Predictors and moderators of marijuana and heavy alcohol use outcomes in adolescents treated for co-occurring substance use and psychiatric disorders in a randomized controlled trial

Abstract

Background:

The current study identifies predictors and moderators of substance use outcomes for 111 adolescents with co-occurring substance use and psychiatric disorders who participated in a randomized controlled trial that compared the effectiveness of two home-based treatments: an integrated cognitive behavioral therapy (I-CBT) protocol, in which masters-level clinic staff received intensive training and ongoing supervision in the use of this protocol versus a treatment-as-usual (TAU) comparison condition in which therapists received a continuing education–style CBT workshop in the same protocol.

Method:

The study conducted exploratory predictor and moderator analyses of marijuana and heavy alcohol use outcomes using candidate variables across four domains of psychological characteristics: adolescent substance use, adolescent psychiatric symptoms, parent, and family.

Results:

Regardless of treatment condition, low parental monitoring at baseline, as assessed by a videotaped interaction task, but not self-report, predicted greater percentage of marijuana use and heavy alcohol use days over the 6-month follow-up period. If parents entered treatment with low levels of parental monitoring, adolescents in the I-CBT condition reduced their percentage of heavy alcohol use days significantly more than adolescents in TAU over the 6-month follow-up period. Greater adolescent aggression and parental emotion dysregulation at baseline also predicted greater percentage of marijuana use days over the 6-month follow-up period for the sample as a whole. Adolescents in the I-CBT condition who reported low positive urgency at baseline reduced their percentage of heavy alcohol use days significantly more than adolescents in TAU care over the 6-month follow-up period.

Conclusion:

The article discusses implications for clinical decision-making, improving treatment effectiveness, and tailoring interventions for adolescents with co-occurring substance use and psychiatric disorders.

1. Introduction

Co-occurring substance use and psychiatric disorders are prevalent in adolescents across various treatment settings. In community-based settings, approximately 60% of adolescents diagnosed with a substance use disorder (SUD) also meet criteria for a comorbid psychiatric disorder (Armstrong & Costello, 2002). In SUD-specific outpatient and inpatient settings, prevalence rates are even higher, with nearly 80% of adolescents with SUD diagnoses experiencing a co-occurring psychiatric disorder (Dennis et al., 2004; Grella, Joshi, & Hser, 2004; Mason & Posner, 2009; Rowe, Liddle, Greenbaum, & Henderson, 2004). Many adolescents with SUD also have more than one comorbid psychiatric disorder (Rowe et al., 2004; Langenbach et al., 2010).

This co-occurrence of substance use and comorbid psychiatric disorders is associated with complex illness presentations and deleterious outcomes, including suicide attempts, death by suicide, and law-breaking behavior (Brent et al, 1993; King et al, 1996; Clingempeel, Britt, & Henggeler, 2008). In general, more severe comorbidity is associated with negative outcomes. More specifically, adolescents with comorbid SUD, externalizing, and internalizing disorders tend to experience a worse course of illness than those without such comorbidity (Rowe et al., 2004; Shane et al., 2003; Tomlinson et al., 2004). Across SUD outpatient, intensive outpatient, and inpatient/residential settings, comorbidity also impedes treatment response (Cornelius et al, 2004; Grella, 2003; Rowe et al., 2004; Shane, Jasiukaitis, & Green 2003; Tomlinson, Brown, & Abrantes, 2004; White et al., 2004).

Treatments designed to target comorbid symptom presentations are both necessary and scarce. In fact, few interventions have been developed expressly for the dual nature of co-occurring substance use and psychiatric disorders (for exceptions, see medication studies by Cornelius et al., 2009 and Riggs et al., 2007, 2011). Instead, many treatment protocols rely on a SUD-specific or psychiatric-specific approach to comorbid presentations. When such treatments are provided clinically, lack of coordination and integration between SUD-specific and mental health–specific treatment settings often detracts from effective implementation of such protocols (see Hawkins, 2009).

One randomized controlled trial (RCT) tested the efficacy of an integrated cognitive behavioral therapy (I-CBT) treatment protocol designed to address co-occurring substance use and psychiatric presentations in adolescents discharged to outpatient care following a psychiatric hospitalization (Esposito-Smythers et al, 2011). Findings from this RCT indicated that the I-CBT protocol was more effective in reducing both substance use and psychiatric symptoms than was treatment-as-usual (TAU). An extension of this study modified the I-CBT protocol for use by masters-level clinicians in a community mental health clinic treating adolescents with broader co-occurring presentations and more significant externalizing behavior than in the original study (Esposito et al, 201). Using a two-group design with follow-up assessments at 3-, 6-, and 12-months, this trial randomized 111 eligible adolescents (ages 12–18 years old) and a participating parent to either I-CBT or TAU. In the I-CBT protocol, masters-level clinic staff received intensive training and ongoing supervision in the use of this protocol. In the TAU comparison condition, therapists received a less intensive, continuing education–style CBT workshop in the same protocol. Primary results over the follow-up period indicated a small, but not statistically significant, reduction in the percent days of heavy drinking and marijuana use over time for both conditions, with the overall effect across the three follow-up points favoring the I-CBT condition. Nonetheless, treatment outcomes were not uniform. Some adolescents in both conditions did well, and some adolescents in both conditions did poorly. Thus, the goal of the current study was to determine which adolescent, parent, and family characteristics predicted the best substance use treatment outcomes, in general, and whether these factors varied by the type of treatment offered.

1.1. Predictors of adolescent substance use treatment

Within the adolescent psychological domain, this study explored characteristics associated with both internalizing and externalizing behavior. The study specifically examined the possible differential effects of internalizing presentations, i.e., depression, versus externalizing presentations, i.e., impulsivity and anger, as predictors and moderators of substance use outcomes. A recent systematic review of 61 studies of adolescent substance use found that depression, more so than anxiety, is a consistent predictor of substance use, after controlling for co-occurring externalizing symptoms (Hussong, Ennett, Cox, & Haroon, 2017). Similarly, impulsivity has been associated with increased substance use (for review, see Stautz & Cooper, 2013), as impulse-control capacities remain immature during this developmental stage (Steinberg et al., 2008). Research has also shown impulsivity to be highly correlated with anger, which significantly predicts adolescent problem behavior (e.g., Colder & Stice, 1998). Trait anger also has been linked to broad externalizing symptoms in adolescence (Li, Hein, Ye, & Liu, 2019). Adolescents with high levels of anger are more likely to experience more severe substance use and psychiatric symptoms than those with low levels of anger (Serafini, Toohey, Kiluk, & Carroll, 2016). This study also examined whether an adolescent’s ability to regulate emotions, such as anger, would predict substance use outcomes, as prior work suggests that anger is a transdiagnostic risk factor for substance use and comorbid psychopathology (for review, see Shadur & Lejuez, 2015).

Parent variables of interest included both parenting strategies, particularly parental monitoring and limit-setting, as well as parental regulation of their own emotions. Meta-analytic and systematic reviews have revealed that parental monitoring is associated with decreases in adolescent marijuana use (e.g., Lac & Crano, 2009) and alcohol use (e.g., Ryan, Jorm, & Lubman, 2010). Relatedly, in a recent study of adolescents with substance use problems, parental supervision was associated with a significantly lower risk of simultaneous marijuana and alcohol use (Lipperman-Kreda, Gruenewald, Grube, & Bersamin, 2017). Though likely linked, the association between parental mental health, in general, and more specific problems, such as emotion dysregulation and adolescent substance use outcomes has been largely understudied. Evidence suggests that parental emotion regulation is related to offspring emotion regulation (e.g., Bariola, Gullone, & Hughes, 2011), and adolescent emotion regulation is related to their substance use (e.g., Weinberg & Klonsky, 2009). Perhaps, then, parental emotion regulation interacts with adolescent emotion regulation to predict substance use. As such, this study explored whether parental emotion regulation might be a relevant predictor or moderator of adolescent substance use outcomes.

Finally, research has shown family functioning to be related to adolescent substance use outcomes. Prior work has implicated family functioning as a predictor of adolescent substance use. In a study of more than 800 high schoolers, Cambron and colleagues (2018) found that low family management, involvement, and bonding were associated with increases in cigarette smoking and alcohol use. Another study found that family cohesion, or the level of affection, support, warmth, and connectedness within a family, had protective effects on adolescent drinking behavior (Reeb et al., 2015). Finally, in a comprehensive review of risk and protective factors for substance use in young adulthood, negative family relations were significantly associated with increases in offspring substance use (Stone, Becker, Huber, & Catalano, 2012). As such, this study explored whether family functioning predicted or moderated adolescent substance use outcomes.

1.2. Current study

The current study tested predictors and moderators of substance use outcomes in an RCT for adolescents with co-occurring substance use and comorbid psychiatric disorders. The study conducted predictor analyses to identify baseline characteristics that predicted which adolescents had either optimal or suboptimal responses to either treatment, regardless of intervention type. The research team conducted moderator analyses to identify baseline characteristics that predicted which adolescents had a better response to a specific treatment (i.e., I-CBT versus TAU; Kraemer, Wilson, Fairburn, & Agras, 2002). Taken together, predictor and moderator analyses allowed for an examination of the conditions under which treatment was most effective. The team selected candidate variables for both the predictor and moderator analyses from three groups of characteristics, including adolescent, parent, and parenting/family.

2. Material and methods

2.1. Participants

Participants in the RCT included 61 families in the I-CBT condition and 50 in the TAU condition. Adolescents were eligible for the study if they were: enrolled in an intensive outpatient, home-based program (IOP) for co-occurring disorders at a community mental health clinic, English-speaking, 12 to 18 years old, and using alcohol and/or other substances in the prior three months, per their report. Parents/guardians also had to speak English (See Wolff et al, 2020).

2.2. Procedure

The study team invited all adolescents, and their parent or legal guardian, who had an intake appointment at the IOP to participate in the study. After obtaining written informed consent and assent from parents and adolescents, respectively, study staff administered a baseline assessment battery. Research assistants blind to treatment condition conducted follow-up assessments at 3, 6, and 12 months. The hospital human subjects’ protection committee approved the study. More details about the procedures and results of the RCT can be found elsewhere (Wolff et al, 2020).

2.3. Treatment conditions

In this trial, masters-level clinicians employed at a community mental health clinic delivered home-based therapy to adolescents in both protocols. In both conditions, participating families received therapy from two therapists: one for the adolescent and one for the parent. The I-CBT protocol focused on the adolescent (e.g., problem-solving, cognitive restructuring, behavioral activation, emotion regulation) parent (e.g., emotion regulation, limit-setting, monitoring), and family (e.g., communication, family problem-solving) skills. The TAU condition afforded therapists more latitude to deliver flexible, eclectic treatment modalities, common to community mental health settings. TAU clinicians implemented wide-ranging strategies, including supportive therapy, advice, skill discussion, and an emphasis on positive clinician-patient relationships (See Wolff et al, 2020 for more details).

Session frequency in both conditions varied based on clinical presentation and insurance coverage. Therapists met with families up to 3 times/week and sessions could range from 30 minutes to 3 hours, based on insurance stipulations for home-based services. Insurers approved treatment for 12 weeks, on average. Doctoral-level psychologists provided CBT training and ongoing weekly supervision to therapists on their I-CBT cases to promote adherence to CBT-based principles from the I-CBT manual throughout treatment. TAU therapists met weekly with the onsite masters-level supervisor who provided clinical supervision. The study assigned therapists to deliver interventions in only one condition but therapists in both conditions received supervision from the clinic supervisor on case management (e.g., insurance authorization procedures). Adolescents could be referred and evaluated for medication by clinic psychiatrists and received random urine drug screens over the course of treatment.

2.4. Measures

The study administered the substance use outcome measures at baseline, 3, 6, and 12 months. We administered all other measures at baseline only. Sociodemographic variables included age, biologic sex, race, sexual orientation, and Hispanic/Latinx ethnicity.

2.4.1. Substance use measures

2.4.1.1 Timeline Followback (TLFB; Sobell & Sobell, 1995) is a calendar-assisted daily drinking and drug use estimation method that research has frequently used to measure alcohol and marijuana use, including in treatment studies with youth (e.g., Waldron et al, 2001). The TLFB has shown satisfactory reliability and validity across multiple studies (see Sobell & Sobell, 2003 for a review). The current study recorded marijuana use per day as well as heavy drinking, defined as 4 drinks or more for females, and 5 drinks or more for males.

2.4.1.2 Adolescent Drinking Questionnaire (ADQ; Jessor, Donovan, & Costa, 1989) consists of 4 items, rated on an 8-point scale, that assess recent drinking frequency, quantity, frequency of heavy drinking (≥ 5 drinks per occasion), and frequency of intoxication. The ADQ has been widely used in survey research as well as in clinical studies (e.g., Spirito et al., 2011). The current study used the heavy drinking subscale as a baseline covariate for the TLFB heavy drinking variable.

2.4.1.3 Drug Use Questionnaire (DUQ) records the number of days that the adolescent used substances, including nicotine, marijuana, cocaine, LSD, etc., over the prior 30 days. Test/retest reliability for mean number of days that they used each substance has been shown to average .83 from 3- to 6-month follow-up and .94 from 6- to 12-month follow-up (Spirito et al., 2004). The current study used marijuana frequency as a baseline covariate for the TLFB marijuana frequency variable.

2.4.2. Adolescent measures

2.4.2.1 Aggression questionnaire.The Aggression Questionnaire (AQ; Buss & Warren, 2000) is a 34-item self-report measure of trait aggression that consists of the five dispositional subtraits of aggression, including psychological (anger, hostility) and behavioral (physical, verbal, indirect) aggression. Responses are given on a 5-point Likert scale, ranging from 1 (Not at all like me) to 5 (Completely like me). Research has reported good to moderate reliability (Buss & Warren, 2000). The current study used the total aggression score as the predictor in the current analyses to provide a broad indicator of aggression. Internal consistency of the total scale was excellent (α = .93) in this sample.

2.4.2.2 Child Depression Inventory – 2 (CDI-2; Kovacs, 2010) is a 28-item self-report measure of depressive symptoms rated on a 3-point Likert scale and corresponding to the prior two weeks. The CDI-2 total raw score ranges from 0 to 56, with higher scores indicating greater depressive symptom severity. The CDI-2, and its predecessor the CDI, have demonstrated good reliability and validity in clinical samples (e.g., Figueras, Amador-Campos, Gómez-Benito, & del Barrio Gándara, 2010; Hodges & Craighead, 1990). In the current study, baseline internal consistency was high (α = .91).

2.4.2.3 Difficulties in Emotion Regulation Scale (DERS; Gratz & Romer, 2004) is a widely used, 36-item measure of emotion regulation difficulties that has been validated for use with adults and adolescents. Items are rated on a 5-point Likert scale ranging from 0 (Almost never) to 4 (Almost always). The DERS assesses six dimensions of emotional dysregulation, including nonacceptance of emotional responses, difficulties engaging in goal-directed behavior, impulse control difficulties, lack of emotional awareness, limited access to emotion regulation strategies, and lack of emotional clarity. In the current study, both parents and adolescents completed the scale; the study used the total score in the analyses. This measure has previously demonstrated acceptable test-retest reliability (Gratz & Romer, 2004). Internal consistency in this study was excellent (α = .97).

2.4.2.4 UPPS Impulsive Behavior Scale- P (UPPS-P; Lynam, Smith, Cyders, Fischer, & Whiteside, 2007) is a 59-item self-report measure that assesses four dimensions of impulsivity, including positive and negative urgency (give in to strong impulses when experiencing either strong positive or negative emotions); lack of perseverance (inability to sustain attention on difficult or boring tasks); lack of premeditation (think through potential consequences of behavior before acting); and sensation seeking (preference for exciting and novel experiences). Items are rated on a 4-point scale ranging from 1 (Agree strongly) to 4 (Disagree strongly). The four subscales showed good internal consistency in the original study (Whiteside & Lynam, 2003). The positive and negative urgency subscales are analyzed here and internal consistency was .94 and .88, respectively, in this study.

2.4.2.5 Impulsive/Premeditated Aggression Scales (IPAS; Stanford et al., 2003) determines whether an individual is predominantly impulsive or premeditating in nature. The study asked adolescents to consider their aggressive acts over the prior six months and complete the 30-item IPAS in relation to those acts. Items are rated on a 5-point Likert scale from 0 (Strongly disagree) to 4 (Strongly agree). Ten of the items focus on impulsive aggressive characteristics (e.g., “I lost control of my temper during the acts”), and the current analyses used that subscale. The IPAS subscales have demonstrated excellent reliability and internal consistency in past work (Cruz et al., 2019). Internal consistency of the impulsive aggressive subscale in this study was high (α = .86).

2.4.2.6 Screen for Child Anxiety Related Disorders (SCARED; Birmaher et al. 1999; Birmaher et al. 1997) is a 41-item self-report questionnaire rated on a 3-point Likert scale. The total score ranges from 0 to 82. This measure has demonstrated good internal consistency and discriminant validity (Birmaher et al. 1997; Birmaher et al. 1999). In the current study, the alpha level was .95.

2.4.3. Parent and parenting measures

2.4.3.1 Brief Symptom Inventory (BSI; Derogatis and Melisaratos 1983; Derogatis 1993) is a self-report measure that assesses nine dimensions of psychiatric symptomatology: depression, anxiety, phobic anxiety, somatization, obsessive compulsive, interpersonal sensitivity, hostility, paranoid ideation, and psychoticism. Items are answered on a 5-point scale from 0 (Not at all) to 5 (Extremely). Internal consistency of the Global Severity Index, the variable used in this study, was high (α = .94).

2.4.3.2 Difficulties in Emotion Regulation Scale (DERS) - described above. Internal consistency for the DERS that parents completed was high (α = .94).

2.4.3.3 Parent Monitoring Questionnaire (PMQ; Stattin & Kerr, 2000) is a 24-item youth and parent report measure designed to assess parental monitoring and sources of parental knowledge (child disclosure, parental solicitation, parental control). The current study used the 9-item monitoring subscale. The study asked parents to rate their knowledge of their adolescent’s whereabouts and various daily activities. Items are rated on a 5-point Likert scale, with higher scores indicating higher levels of monitoring. Internal consistency in this study was high (α=.88).

2.4.3.4 Videotaped Family Assessment Task (FAsTask; Dishion, Nelson, & Kavanaugh, 2003) was used to assess monitoring, limit setting, and communication. To assess parental monitoring and listening, the teen leads a 5-minute discussion about a time without parental supervision and parents seek information about it. To assess limit setting, the parent and teen discussed a time when a limit was set on the teen’s behavior. If no limit was set, they discussed how they would address a problem behavior if it occurred in the future. A structured clinical code is used to derive variables. Norms were originally derived from 120 multi-ethnic families whose children had good school attendance and no behavior problems (Dishion et al, 2003).

A trained coder rated individual items on a 9-point scale (1=Not at all to 9=Very much), with higher scores reflecting greater levels of monitoring and limit setting. A total score was generated for each of these domains via an average of subscale items (Dishion & Kavanaugh, 2003). When both parents participated in the FAsTask, we averaged across both parents. Approximately 20% of recordings were double coded to calculate inter-rater reliability. Intraclass correlation coefficients (ICCs) for the two raters were .96 for monitoring and .95 for limit setting for the family as a whole.

2.4.4. Family measure

2.4.4.1 Family Assessment Device (FAD; Epstein, Baldwin, & Bishop, 1983) was used as a measure of parental perception of family functioning. The current study used the 12-item General Functioning subscale. Item scores ranged from 1 to 4, with lower scores representing better family functioning. The General Functioning subscale has been used in prior research as a brief measure of overall family functioning (e.g., Weinstock & Miller, 2010). The FAD has demonstrated good reliability and validity across ages and populations (Miller, Epstein, Bishop, & Keitner, 1985; Youngstrom, Youngstrom, Freeman, DeLosReyes, & Feeney, 2011). Internal consistency of the General Functioning subscale in the current study was strong (α=.92).

2.5. Data analyses

The study set the significance level at p < .05 and calculated effect sizes to improve ease of interpretation of the magnitude of effects across analyses. We used means and standard deviations to describe continuous variables, and frequencies to describe categorical variables. In our post–baseline analyses, we included data from months 3, 6, and 12. For longitudinal post–baseline group comparisons of continuous variables across months 3–12, we used hierarchical linear modeling (HLM) for repeated measures (Raudenbush & Bryk, 2002) since it can accommodate some missing data, covarying the baseline score. We used two dependent variables in all of our outcome analyses: percent of days of heavy drinking and percent of days of use of marijuana as measured by the TLFB. We controlled for baseline heavy drinking and marijuana use in the analyses. Due to missing follow-up data, we used maximum likelihood estimation to take advantage of all available information for each case (Hox,1999).

Because of the number of predictors, candidate variables, all measured at baseline, were screened by a priori groups, which the study team selected based on the research literature. The five variable groups were: (1) Socio-demographics, including gender, race (dichotomized to White vs. People of Color), Hispanic/Latinx ethnicity, and age; (2) Teen substance use, including MJ frequency (from the DUQ), and alcohol use frequency (from the ADQ); (3) Teen psychiatric symptoms/disorders, SCARED total score, CDI2 score, and AQ total; (4) Family/parent variables, including BSI total, DERS total score (parent) FAD general functioning, PMQ monitoring, FAsTask monitoring, and FAsTask limit setting; and (5) Teen reactivity variables, including DERS total score (adolescent), UPPS positive and negative urgency, and IPAS total score. So as not to overload our limited sample size, we conducted analyses to select the most promising predictors within each group, separately for each dependent variable. We selected the optimum predictors by backward elimination, starting with all the predictors in the group, dropping the weakest one at each step, ending with all variables in each category, if any, that had p-values of < .05. Last, the study team conducted a final series of analyses to select the best final predictors from the finalists in each of the variable groups, also by backward elimination.

Analyses of treatment moderators used the same two dependent variables, percent of days of heavy drinking from the TLFB and percent of days of use of marijuana from the TLFB. Based on correlational analyses, we selected variables from the four symptom groups described above (excluding socio-demographics) as the final potential baseline predictors: (1) Teen substance use: Because our primary outcomes were assessed using the TLFB, the baseline TLFB score was covaried in every model. The study used the ADQ and DUQ to test whether those different substance use measures predicted outcomes over and above the baseline TLFB score; (2) Teen psychiatric symptoms: CDI2 score and AQ; (3) Family/parent variables: FAD general functioning, Parent DERS, PMQ monitoring, FAsTask monitoring, and FAsTask limit setting; and 4) Teen reactivity variables: Teen DERS, IPAS total score, UPPS positive urgency, and negative urgency. Given the limited sample size, the study first tested each candidate moderator individually, ending with a combined model incorporating all moderators whose effects were significant at the p<.05 level.

3. Results

3.1. Baseline demographics and clinical characteristics

The sample was 58% male, 72% heterosexual, 85% White, and 31% Hispanic/Latinx (Wolff et al, 2020, for more details). No differences between treatment conditions existed on any of the socio-demographic variables.

Table 1 presents the means and standard deviations (or percentages) for the baseline assessment measures and the outcome variables by treatment condition. Table 2 presents the final results for the backward elimination analyses predicting percent days of heavy drinking. After the initial round examined predictors within each of the variable groupings, the only grouping that was significant was the Family/parent variables. The final predictor after backward elimination was FAsTask monitoring with a small to medium effect size.

Table 1.

Baseline Clinical Characteristics by Treatment Condition

Variable	EXP		TAU
	n	%	n	%	X2
DSM-IV Diagnoses
Alcohol Abuse	14	22.95	7	14.58	1.21
Alcohol Dependence	8	13.11	9	18.75	0.65
Marijuana Abuse	14	22.95	13	27.08	0.25
Marijuana Dependence	38	62.30	29	60.42	0.04
Other Substance Abuse	2	3.28	2	4.17	0.06
Other Substance Dependence	8	13.11	7	14.58	0.05
ADQ - % reporting a heavy drinking day in prior 90 days	25	51.02	29	47.54	0.44
	EXP		TAU
	n	M(SD)	n	M(SD)	t
Adolescent measures
Timeline Follow-back
Proportion of Alcohol use Days	61	0.07 (0.17)	49	0.09 (0.20)	0.56
Proportion of Marijuana use Days	61	0.36 (0.38)	49	0.39 (0.38)	0.38
DUQ - # marijuana use days past 90 days	61	37.39 (35.11)	49	35.08 (30.96)	0.36
Children’s Depression Inventory – 2	61	17.13 (10.21)	49	16.09 (10.04)	0.54
Screen for Child Anxiety-Related Disorders	60	21.53 (17.26)	47	24.21 (16.29)	0.82
Aggression Questionnaire total score	60	85.07 (24.02)	49	96.03 (26.02)	2.28
DERS (Adolescent) total score	61	97.91 (22.05)	48	96.27 (30.15)	0.33
UPPS-P Positive urgency	60	2.35 (0.81)	48	2.40 (0.70)	0.33
UPPS-P Negative urgency	60	2.75 (0.67)	48	2.85 (0.54)	0.79
IPAS total score	61	3.24 (0.97)	48	3.36 (0.10)	0.75 *
Parent measures
Brief Symptom Inventory total score	61	33.51 (29.01)	48	24.94 (36.61)	0.23 +
DERS (Parent) total score	59	77.87 (21.34)	48	79.91 (25.47)	0.45
PMQ – Monitoring subscale	60	29.23 (7.20)	48	30.23 (8.06)	0.68
FAD – General Functioning	60	2.08 (0.46)	48	2.16 (0.53)	0.86
FAST – Monitoring	55	6.09 (1.80)	45	5.51 (1.86)	1.58
FAST – Limit Setting	53	5.45 (1.90)	44	4.77 (2.10)	1.67

Note: EXP = Experimental treatment condition; TAU = Treatment-as-Usual condition

Note: df range from 105 – 108 due to some missing data

ADQ = Adolescent Drinking Questionnaire; DUQ = Drug Use Questionnaire; DERS = Difficulties in Emotional Regulation; UPPS = Urgency-Premeditation-Perseverance-Sensation Seeking; IPAS= Impulsive/Premeditated Aggression Scales; PMQ= Parental Monitoring Questionnaire; FAD = Family Assessment Device; FAST = Family Assessment Task

^*p <.01

⁺.05 > p <.10

Table 2.

Prediction of Percent Days of Heavy Drinking on the TLFB at 6 – 12 Month Follow-up by Baseline Variables: Backwards Elimination Final Results

Baseline Variable	B	SE	Lower 95%	Upper 95%	Effect Size	Error df	t	p
Time	−0.0044	0.0053	−0.0149	0.0060	0.000	90	0.84	.4010
% Days Heavy Alcohol Use on the ADQ	0.0334	0.0156	0.0025	0.0644	0.035	95	2.14	.0344
FAST Monitoring	−0.0034	0.0015	−0.0064	−0.0003	0.039	95	−2.	.0279

Note: SE = Standard Error of the regression coefficient B. Effect size is the adjusted partial eta squared, a measure of proportion of variance explained (Mordkoff, 2019). t = Student’s t. TLFB = Timeline Follow-back; ADQ = Adolescent Drinking Questionnaire FAST = Family Assessment Task

Table 3 presents the final results for the backward elimination predicting percent days of marijuana use. The initial round examined predictors within each of the variable groupings with significant predictors in the Family/parent and Teen psychiatric symptoms groupings. The final predictors after backward elimination were FAsTask monitoring (approaching a medium effect), teen Aggression Questionnaire (medium effect), and parent DERS (small-to-medium effect).

Table 3.

Prediction of Marijuana Frequency on the TLFB at 6 – 12 Month Follow-up by Baseline Variables: Backwards Elimination Final Results

Baseline Variable	B	SE	Lower 95%	Upper 95%	Effect Size	Error df	t	p
Time	0.0255	0.0505	−0.0749	0.1259	0.000	88	0.50	.6149
% Days MJ Use on the DUQ	0.2921	0.0683	0.1564	0.4278	0.154	91	4.28	<.0001
Aggression Questionnaire total score	0.0040	0.0010	0.0020	0.0060	0.133	91	3.96	.0001
FAST Monitoring	−0.0410	0.0146	−0.0704	−0.0120	0.067	91	−2.81	.0061
Parent DERS	0.0030	0.0011	0.0008	0.0051	0.062	91	2.72	.0079

Note: SE = Standard Error of the regression coefficient B. Effect size is the adjusted partial eta squared, a measure of proportion of variance explained (Mordkoff, 2019). t = Student’s t. TLFB = Timeline Follow-back; DUQ = Drug Use Questionnaire; FAST = Family Assessment Task; DERS = Difficulties in Emotion Regulation Scale total acore; MJ = marijuana

The study detected no significant moderators for percent days of marijuana use. Table 4 presents the results for the moderator analyses examining heavy alcohol use. After the study tested each candidate moderator individually, two Teen psychiatric symptom variables (IPAS Impulsive Aggression subscale and UPPS Positive Urgency subscale) and one Family/parent (FAsTask monitoring) were significant at the p<.05 level and entered into a combined model.

Table 4.

Moderation Effects by Treatment Condition on Percent Days of Heavy Drinking on the TLFB at 6 – 12 Month Follow-up by Baseline Variables

Baseline Variable	B	SE	Lower 95%	Upper 95%	Eff Size	Error df	t	p
Time	−0.0042	0.0051	−0.0144	0.0060	0.000	89	0.83	.41
% Days Heavy Alcohol Use on ADQ	0.0386	0.0154	0.0081	0.0692	0.169	87	2.52	.0137
Condition	−0.1293	0.0298	−0.1885	−0.0701	0.057	87	−4.34	<.0001
IPAS	−0.0037	0.0063	−0.0163	0.0089	0.000	87	−0.58	0.56
IPAS*Condition	0.0042	0.0075	−0.0106	0.0190	0.000	87	0.56	0.58
UPPS - Positive Urgency	−0.0117	0.0064	−0.0244	0.0010	0.000	87	−1.83	0.0703
UPPS - Positive Urgency*Condition	0.0166	0.0078	0.0011	0.0322	0.038	87	2.12	0.0365
FAST Monitoring	−0.0087	0.0023	−0.0134	−0.0041	0.038	87	−3.73	0.0003
FAST Monitoring *Condition	0.0110	0.0031	0.0048	0.0171	0.117	87	3.56	0.0006

Note: SE = Standard Error of the regression coefficient B. Effect size is the adjusted partial eta squared, a measure of proportion of variance explained (Mordkoff, 2019). t = Student’s t. TLFB = Timeline Follow-back; ADQ = Adolescent Drinking Questionnaire; IPAS Impulsive Aggression subscale; UPPS – Positive Urgency = Urgency-Premeditation-Perseverance-Sensation Seeking-Positive Urgency; FAST = Family Assessment Task; Condition: Integrated Cognitive-Behavioral Therapy, Treatment-as-Usual

As Table 4 shows, the study found moderator effects for the UPPS Positive Urgency subscale (small effect size) and FAsTask monitoring (medium effect). The two moderator effects are presented graphically in Figures 1 and 2.

Fig. 1.

Moderator effects of parental monitoring on heavy alcohol use by treatment condition.

Fig. 2.

Moderator effects of UPPS-P positive urgency subscale on heavy alcohol use by treatment condition.

4. Discussion

The current study extends prior research by identifying predictors and moderators of adolescent substance use outcomes across two community-based interventions: I-CBT and TAU. Regardless of treatment condition, baseline adolescent self-reported aggression and parental baseline emotional dysregulation predicted a greater percentage of adolescent marijuana use days over the 6-month follow-up period. Additionally, parental monitoring at baseline was inversely related to the percentage of marijuana use and heavy drinking days over the follow-up period, regardless of treatment condition. In terms of moderating effects, adolescents whose parents engaged in lower levels of monitoring at baseline reported a significantly lower percentage of days of heavy alcohol use if they received I-CBT compared to TAU. In addition, treatment condition moderated heavy drinking outcomes such that lower levels of baseline adolescent positive urgency (giving into strong impulses when experiencing positive emotions), were associated with a lower percentage of heavy alcohol use days in the I-CBT condition but not in TAU.

To put the findings into perspective, the adolescent predictor variables were mostly within the average or slightly elevated ranges while the parent predictors were more elevated. The mean score on the Aggression Questionnaire was below the clinical cut off point reported by the developers (Buss & Warren, 2000) while the UPPS Positive Urgency score was in the midrange for the scale (Lynan et al, 20070 Although the IPAS is typically used categorically, i.e., a person either demonstrates more impulsive or more premeditated aggression, the scores on the Impulsive aggression subscale were in the mid-range of possible scores.

With respect to parent predictors, the mean score of parents on the DERS approached the clinical cut-off range suggested in the literature (Hallion, Steinman, Tolin, & Diefenbach, 2018; Haynos, Roberto, & Attia, 2015). The mean score on the monitoring subscale of the PMQ was much lower than that found in other similar samples (Abar, Jackson, Colby, & Barnett, 2015).

Adolescents whose parents endorsed greater emotional dysregulation at baseline had a greater percentage of marijuana use days across treatment conditions. These findings are consistent with past work supporting the critical role of parents in adolescents’ emotional development and psychosocial adjustment (Repetti et al., 2002). Indeed, evidence suggests that parental emotion dysregulation is strongly linked to problems with emotion regulation strategies in youth (Zeman et al., 2013). Several processes may account for this association. First, parents experiencing dysregulated emotions may not teach or model effective emotion regulation strategies to youth. In addition, parents who experience problems with emotion regulation may engage in more frequent and intense expressions of negative affect. These displays may elicit affective overarousal and cognitive control difficulties in youth, promoting emotion dysregulation (Hoffman, 2000). Moreover, adolescent emotional dysregulation is associated with internalizing and externalizing problems (Southam-Gerow & Kendall, 2022; Silk et al., 2003), increasing risk for engagement in substance use. Past work demonstrates that youth with emotion regulation difficulties are more likely to use ineffective coping strategies, including marijuana use, to regulate their emotions and alleviate emotional distress (Wills, Simons, Manayan & Robinson, 2017).

In this sample, adolescent emotional dysregulation, in the form of aggression, predicted marijuana use. Several explanations for this finding are possible. First, aggression may be associated with negative emotion, such as anger, which marijuana may help to regulate. Adolescents who have problems controlling aggressive behavior may use marijuana to decrease aggressive behaviors, at least temporarily (Hoaken & Stewart, 2003). Aggression is also a common symptom of a conduct disorder diagnosis so this dimensional characteristic may be indicative of a broader underlying categorical diagnosis. Alternatively, aggression may be a symptom of underlying internalizing distress, and in such cases, marijuana may serve to reduce symptoms of depression and anxiety. Given that this sample had co-occurring mental health symptoms, both of these explanations might apply. In contrast to aggression, impulsivity in response to positive emotion also predicted marijuana use. This finding might be driven by peer influence, but regardless suggests that different emotions, and motives, can drive marijuana use and must be considered in treatment planning.

Parental monitoring, as assessed by a videotaped interaction task, but not self-report, was the other significant predictor of marijuana use and heavy alcohol use outcomes following treatment for substance use. These results align with past work that offers support for protective effects of parental monitoring on adolescent drinking (Fagan, Van Horn, Hawkins & Jaki, 2013; O’Brien, Hernandez & Spirito, 2015; Ryan, Jorm, & Lubman, 2010) and marijuana use (Lac & Crano, 2009).

In terms of moderating effects, the study found low parental monitoring upon entry into treatment to be associated with a decreased percentage of adolescent alcohol use days in the I-CBT condition compared to TAU. If parents are already adequately monitoring their teen, the more structured I-CBT condition does not appear to add a measurable difference to adolescent heavy drinking outcomes. However, if parents are less consistent in their monitoring, the I-CBT treatment improved outcomes. Several potential explanations for this finding may exist. First, the I-CBT protocol specifically targets parenting skills and evidence suggests that parent/family work is often neglected by community practitioners and leads to more negative outcomes (Farahmand et al., 2012; Spirito et al., 2011). In parent-only sessions, I-CBT therapists specifically focused on improving parental monitoring, so parents in the I-CBT condition may have been more likely to supervise adolescents’ social interactions and promote socialization with appropriate peer groups. These parents may also have been more likely to detect adolescent drinking behavior and/or more likely to deliver consequences for drinking, deterring adolescents from engaging in heavy drinking in the future.

The fact that the study detected the parental monitoring result only with observational, not self-report, measures in both the predictor and moderator analyses, is an interesting research methodology finding with direct implications for patient care. Research has found self-report parenting measures to have methodological weaknesses, including variable psychometric properties and parental interpretations/biases (e.g., over-reporting positive aspects of their parenting). In contrast, observational approaches offer real-time sampling of behavior as well as a snapshot of how rapid and dynamic parental behaviors unfold and are altered by youth reactions (Gardner, 2000). Thus, use of FAsTask observational codes in this study offered a more nuanced assessment of parental monitoring and may be important to incorporate in future research examining parenting, especially for youth who report co-occurring mental health symptoms and substance use. One advantage of observer ratings is the ability to understand the contextual cues that help to differentiate coercive monitoring, versus concerned supervision, which can result in very different responses from an adolescent.

Finally, the study found low levels of baseline positive urgency, or giving in to strong impulses when experiencing positive emotions, to be associated with a lower percentage of heavy alcohol use days in I-CBT compared to TAU. At high levels of positive urgency, neither treatment had much impact on outcomes, suggesting that once positive urgency reached a certain threshold, the interventions offered were ineffective. Thus, while the skills taught in I-CBT may have aided in tampering heavy drinking among those with less tendency to drink heavily in response to positive emotions, more skill practice or a different set of skills may be needed for those with stronger positive urgency. Meta-analytic reviews have revealed that positive urgency is associated with greater alcohol consumption and problematic alcohol use in adolescent samples (Stautz & Cooper, 2013). Compared to other impulsivity-related traits, positive urgency showed among the largest associations with these alcohol outcomes (Stautz & Cooper, 2013). Therefore, during adolescence, excessive alcohol consumption and problematic use may be driven, in part, by the tendency to act impulsively when in a positive mood state.

This study has several limitations. First, the relatively small sample size limited the number of predictor variables that we could analyze. We selected variables on a conceptual basis but there could be other factors that could account for the findings in this study. For example, research has found sensation seeking to be a strong predictor of adolescent substance use (Vassileva & Conrod, 2019). Missing data at follow-up may also have affected analyses. Second, the small percentage of LGBT and participants of color precluded their inclusion as interaction terms in the moderator analyses. Additionally, this study did not include factors associated with race and/or ethnicity (i.e., sociodemographic status, such as parental formal education, income). Third, the findings reported here may not be generalizable to other samples, such as adolescents who use substances but do not have clinically significant levels of psychiatric symptomatology.

5. Conclusion

Taken together, this research highlights the importance of examining individual youth, parent, and parenting/family characteristics for understanding treatment response among youth with co-occurring disorders. Information derived from predictor and moderator analyses is clinically valuable for several reasons. First, predictor variables provide general prognostic information about how an individual might respond to any intervention. For clinicians, this broad knowledge may prove useful in generating patient goals, managing treatment expectations, and monitoring progress within a therapeutic setting. In addition, moderator identification could inform the development of intake assessment protocols, for both parents and adolescents, and guide intervention-tailoring within treatment as a way to maximize factors associated with better outcomes for specific patients.

Our findings highlight the importance of addressing and enhancing parental monitoring of teens with co-occurring mental health and substance use symptoms and the added benefit of using structured parent training, as reflected by the I-CBT protocol, to do so. In addition, this study also underscores the importance of considering both parent and adolescent functioning, in this case, emotion regulation in particular, when devising treatments for adolescents, especially in those with complicated presentations and comorbid conditions.

Highlights.

Regardless of treatment condition, low parental monitoring at baseline, as assessed by a videotaped interaction task, but not self-report, predicted adolescent marijuana and heavy alcohol use days over a 6-month follow-up period.

If parents entered treatment with low levels of parental monitoring, adolescents in the CBT condition reduced their percentage of heavy alcohol use days significantly more than adolescents in treatment-as-usual over a 6-month follow-up period.

Regardless of treatment condition, greater adolescent aggression and parental emotion dysregulation at baseline predicted greater marijuana use days over a 6-month follow-up period.

Adolescents in the CBT condition who reported low positive urgency at baseline reduced their heavy alcohol use days significantly more than adolescents in treatment-as-usual over the 6-month follow-up period.