odds ratio [OR] = 27.63). Youth who reported thoughts of self-harm did not report significantly higher rates of CU, and no significant effects were observed due to sex, age, time since discharge, or compliance. When diagnoses of mood and anxiety disorders were added as predictors of CU, results indicated that youth with CUD were highly likely to report CU during this time (OR = 31.69).

Table 2.

Results of multilevel logistic regressions predicting in vivo cannabis use among ever users (n = 62)

Variable	Estimate	SE	p	OR
Intercept	8.78	5.34	.100
Self-harm (within)	0.47	0.41	.26	1.59
Self-harm (between)	-1.59	1.85	.390	0.20
Other drugs (within)	3.32	0.86	<.001	27.63
Other drugs (between)	3.11	3.33	.350	22.49
Time	0.01	0.02	.518	1.01
Male	0.62	0.73	.400	1.86
Age	0.29	0.35	.396	1.34
Compliance	-0.23	1.73	.894	0.79
Residual variance	5.65	1.83	.002
Between-person models including baseline diagnosisa
PTSDb	0.03	0.87	.976	1.03
MDD	-0.83	0.71	.242	0.43
GAD	-0.43	0.84	.604	0.65
CUD	3.46	0.71	<.001	31.69

Notes: Bold indicates p < .05. OR = odds ratio; PTSD = posttraumatic stress disorder; MDD = major depressive disorder; GAD = generalized anxiety disorder; CUD = cannabis use disorder.

^a Models assessed separately for each diagnosis;

^b n = 57 because of missing data for PTSD diagnosis.

In vivo association between reported cannabis use and subsequent affect

ML-CFA was conducted in the full sample of 172 youth. ICCs for individual affect items ranged from .36 to .59. Although we hypothesized a three-factor model based on prior work and affective categories, to explore potential alternate solutions and determine the best fitting model, one-, two-, and three-factor models were fit (Table 3). The three-factor solution demonstrated superior model fit and was supported by theory, with correlated factors for PA, NA, and anger/irritability items. Standardized factor loadings and correlations for the three-factor model are presented in Table 4. The path model is presented in Figure 1. At the within level, all affect factors were significantly correlated with each other. At the between level, PA was significantly negatively correlated with NA, but not anger/irritability.

Table 3.

Fit indices for three multilevel confirmatory factor models among all youth (n = 172)

Variable	1-factor	2-factor	3-factor
Model χ2	2,692.89	1,066.81	1,022.37
df	108	106	102
p	<.001	<.001	<.001
Scaling correction	2.44	2.39	2.37
CFI	.70	.89	.89
RMSEA	.06	.04	.04
SRMR (within)	.09	.05	.05
SRMR (between)	.16	.08	.07
AIC	191,525.00	187,511.70	187,384.10
BIC	191,934.60	187,935.00	187,834.70
aBIC	191,743.90	187,737.90	187,625.00

Notes: CFI = comparative fit index; RMSEA = root mean square error of approximation; SRMR = standardized root mean squared residual; AIC = Akaike information criterion; BIC = Bayesian information criterion; aBIC = sample size–adjusted BIC.

Table 4.

Standardized item factor loadings, correlations, and intraclass correlation coefficients (ICCs) for three-factor multilevel confirmatory factor analysis among all youth (n = 172)

Construct	Item	Within	SE	Between	SE	ICC
Positive affect	Happy	0.78	0.02	0.93	0.04	.47
	Excited	0.77	0.02	0.92	0.05	.45
	Confident	0.55	0.03	0.80	0.05	.36
Negative affect	Sad	0.70	0.02	0.91	0.03	.38
	Guilty	0.62	0.04	0.86	0.04	.50
	Worried	0.54	0.03	0.82	0.04	.57
	Lonely	0.65	0.03	0.84	0.05	.52
	Hopeless	0.71	0.02	0.90	0.03	.57
	Shame	0.62	0.04	0.86	0.04	.59
Anger/irritability	Angry at self	0.70	0.03	0.91	0.03	.57
	Angry at others	0.52	0.03	0.77	0.08	.48
	Irritable	0.56	0.03	0.79	0.05	.41
Positive with negative affect		-0.40	0.05	-0.29	0.10
Positive affect with anger/irritability		-0.37	0.05	-0.18	0.09
Negative affect with anger/irritability		0.90	0.02	0.95	0.03

Figure 1.

Three-factor multilevel confirmatory factor analysis path model of in vivo affect. Note: Standardized parameter estimates presented with standard errors in parenthesis. Bold indicates p < .05. Top portion represents the between-person model, bottom portion presents the within portion model.

We then computed mean scores for each affective domain to examine structural relations among EMA-rated affect and time-varying associations with CU among ever users (Table 5). Results indicated that youth with lower compliance reported higher levels of anger/irritability, but not PA or NA. Results also revealed a significant time trend for all affective domains, indicating that the mean level of PA decreased over time while the mean level of NA and anger/irritability increased. Regarding CU, within-person association revealed that when youth reported having used cannabis since their last assessment, they rated higher PA and lower anger/irritability, but no significant associations for NA emerged. None of the between-person associations between affect factors and CU were significant, suggesting that individuals who reported using cannabis more often did not have significantly higher/lower levels of affect overall. Reports of any drug use at the momentary level were associated with higher NA and anger/irritability, but not PA. Thoughts of self-harm were associated at both the within and between level for all affect factors, revealing that the presence of self-harm at the momentary level and overall was associated with lower PA and higher NA/anger/irritability. Results for demographic covariates indicated that only male sex was associated with higher PA at the between-person level and that better compliance was associated with lower overall anger/irritability.

Table 5.

Results of multilevel structural equation models of cannabis use predicting affect among ever users (n = 62)

Affect	Predictor	Estimate	SE	p
Within-person model
Positive affect	Time	-0.03	0.01	<.001
	Cannabis use	0.28	0.07	<.001
	Other drugs	-0.15	0.12	.232
	Self-harm	-0.22	0.09	.012
Negative affect	Time	0.01	0.01	.02
	Cannabis use	-0.08	0.05	.17
	Other drugs	0.14	0.07	.053
	Self-harm	0.55	0.12	<.001
Anger/irritability	Time	0.01	0.01	.032
	Cannabis use	-0.17	0.07	.012
	Other drugs	0.27	0.11	.012
	Self-harm	0.51	0.08	<.001
Between-person model
Positive affect	Cannabis use	0.15	0.49	.766
	Other drugs	0.51	0.61	.407
	Self-harm	-0.95	0.31	.002
	Male	0.44	0.22	.046
	Age	-0.12	0.06	.064
	Compliance	-0.16	0.45	.716
Negative affect	Cannabis use	-0.17	0.33	.614
	Other drugs	0.31	0.42	.459
	Self-harm	2.18	0.60	<.001
	Male	-0.07	0.15	.632
	Age	0.01	0.08	.899
	Compliance	-0.16	0.31	.612
Anger/irritability	Cannabis use	0.16	0.37	.663
	Other drugs	0.52	0.79	.507
	Self-harm	1.58	0.65	.015
	Male	-0.01	0.14	.962
	Age	-0.05	0.08	.526
	Compliance	-0.76	0.27	.005

Note: Bold indicates p < .05.

When current mood and anxiety disorders were included as between-person predictors of affect (Table 6), the previously observed associations between moments of CU with higher PA and lower anger/irritability remained significant. Main effects were observed such that MDD was associated with lower PA and higher NA, but no main effects were observed for any other diagnostic category. When current psychiatric diagnosis was entered as a moderator of within-/ between-level CU, results indicated that individuals with PSTD or GAD reported higher PA following moments of CU (i.e., cross-level interactions). No interactions were observed for MDD or CUD and no interactions were observed at the between-person level.

Table 6.

Results of main and interaction effects of cannabis use (CU) and psychiatric diagnosis predicting affect among ever users (n = 62)

Diagnosis	Affect	Effects	Estimate	SE	p
PTSDa	Positive affect	Main effect of diagnosis	0.05	0.20	.812
		CU within interaction	0.29	0.13	.028
		CU between interaction	-0.25	0.96	.798
	Negative affect	Main effect of diagnosis	0.06	0.16	.688
		CU within interaction	-0.05	0.10	.640
		CU between interaction	0.54	0.70	.438
	Anger/irritability	Main effect of diagnosis	0.03	0.15	.818
		CU within interaction	-0.16	0.12	.195
		CU between interaction	0.26	0.65	.688
MDD	Positive affect	Main effect of diagnosis	-0.65	0.22	.003
		CU within interaction	0.11	0.15	.491
		CU between interaction	1.06	6.04	.861
	Negative affect	Main effect of diagnosis	0.39	0.16	.012
		CU within interaction	-0.18	0.24	.450
		CU between interaction	1.11	4.50	.806
	Anger/irritability	Main effect of diagnosis	0.27	0.15	.081
		CU within interaction	-0.31	0.29	.281
		CU between interaction	0.84	4.57	.854
GAD	Positive affect	Main effect of diagnosis	-0.32	0.18	.077
		CU within interaction	0.29	0.14	.033
		CU between interaction	0.73	0.96	.445
	Negative affect	Main effect of diagnosis	0.12	0.19	.546
		CU within interaction	-0.12	0.09	.190
		CU between interaction	-0.30	0.63	.641
	Anger/irritability	Main effect of diagnosis	0.09	0.18	.623
		CU within interaction	-0.16	0.11	.155
		CU between interaction	0.28	0.83	.738
CUD	Positive affect	Main effect of diagnosis	0.44	0.43	.304
		CU within interaction	-0.02	0.16	.887
		CU between interaction	0.32	1.88	.867
	Negative affect	Main effect of diagnosis	-0.22	0.23	.348
		CU within interaction	-0.10	0.13	.418
		CU between interaction	-0.08	2.09	.971
	Anger/irritability	Main effect of diagnosis	-0.21	0.22	.328
		CU within interaction	-0.20	0.15	.178
		CU between interaction	-2.40	1.63	.141

Notes: Bold indicates p < .05. Separate models fit for main effects and interaction effects by diagnostic category; all models are adjusted for the following: other drug use, time, thoughts of self-harm, age, sex, and compliance. PTSD = posttraumatic stress disorder; MDD = major depressive disorder; GAD = generalized anxiety disorder; CUD = CU disorder.

^a n = 57 because of missing data for PTSD diagnosis.

Discussion

We examined CU among youth recruited during psychiatric hospitalization. Nearly half of youth reported having ever tried cannabis—a number much higher than the 15% lifetime use reported among 12- to 17-year-olds in U.S. epidemiologic studies (SAMSHA, 2020). Further, 14% reported being regular users, and 12% reported being experimental users. Analyses revealed that 22% of all youth, and 47% of ever users, reported using cannabis at least once during the 3 weeks following discharge. Youth who reported moments of other substance use were more likely to report CU, as were youth with a diagnosis of CUD. Although this sample represented a high-risk group of youth, no evidence was found that thoughts of self-harm increased the risk for CU.

Although several prior EMA studies point to NA as one of the most consistent processes related to CU (Wycoff et al., 2018), we did not observe differences in NA associated with CU. Given that our sample comprised adolescents who may have recently initiated CU or may not use heavily/frequently, the negative reinforcement cycles that often drive addiction (i.e., use to alleviate cravings, avoid withdrawal, or as a maladaptive coping mechanism) may not yet be established. Instead, it is possible that CU among youth during this time may be predominantly driven by positive reinforcement. This early influence of positive reinforcement, which functions through experiences of pleasure and reward, as a motivator for initial use has been hypothesized to lead to addiction as early use becomes sustained and maintained by negative reinforcement (Koob, 2013).

Thus, although more research is needed to tease out causal effects, especially among youth, our findings are consistent with a growing body of literature of positive reinforcement following CU (Gruber et al., 2012; Henquet et al., 2010; Trull et al., 2016; Tyler et al., 2015). Indeed, our results indicated that when youth reported having used cannabis, they had higher subsequent PA and lower subsequent anger/irritability. This finding is notable relative to other explored predictors such as other drug use or current MDD. Specifically, reports of any drug use (e.g., smoking, alcohol, or other drugs) or having current MDD were associated with higher NA, whereas having current MDD was also associated with lower PA. Our findings are consistent with prior work suggesting that certain individuals may use cannabis as a means to heighten PA, serving as positive reinforcement (Wycoff et al., 2018). However, as individuals progress toward a cycle of addiction, neurological adaptations associated with heavier and more frequent use begin to drive a negative reinforcement model in which cannabis may be used to avoid symptoms of NA and withdrawal (Koob & Volkow, 2010). Although we tested for moderation of effects by current CUD, which may indicate differing associations among those with more problematic use, no significant results emerged. These findings should be considered preliminary in light of the small number of youth with CUD, and these effects should be tested in larger samples of cannabis users.

Instead, our results indicated a significant interaction between momentary CU and PA such that those diagnosed with current PTSD or GAD were more likely to report higher levels of PA than others. Thus, youth experiencing anxiety disorders may be most likely to use cannabis to enhance PA. Prior studies have consistently found an association between both PTSD and anxiety with CU (Cougle et al., 2011; Gentes et al., 2016; Kedzior & Laeber, 2014). Indeed, enhancement and coping mechanisms have been reported as the most common reasons for CU among users (Buckner et al., 2015), but some research paradoxically suggests a lower quality of life reported among regular users of cannabis (vs. nonusers) with anxiety disorders (Lev-Ran et al., 2012). Likewise, EMA studies of CU among individuals with PTSD have supported PTSD symptom reductions in response to use (LaFrance et al., 2020), but little research exists with regard to enhancement among those with PTSD. Although the momentary enhancement of PA may be a motivating factor for CU for some individuals, it remains unclear whether short-term alleviation translates to longer-term effects, especially in the context of co-use of other substances, which demonstrated an opposing effect such that other drug use was associated with greater NA and anger/irritability.

This study should be considered within the context of several limitations. First, the sample consisted of adolescents who were being discharged from inpatient psychiatric care and are considered a high-risk population for suicide. Although we observed higher rates of CU relative to the general U.S. population, youth were not recruited on the basis of substance use. In addition, although our results were robust even after controlling for thoughts of self-harm, our findings may not generalize to other clinical and nonclinical populations and may look different in a population of youth characterized by more frequent or regular use or those at lower risk for suicide. Furthermore, because these findings represent secondary analyses from a larger study, information specific to CU—such as method of administration, quantity, THC (i.e., delta-9-tetrahydrocannabinol, the main psychoactive component of cannabis) content, or age at initiation—was not collected. Associations of CU with affective changes following consumption may vary as a function of the quantity and potency of the cannabis product. In addition, our analyses only examined associations at two levels, representing between- and within-person results. It is possible that daily fluctuation in CU and/or affect may play an important role in understanding use patterns (i.e., weekend vs. weekday effects). Future studies should consider integrating day-level predictors of CU and affect. Finally, because of the design and wording of the questions in the EMA, we were unable to examine reverse associations (i.e., affect predicting subsequent CU) because it would require lagging the data, which would result in considerable data loss. Thus, we were unable to directly test whether use tended to precede or succeed elevations in affect. If positive reinforcement is driving CU, it is possible that anticipatory effects representing higher affect may precede use. Future studies designed to specifically disentangle the directionality—and potential bidirectionality—of affect and CU may help to shed novel light on reinforcement processes assessed in vivo and should consider whether these processes are particularly strong for those with anxiety disorders like PTSD/GAD.

Finally, EMA compliance represented a relative weakness in the present sample. In a meta-analysis of EMA compliance of children and adolescents, Wen et al. (2017) reported that compliance for clinical samples using a frequency of four to five prompts per day was 66.9%. However, none of the clinical studies included in the 2017 review were conducted using samples of adolescents at risk for suicidal thoughts and behavior. A systematic review of EMA studies of suicide phenotypes demonstrated a wide range of compliance rates across studies (i.e., 52%–90%), although few had comparable sample sizes, sampling frequencies, or follow-up lengths to ours and studies were not limited to adolescents (Sedano-Capdevila et al., 2021). Potential influences on compliance in our study are likely complex and may be related to the characteristics of high-risk youth (i.e., hesitation to complete EMA during times of distress out of fear of rehospitalization) but were found to be unrelated to clinical diagnosis. We did find, however, that poorer compliance was associated with higher reported anger/irritability. Thus, it is evident that youth in our study who struggled emotionally during their transition home from hospitalization may have struggled more with compliance. Despite the significant associations between compliance and affect, however, our main findings remain unchanged.

Clinical implications

Findings from this work have several clinical implications. First, our results suggest that CU among adolescents may be driven in part by positive reinforcement and reward seeking and that the effect is strong among youth with PTSD or GAD. Thus, future prevention and intervention strategies may seek to identify alternate ways for youth to boost positive emotions rather than focusing on negative reinforcement, which comes later in the addiction cycle and may be less relevant to youth who are younger and/or have not yet developed a substance use disorder. Indeed, some researchers are highlighting the PA system as an underexplored treatment target for those with anxiety and depression (Taylor et al., 2017). Further, evidence-based interventions aimed at reducing CU in adolescents may include individual and multisystem approaches (e.g., behavioral interventions, cognitive–behavioral interventions, motivational approaches, Multisystemic Therapy; Bender et al., 2011), with most orientations encompassing a component addressing positive emotions (e.g., behavioral activation techniques). Perhaps not surprisingly, findings also suggest that the use of other drugs was related to CU. Findings support previous calls for taking an upstream approach to CU prevention by strengthening peer networks (Hyshka, 2013), interventions that target perceived social norms of CU among adolescents using cannabis (Blevins et al., 2018), as well as those that focus on parent training (Allen et al., 2016), which may help deter high-risk adolescents from turning to substance use for coping, enhancement, and other self-regulatory behaviors.

Conclusions

Results highlight the need for future research among these youth, as the weeks and months following psychiatric discharge are a known period of increased psychiatric risk (Mirkovic et al., 2020). Current findings suggest that nearly one in four youth reported using cannabis during this high-risk period. Concurrent use of other drugs was associated with greater odds of CU. Reports of CU were followed by significantly higher PA and lower anger/irritability, and the effect of CU on enhanced PA was greater for those with PTSD/GAD. In light of these findings, more research is needed to determine whether these types of short-term effects translate into long-term benefits, especially among youth leaving psychiatric care. Given that self-regulation through the use of cannabis may function as a maladaptive coping mechanism for some youth, these findings also underscore the potential for tailored interventions that focus on providing more adaptive alternative methods to enhancing PA.