Association of Mental Health Symptoms and Peer Behaviors with Risk for Substance Use and Condomless Sex among Youth in Juvenile Drug Court

Abstract

Juvenile drug courts are a growing response to adolescent substance use, but a better understanding of modifiable risk factors is needed to improve program outcomes. Youth’s mental health symptoms and peers’ activities may impede the effectiveness of these “therapeutic” courts. In a unique longitudinal sample of 105 adolescents involved in juvenile drug court, we find elevated internalizing symptoms and deviant behavior of peers were each associated with increased risk of alcohol and marijuana use. Similar effects were seen on risk for condomless sex. Mental health and peer behaviors should be intervention targets for evidence-based juvenile drug court programming.

Juvenile Drug Courts: A Growing Public Health Opportunity

Communities across the United States increasingly rely on JDCs to address the problem of adolescent drug use (Shaffer, 2011). As of December 2014, there were 420 JDCs in 44 States and the District of Columbia, serving thousands of youth annually (National Association of Drug Court Professionals, 2014; National Institute of Justice [NIJ], 2016a). Modeled after adult drug courts, these “therapeutic” special court dockets for young people involved with the criminal justice system because of substance misuse screen and assess adolescents for substance use problems, coordinate needed services across agencies, help youth and families connect with services, promote service engagement, and assist with transition out of court services and into long-term supports if necessary (NIJ, 2016b). Adolescents involved in JDCs engage in many harmful health behaviors and are at risk for poor health outcomes, including serious substance use (Belenko & Dembo, 2003; Tolou-Shams, Brown, Gordon, Fernandez, & Project SHIELD Study Group, 2007); high-risk sexual behavior, including sex without a condom and sex in exchange for drugs or money (Belenko & Dembo, 2003; Tolou-Shams et al., 2007; Belenko et al, 2008; Castrucci & Martin, 2002; Letourneau, McCart, Asuzu, Mauro, & Sheidow, 2013); and HIV and other sexually transmitted infections (Belenko & Dembo, 2003). As such, JDCs aim to serve an important and growing public health function.

The Need for Improved Juvenile Drug Court Programming

There is strong evidence that adult drug courts can cut rates of criminal recidivism by nearly half and that substance use outcomes also are very positive (Mitchell, Wilson, Eggers, & MacKenzie, 2012). However, research on JDCs is decidedly more mixed (Mitchell et al., 2012). As noted in reviews of the JDC literature (Henggeler & Marlowe, 2010; Tanner-Smith, Lipsey, & Wilson, 2016), most randomized and quasi-randomized studies of JDCs have indicated these programs produce only small reductions in general recidivism and substance use among youth. A recent federal investigation of nine JDCs found that youth in seven of these courts were actually at higher risk for recidivism than a matched-comparison group of young people placed on probation (Latessa, Sullivan, Blaire, Sullivan, & Smith, 2013; Blaire, Sullivan, Latessa, & Sullivan, 2015). The investigation also found that these seven JDCs referred youth to non-evidence-based talk therapy or education programs for substance use treatment (Latessa et al., 2013). In the two courts that referred to evidence-based treatment services, there was evidence that drug court was beneficial. This is consistent with other research indicating that JDCs are more effective when they incorporate evidence-based treatment services into the program (Henggeler, Halliday-Boykins, Cunningham, Randall, Shapiro, & Chapman, 2006; Schaeffer, Henggeler, Chapman, Halliday-Boykins, Cunningham, Randall, & Shapiro, 2010). Indeed, recognizing the importance of scientific evidence in therapeutic programming, the federal Office of Juvenile Justice and Delinquency Prevention (2016) recently published a set of treatment guidelines for juvenile drug courts designed to promote effective treatment practice that reduce substance use and criminogenic behavior. However, even these recommendations for the most part do not discuss sexual health or target youths’ risky sexual behavior.

A better understanding of the specific modifiable risk and protective factors that contribute directly to outcomes of youth participating in JDC could help improve services. Such specific findings would carry more weight with policy makers and court officials, who have limited resources to improve services for this high-risk population. Enhancements and modifications to JDC programs could then focus on results for this specific population rather than results from general populations of youth offenders. Therefore, as JDCs proliferate, there is growing need to understand factors associated with participants’ continued substance use and risky sexual behavior. This understanding will improve the ability of JDCs to fulfill their dual public health and justice functions.

Mental Health and Peer Behaviors as Proximal Intervention Targets

This paper examines two important factors that deserve attention as potential targets for improving the public health and justice outcomes of JDCs: a) underlying mental health problems that may both impair judgment during sex and motivate youth to self-medicate with substances; and b) the peer networks of youth involved in JDCs.

Mental health

While there are, to our knowledge, no prevalence estimates of psychological disorders in the JDC population specifically, several studies have identified elevated rates of mental health problems among youth in the juvenile justice system. One study found 70% of youth involved with juvenile court have a diagnosable mental disorder (Shufelt & Cocozza, 2006). Disruptive behavior disorders (e.g., attention deficit hyperactivity disorder [ADHD]) and substance use disorders were particularly common with prevalence rates around 50%, but internalizing disorders such as depression and anxiety were highly prevalent as well, with anxiety present in a quarter of all boys and more than half of all girls. Other analyses have found mental and behavioral disorder prevalence rates as high as 50% among youth on probation (Wasserman, McReynolds, Ko, Katz & Carpenter, 2005), and about 30% among boys in juvenile detention (Wasserman, McReynolds, Lucas, Fisher, & Santos, 2002).

Considerable research has examined associations between substance use and major mental health problems in youth, including depression, anxiety, and posttraumatic stress disorder (PTSD). In fact, studies have documented strong associations between these mental health domains and substance use in both community-based (Armstrong & Costello, 2002; Wolitzky-Taylor, Bobova, Zinbarg, Mineka, & Craske, 2012) and justice-involved (Abram, Teplin, McClelland, & Dulcan, 2003; Davis, Dumas, Wagner, & Merrin, 2016) youth samples.

For youth in the general population, depression, anxiety, and PTSD also have all been linked with adolescents’ engagement in sexual risk behaviors (Stiffman, Dore, Earls, & Cunningham, 1992). For justice-involved youth specifically, research has focused predominantly on the association between depression and risky sex, and results have thus far been mixed (Voisin, Hong, & King, 2012). For example, in one study of adolescent offenders (Lucenko, Malow, Sanchez-Martinez, Jennings, & Devieux, 2003), those scoring above the clinical cut-off on a depression screener reported having more sexual partners and higher rates of unprotected sex relative to offenders scoring below the cut-off on the depression screener. However, in another study with justice-involved youth (Elkington et al., 2008), those with major depressive disorder where significantly less likely to report engaging in unprotected sex relative to those with a substance use disorder or neither disorder. Thus, more research clearly is needed to examine the potential impact of depression (as well as anxiety and PTSD symptoms) on risky sex among justice-involved adolescents, including those involved in JDC programs.

Peer delinquent and prosocial behaviors

There also is experimental evidence that adolescents tend to engage in more risk-taking behavior when in the company of peers (Gardner & Steinberg, 2005). This evidence has been extended through correlational studies to both substance use and risky sex. Indeed, research shows that adolescents who associate with delinquent peers are more likely to use drugs and alcohol (Fergusson, Swain-Campbell, & Horwood, 2002; Steinberg, Fletcher, & Darling, 1994) and to engage in unsafe sexual behavior, such as sex with multiple partners and sex without a condom (Voisin et al., 2012). Unfortunately, the potential protective effects of positive peer influence have received far less attention in the empirical literature. In one study, Prinstein, Boergers, and Spirito (2001) failed to find any association between adolescents’ use of substances and the involvement of their peers in prosocial activities (e.g., participation in school clubs or athletic programs). However, others have found that youth are less likely to use substances if they have prosocial peers who actively discourage such use (Coyle, Bramham, Dundon, Moynihan, & Carr, 2016). To the best of our knowledge, studies have yet to examine the potential link between peer prosocial activities and adolescents’ engagement in sex-risk behavior. Research also is lacking on the influence of peers among youth involved in JDCs, which highlights the contributions of the present study.

Need for Research

Taken together, findings suggest that mental health problems and peer influences might be important proximal targets for improving the effectiveness of JDCs to achieve a range of public health outcomes. Studies, however, have not thoroughly examined the role of these factors for justice-involved adolescents who have substance-related problems, and particularly have not had longitudinal data to examine these factors during intervention. Further, to our knowledge there are no studies that document the effects of these factors on substance use and sexual risk behaviors of youth in JDCs. In fact, most existing research on this population does not examine the determinants of juveniles’ sexual risk behavior at all, focusing instead only on their drug use and recidivism (Henggeler & Marlowe, 2010). Thus, there is a weak evidence base to inform scientists and policy makers who seek to improve the public health impact of JDC programs.

This paper examines associations between mental health and peer behaviors and substance use and risky sex among youth in JDCs. It uses six waves of data collected as part of a randomized trial of a behavioral intervention for youth in JDCs (Letourneau, McCart, Sheidow, & Mauro, 2017). These data offer the opportunity to answer several novel questions. First, because the original intervention targeted HIV risk, this study can examine determinants of risky sex. Second, this longitudinal dataset offers a unique opportunity to describe trends in substance use and risky sex over a full 18-month period beginning at the start of youths’ court involvement until after court supervision. Third and perhaps most important, with up to six repeated measurements on each youth over the trial period, this study offers a unique opportunity to control for a youth’s underlying predisposition for risky behavior. It also offers an opportunity to control for secular trends in substance use and sex behavior influenced by court involvement (for example, if all youth stop substance use and then reinitiate following the end of court involvement). This makes it possible to better isolate the association between mental health and peer behaviors and substance use and risky sex.

We hypothesize that, after controlling for sex, age, temporal fluctuations that may be influenced by the court process, and a youth’s underlying predisposition for risky behavior, at assessments when a youth reports more mental health symptoms, he or she will also be more likely to report engaging in substance use and risky sexual behaviors in the past 90 days. We further hypothesize that, when youth or their caregivers report that peers engage in delinquent activity, youth will be more likely to report engaging in substance use and risky sexual behaviors. Finally, we hypothesize that, when youth or their caregivers report that peers engage in conventional activities like participating in school activities, they will be less likely to report engaging in substance use and risky sexual behavior.

Methods

Design

The data for this analysis come from a completed randomized clinical trial (RCT) evaluating the effectiveness of a substance use and sexual risk reduction intervention for youth involved in JDC (Letourneau, McCart, Sheidow, & Mauro, 2017; ClinicalTrials.gov Identifier: NCT01511380). This study was conducted with two courts in the southeastern United States.

All youth who entered the courts as new referrals were screened by research staff for study eligibility. Following screening, research staff met with youth and their caregivers to describe the study and obtain informed consent. Participants in the trial completed interviews and outcome measures at each of six time points – at baseline assessment, and at follow-up assessments occurring at 3-, 6-, 9-, 12-, and 18-months post-baseline. Trained research assistants administered clinical measures at times and places convenient to families. Families were compensated $30 for completing each assessment. The Institutional Review Board at the Medical University of South Carolina approved all study procedures prior to data collection.

Youth enter JDC primarily through referral from juvenile justice authorities following arrest and adjudication for a substance-related offense. To have been included in the trial, youth had to a) be 12–17 years of age (one eleven year old was included), b) have formal or informal probationary status, and c) be fluent in English and have a caregiver fluent in English. Children were excluded if they had gross neurological problems, significant medical disorders, significant intellectual disabilities, or active psychosis.

The full randomized trial began with 105 participants at baseline, 40 of whom were assigned to the “treatment” condition and 65 of whom were assigned to the “control” condition. The treatment condition targeted substance use and risky sex; the control condition targeted substance use. Neither condition targeted mental health problems or peer behaviors directly. There was dropout at each of the five follow-up assessments particularly between the final two assessments. Whereas 83% of participants completed the 12-month follow-up, only 56% completed the final 18-month follow-up. Age, internalizing symptoms, and engaging in sex at baseline were weakly associated with dropping out of the study by assessment six; mixed effects models are used to account for potential differential dropout by exposure status (see “Analytic Strategy). The demographics of the study sample at each assessment are shown in Table 1.

Table 1.

Demographics of Full Study Sample by Visit

Study Visit	Assessment 1	Assessment 2	Assessment 3	Assessment 4	Assessment 5	Assessment 6
Population	105	96	95	92	87	59
Age	14.89 [1.4]	14.88 [1.38]	14.87 [1.39]	14.87 [1.41]	14.87 [1.4]	15.24 [1.19]
Male	88 (84%)	81 (84%)	80 (84%)	77 (84%)	73 (84%)	50 (85%)
Any Sex	37 (35%)	33 (35%)	34 (38%)	30 (38%)	35 (44%)	27 (51%)
Unprotected Sex (Self-Report)	12 (12%)	10 (11%)	9 (10%)	12 (15%)	11 (14%)	12 (23%)
Marijuana Use (Self-Report)	91 (87%)	18 (19%)	7 (8%)	13 (16%)	19 (24%)	26 (49%)
Alcohol Use (Self-Report)	42 (40%)	8 (9%)	5 (6%)	4 (5%)	6 (8%)	8 (15%)
Internalizing Symptoms (Self-Report)	0.36 [1.03]	0.07 [1]	0 [1]	−0.07 [1.05]	−0.31 [0.85]	−0.25 [0.86]
Depressive Symptoms (Self-Report)	0.25 [1.08]	0.07 [0.98]	0.1 [1.06]	0 [1.03]	−0.27 [0.82]	−0.37 [0.81]
Anxiety Symptoms (Self-Report)	0.35 [1.13]	0.07 [1.05]	−0.09 [0.93]	−0.09 [1.06]	−0.27 [0.77]	−0.13 [0.79]
Trauma Symptoms (Self-Report)	0.34 [1.1]	0.04 [1]	−0.03 [0.97]	−0.03 [1.03]	−0.29 [0.84]	−0.2 [0.84]
Peer Delinquency (Self-Report)	0.37 [1.04]	−0.06 [0.99]	−0.13 [0.91]	−0.05 [0.95]	−0.25 [0.95]	0.23 [1.1]
Peer Delinquency (Caregiver-Report)	0.45 [1.07]	−0.2 [0.85]	−0.06 [1.06]	−0.29 [0.81]	−0.08 [0.99]	0.26 [1.07]
Peer Conventional Activities (Self-Report)	0.01 [0.82]	0.09 [0.99]	−0.04 [0.98]	−0.02 [1.03]	0.05 [1.13]	−0.13 [1.08]
Peer Conventional Activities (Caregiver-Report)	−0.16 [1.06]	0.12 [0.91]	0.02 [1.02]	−0.05 [1.07]	0.13 [0.99]	−0.11 [0.97]

Note: Scale scores are log-transformed, centered at their mean, and scaled by their standard deviation.

A note on the study sample

Scales measuring peer delinquent and prosocial behaviors were introduced partway through the trial, so a subset of youth did not complete these scales. Furthermore, at some study assessments, some youth did not complete the scales or tests for other exposure or outcome measures. Because this study begins with a small sample, to maximize statistical power, each analysis uses the largest available subset of study data. For example, a youth who responded to questions about risky sex and completed all mental health scales but did not complete the peer behavior scales would be included in the analyses of the association between risky sex and mental health, but not in the analysis of the association between risky sex and peer behaviors. The demographic composition of the sample used in each regression model is shown in the Appendix (Tables 3 – 6).

Outcomes of Interest

There are four outcomes of interest, two pertaining to substance use, and two pertaining to sexual risk:

1. Any self-reported alcohol use in the past three months.

2. Any self-reported marijuana use in the past three months.

3. Any self-reported vaginal or anal sex in the past three months.

4. Any self-reported vaginal or anal sex without a condom in the past three months.

Substance use

Self-reported alcohol and marijuana use were examined using a variation of the Form 90 (Miller, 1991), which is an interview based on the timeline follow back methodology of quantifying specific amounts of substances consumed by individuals during the previous 90 days. Research with adolescents indicates that the timeline follow back method is reliable (Waldron, Slesnick, Brody, Turner, & Peterson, 2001) and yields data that correspond with biological markers and collateral reports of youth substance use (Donohue, Azrin, Strada, Silver, Teichner, & Murphy, 2004). Because the distribution of frequency of use was highly skewed, we dichotomized the two substance use outcomes to differentiate any vs. no self-reported marijuana use and any vs. no self-reported alcohol use in the last 90 days.

Marijuana use was also assessed via urine drug screens (UDS). The specific test was the “Integrated Key Cup” supplied by BioTechNostix, with a detectable level for cannabis of 50 ng/ml, and a sensitivity of 80%. However, this study relied primarily on self-reported marijuana use because the time window of marijuana use assessed by UDS was shorter than the 90-day window assessed in timeline follow back interviews and therefore, UDS marijuana assessments would not cover the entire three-month period separating study assessments and would not be comparable to alcohol assessments. Nevertheless, as described later, a marijuana sensitivity analysis was conducted using the UDS assessments.

Risky sex

Sexual risk behaviors were assessed using a standardized set of items validated across numerous studies with at-risk adolescents (Jemmott, Jemmott, & Fong, 1992; Jemmott, Jemmott, Spears, Hewitt, & Cruz-Collins, 1992; Jemmott, Jemmott, Fong, & McCaffree, 1999). At baseline, youth reported on their lifetime history of vaginal and anal sex. At baseline and all follow-up assessments, youth reported whether they had engaged in vaginal or anal sex in the past three months and, if so, the number of intercourse acts and number of acts in which condoms were used. We created dichotomous variables that indicated “any” versus “no” sex in the past three months and “any” versus “no” sex without a condom in the past three months.

Predictors of Interest

Mental health symptoms

The primary mental health predictors were symptom counts for three internalizing mental disorders – depressive symptoms, anxiety/fear symptoms, and post-traumatic stress symptoms. At each assessment point, these symptoms were assessed using three subscales of the Global Appraisal of Individual Needs (GAIN; Dennis, White, Titus, & Unsicker, 2008). The GAIN is used in hundreds of peer-reviewed studies (GAIN, 2015), and its mental health subscales exhibit good internal consistency and validity (Dennis et al., 2008). The 9-item depressive symptom scale assesses the presence of mood, anhedonia, and common somatic symptoms. The 12-item anxiety/fear scale assesses the presence of common psychological and somatic (e.g., heart racing) symptoms of panic, as well as common phobias (e.g., fear of open spaces). The 12-item post-traumatic stress scale assesses the presence of common symptoms of PTSD like flashbacks, nightmares, and numbness. Symptom scales were scored as the total number of symptoms present. Separate subscale scores were also summed to form a “total internalizing symptoms” score.

Peer delinquent and prosocial behaviors

Peer relations were measured using scales developed for the Pittsburgh Youth Study (Loeber, Farrington, Stouthamer-Loeber, & Van Kammen, 1998). These scales have high reliability and validity (e.g., Pardini, Loeber, & Stouthamer-Loeber, 2005). The Peer Delinquency and Drug Activity Scales assess the proportion of youths’ friends who engage in various antisocial behaviors. The 11-item Peer Delinquency Scale assesses general delinquency/criminal behavior (e.g., strong armed robbery, destruction of property) commited by peers during the past 90 days, and the 4-item Peer Drug Activity Scale assesses peer drug related behaviors (e.g., used alcohol, sold drugs) during that same time frame. The Prosocial Activities of Peers Scale is an 8-item measure that assesses the proportion of participants’ friends who engage in prosocial activities at school (e.g., athletics, clubs), in the community (e.g., church groups), and at home (e.g., doing things with family members). Consistent with prior research, both the youth and primary caregiver completed each scale. Respondents were asked if “none,” “a few,” “half,” “more than half,” or “all” of the youth’s friends engaged in each behavior or interest. Each item response was converted to a corresponding integer ranging from 0 (none) to 4 (all). These integers were averaged over all questions to form a scale score (Pardini, Loeber, & Stouthamer-Loeber, 2005).

Each exposure measure was log-transformed to reduce skew. To ease comparison across scales, we created z-scores for each exposure measure by subtracting the overall mean score (based on all participants and all study assessments) and dividing by the standard deviation.

Demographic Covariates

We recorded participants’ age (in years), sex (male/female), and the study assessment at which data were collected.

Analytic Strategy

Trends over time and associations between outcomes of interest and exposures of interests were examined using generalized linear mixed effects models using a random intercept (Fitzmaurice, Laird, & Ware, 2011). Random intercept models extend regression models to the analysis of clustered data such as repeated measures in a longitudinal study. These models allow each participant in a longitudinal study to have his/her own model intercept, effectively accounting for a study participant’s unmeasured “underlying propensity” for experiencing the outcome. All other model coefficients are treated as “fixed,” i.e. the association between each exposure and the outcome does not vary across participants. In addition to the ability to account for an underlying predisposition for experiencing the outcome, an advantage of mixed effects models is that they yield unbiased estimates even when the probability of dropping out of a study is associated with a past exposure as is true in this cohort (Bell, Kenward, Fairclough, & Horton, 2013); thus these models are a strong analytic approach in studies (like this one) with high risk samples that have a degree of dropout over time. As a sensitivity analysis, we also reanalyzed all results excluding the 18-month study visit, which had only 56% participation.

All regression models used logistic regression for binary outcomes. First, temporal trends in the outcomes were examined by regressing each outcome on a cubic polynomial of time (in months) from the baseline assessment with no other covariates; this approach flexibly accommodates non-linear trends over time. Separate logistic regression models were then fitted for each possible exposure-outcome combination. All exposure/outcome regression models also controlled for sex (dichotomous), age (analyzed as a linear continuous variable), and for time trends as described above. Regression fixed effect coefficients were exponentiated for interpretation as odds ratios. These odds ratios have a conditional interpretation – they represent the average effect of a participant’s exposure on his/her outcome probability after accounting for age, sex, shared temporal trends that may have been influenced by the court process, and that participant’s underlying predisposition to experience the outcome. Because exposures were transformed to z-scores, regression odds ratios can be interpreted as the average increase in the odds of the outcome for each one standard deviation increase in the participant’s logged exposure scale of interest. Because of the many models we fit, differentiation from a null effect (statistical significance) was assessed using Wald tests on the coefficients at a conservative alpha = .01 level rather than the traditional .05 level.

Analyses were conducted in R 3.3.1. Mixed effects models were estimated using “lme4” (Bates, Maechler, Bolker, & Walker, 2015) using restricted maximum likelihood estimation.

Results

Trends Over Time

Estimated individual and population average trends in substance use and sex behaviors are shown in Figure 1. Rates of past 90-day marijuana and alcohol use were unsurprisingly very high at baseline, immediately following entry into drug court. Self-reported use then dropped sharply to near zero, but began to rebound at 12- and 18-month assessments, following the end of drug court involvement. By contrast, rates of any sex and condomless sex increased only slightly over the course of the study, as would be expected as youth age, and do not appear to be influenced by the drug court program.

Figure 1.

Population Average and Individual Estimated Substance Use and Sex Behavior Trajectories for 105 Youth in Juvenile Drug Court over 18-Month Study Period

Summary of Associations

Results of exposure/outcome analyses are shown in Table 2 (with odds ratios and 99% confidence intervals). In all cases, the direction of the observed association is consistent with the proposed hypotheses, but not all associations achieved statistical significance. Specifically, while there was strong evidence for an association between mental health symptoms and risky sex, only post-traumatic stress symptoms were significantly associated with substance use. Peer delinquency was by all measures significantly associated with substance use, and by some associated with risky sex. Effects of peer conventional activities were mostly not significant. Associations significant at the p<.01 level are described in greater deatail below; non-significant associations all trended in the expected direction, and in a direction consistent with other results.

Table 2.

Associations Between Mental Health Symptoms and Peer Activities and Substance Use and Sex Behaviors

	Any Sex	Unprotected Sex	Self-Reported Marijuana Use	Self-Reported Alcohol Use
Total Internalizing Symptom Score	1.86 [1.12,3.11]	2.79 [1.48,5.26]	1.41 [1.05,1.91]	1.75 [1.19,2.58]
Depressive Symptom Score	1.19 [0.74,1.93]	2.26 [1.24,4.1]	1.24 [0.93,1.64]	1.4 [0.97,2.02]
Anxiety/Fear Symptom Score	1.85 [1.16,2.95]	2.34 [1.36,4.04]	1.25 [0.94,1.67]	1.24 [0.88,1.75]
Post-traumatic Stress Symptom Score	1.81 [1.11,2.96]	2.54 [1.41,4.57]	1.4 [1.05,1.88]	1.72 [1.2,2.47]
Peer Delinquent Behavior Score (Self-Report)	1.63 [0.86,3.09]	2.14 [1.09,4.21]	1.81 [1.24,2.63]	2.36 [1.38,4.04]
Peer Delinquent Behavior Score (Caregiver-Report)	1.97 [1,3.85]	1.67 [0.84,3.29]	2.34 [1.49,3.69]	2.24 [1.29,3.92]
Peer Conventional Activity Score (Self-Report)	0.63 [0.33,1.21]	0.39 [0.14,1.1]	0.8 [0.57,1.14]	0.69 [0.4,1.21]
Peer Conventional Activity Score (Caregiver-Report)	0.4 [0.18,0.87]	0.3 [0.06,1.38]	0.73 [0.51,1.05]	0.67 [0.41,1.1]

Note: Odds ratios correspond to the increase in odds associated with a one standard deviation increase in the logged scale.

Note: All odds ratios are adjusted for age, sex, and time from baseline.

Note: Intervals are 99% confidence intervals. Results Significant at the p < .01 level are bold.

Substance Use

Total internalizing symptoms were significantly associated with increased past 90-day use of alcohol (Adjusted Odds Ratio (aOR): 1.75, 99% Confidence Interval (CI): 1.19 – 2.58) and marijuana (aOR: 1.41, 99% CI: 1.05 – 1.91). There were positive, significant associations between post-traumatic stress symptoms specifically and both alcohol use marijuana use (see Table 2 for detailed odds ratios).

There was a positive significant association between delinquent peer activity and alcohol use, as reported by youth (aOR: 2.36, 99% CI: 1.38 – 4.04) and as reported by caregivers (aOR: 2.24, 99% CI: 1.29 – 3.92). There was also a positive significant association between delinquent peer activity and marijuana use, as reported by youth (aOR: 1.81, 99% CI: 1.24 – 2.63) and caregivers (aOR: 2.34, 99% CI: 1.49 – 3.69). Peer prosocial behaviors were all negatively but not significantly associated with alcohol and marijuana use (see Table 2 for detailed odds ratios).

Risky Sexual Behavior

There was a positive significant association between total internalizing symptoms and any sex in the past three months (aOR: 1.86, 99% CI: 1.12 – 3.11) and a stronger positive association between total internalizing symptoms and condomless sex (aOR: 2.79, 99% CI: 1.48 – 5.26). There were also significant positive associations between all three symptom scale scores and condomless sex; anxiety and post-traumatic stress symptoms were also associated with increased odds of any sex in the past 90 days (see Table 2).

Peer delinquent behaviors were significantly and positively associated with any sex when peer behaviors were reported by caregivers (aOR: 1.97, 99% CI: 1.3 – 3.85), and with condomless sex when reported by youth (aOR: 2.14, 99% CI: 1.09 – 4.21). Finally, peer prosocial behaviors were significantly negatively associated with any sex when reported by caregivers (0.40, 0.22 – 0.72).

Sensitivity Analyses

In all cases, effects were in the same direction and of similar magnitude in the sensitivity analysis excluding the 18-month visit. The sensitivity analysis of marijuana use using UDS showed all effects in the same direction as the analysis using self-report.

Discussion

Juvenile drug courts serve dual public health and justice functions, and have become increasingly common, but outcomes of these programs are mixed. Indeed, in this sample of drug-court involved adolescents, the prevalence of marijuana and alcohol use fell sharply following entry into drug court, but and then rose again over the course of the 18 months of follow-up, with the sharpest increases after regular urine testing ended. The prevalence of sex increased slowly and steadily, consistent with sexual development over adolescence. Our findings offer guidance to JDC proponents and professionals who seek to improve the public health impact of drug court programming by identifying potential therapeutic targets that could be the subject of intervention enhancements for justice-involved youth who are using drugs.

Specifically, we present new evidence that mental health problems are associated with increased risk of condomless sex and alcohol and marijuana use even after accounting for youth’s underlying predisposition for risk-taking behavior. To our knowledge, this is the first study to show that mental health problems, particularly PTSD symptoms, are associated with condomless sex among youth involved with the criminal justice system. Past research focused on depression has been inconsistent (Lucenko et al., 2003; Elkington et al., 2008). Our study also replicates findings that mental health problems are associated with substance use in criminal justice (Abram, Teplin, McClelland, & Dulcan, 2003; Davis, Dumas, Wagner, & Merrin, 2016) and treatment settings (e.g., Rowe, Liddle, Greenbaum, & Henderson, 2004), and extends these findings to youth involved with juvenile drug court. Thus mental health problems may be an important barrier to the therapeutic aims of JDC throughout the court process and after, suggesting that targeting these factors may be a good use of resources within JDC programs. More research should examine the extent to which the effects of mental health symptoms on risky sex may be mediated by substance use.

Our study also replicates past research linking delinquent peer behavior to both substance use (Fergusson et al., 2002) and unprotected sex (Voisin et al., 2012). Peer delinquency has been shown to mediate treatment effects on other delinquent behaviors (Henggeler, Letourneau, Chapman, Borduin, Schewe, & McCart, 2009). While JDCs may order youth not to associate with certain peers, they have limited influence over peer behavior. And, in some cases, JDCs inadvertently encourage involvement with negative peers via group-based treatments and court-arranged activities. However, parents exert a powerful influence over whom their children associate with, and parental monitoring has been shown to protect against youth substance abuse (Steinberg et al., 1994) and mediate treatment effects on other delinquent behaviors (Henggeler, Brondino, & Pickrel, 2000; Eddy & Chamberlain, 2000). In fact, other research shows that, when drug court interventions are effective at reducing problem substance use, it is often because they increase parental monitoring behavior (Schaeffer, Henggeler, Chapman, Halliday-Boykins, Cunningham, Randall, & Shapiro, 2010). Thus, our findings build upon and extend the research base supporting that parenting behaviors that aim to reduce youth association with deviant peers may be an important target for effective therapeutic drug-court interventions.

By contrast, past research findings on whether peers engaged in prosocial activities protects against substance use and risky sex are contradictory, and our own findings are mixed. Our findings are similar to those reported by Prinstein and colleagues (2001) – we failed to find evidence that simply having prosocial peers is a significant protective factor for adolescents’ engagement in substance use. As suggested in the study by Coyle and colleagues (2016), perhaps the protective impact of prosocial peers emerges only when those peers actively discourage substance use. By contrast, we find that parent-reported prosocial peer behavior is associated with significantly reduced odds of reporting any sex, and reduced odds of condomless sex, although the latter effect did not achieve significance. Like the findings for substance use, perhaps this protective effect only emerges when prosocial peers actively communicate healthy sexual behavior norms. Consistent with that view, there is evidence that adolescents – both in juvenile detention specifically, and in the general population – are more likely to use a condom during sex if they perceive their peers are doing the same (Romer, Black, Ricardo, Feigelman, Kaljee, Galbraith, et al., 1994; DiClemente, 1991). However, it is also important to acknowledge that, with a sample of only around 50 youth who received peer behavior assessments at each visit, we had little statistical power to detect effects of prosocial peer behaviors.

This study has several limitations: (1) While 83% of participants completed the 12-month follow-up, only 56% completed the final 18-month follow-up. While the use of mixed effects models protects against bias in the case that observed covariates are differentially associated with dropping out of the study, if unobserved substance use or risky sexual behavior at these timepoints are also associated with study exit, then the associations presented here would be biased (Donders, van der Heijden, Stijnen, & Moons, 2006). This problem is ubiquitous in longitudinal studies, particularly with transient populations like families involved with JDC. Most youth who exited the study early were lost to follow-up, for example because they moved, or had their phone service terminated and could not be located via other means. Our sensitivity analysis excluding the 18-month visit had qualitatively similar results in all cases to the analysis including the 18-month visit. (2) This analysis also does not examine the independent contributions of each of the risk factors examined. For example, young people who have higher levels of mental health symptoms may also be more likely to associate with delinquent peers. Both of these were found to be risk factors for alcohol and marijuana use but the present analyses do not determine how strongly each of these risk factors is independently associated with alcohol or marijuana use after controlling for the other. (3) Because this was an exploratory analysis, hypothesis testing for statistical significance is not corrected for multiple comparisons, although we do adopt a conservative p<.01 threshold. However, the significant associations we found are all in the expected direction, and are mostly consistent across related measures, suggesting our findings are not spurious. (4) While this analysis looked at both substance use and risky sex outcomes, we did not collect data on whether youth were having sex while under the influence of marijuana or alcohol. (5) The small sample prevented the use of more flexible “random slopes” models. These models not only allow each youth to have his/her own model intercept; they also allow each youth to have his/her own time trend. This lack of flexibility may incrase the risk for residual confounding. (6) Finally, this was a relatively small sample of youth drawn from two southeastern JDCs, and results may not generalize to youth in other drug courts or in other juvenile justice settings throughout the United States.

Despite these limitations, the findings presented here are the first to exploit multiple measurements on youth over an extended time window to show that when youth in juvenile court are experiencing more mental health challenges and are spending time with deviant peers, they are also more likely to be engaging in substance use and risky sex, even after accounting for underlying predisposition for risky behavior. In fact, these findings are some of the first to track trends in and examine risk factors for the risky sexual behavior of youth in JDC at all.

These findings offer guidance to policy makers and court officials who seek to improve the quality of therapeutic services provided by the increasingly important JDC system. Past research has shown that most services provided to youth in JDC are not evidence-based (Henggeler & Marlowe, 2010; Latessa, et al., 2013). That research also shows that JDCs are most effective when evidence-based services are provided (Henggeler & Marlowe, 2010), and in fact may be harmful when services are not evidence-based (Latessa, et al., 2013). This paper extends this research by identifying two important factors – mental health and peer behaviors – associated with substance use and risky sex even after accounting for youths’ observed underlying propensity for risky behavior, suggesting that these factors could be important targets for treatment. And, in fact, there are some evidence-based treatments for adolescent substance abuse that target these issues simultaneously (Hogue, Henderson, Ozechowski, & Robbins, 2014). This also highlights the need for inter-agency and inter-sector collaboration between justice, substance use, and mental health professionals; each of these service systems can uniquely contribute to promoting healthy adolescent behavior. Future research should examine whether juvenile justice interventions that provide evidence-based mental health services, and that seek to incorporate families as allies in treatment to prevent engagement with delinquent peers and promote engagement with peers who convey healthy messages about substance use and sex, are more effective than usual services.

Appendix: Demographics of Analysis Samples

Table 3.

Demographics of Mental Health Symptoms and Substance Use Analysis

	Assessment 1	Assessment 2	Assessment 3	Assessment 4	Assessment 5	Assessment 6
Study Visit	103	93	90	78	79	53
Age	14.87 [1.41]	14.88 [1.4]	14.94 [1.33]	14.9 [1.4]	14.85 [1.42]	15.23 [1.23]
Male	86 (83%)	79 (85%)	76 (84%)	65 (83%)	67 (85%)	47 (89%)
Marijuana Use	89 (86%)	18 (19%)	7 (8%)	13 (17%)	19 (24%)	26 (49%)
Alcohol Use	40 (39%)	8 (9%)	5 (6%)	3 (4%)	6 (8%)	8 (15%)
Internalizing Symptoms (Self-Report)	0.36 [1.03]	0.07 [1]	0 [1]	−0.07 [1.05]	−0.31 [0.85]	−0.25 [0.86]
Depressive Symptoms (Self-Report)	0.25 [1.08]	0.07 [0.98]	0.1 [1.06]	0 [1.03]	−0.27 [0.82]	−0.37 [0.81]
Anxiety Symptoms (Self-Report)	0.35 [1.13]	0.07 [1.05]	−0.09 [0.93]	−0.09 [1.06]	−0.27 [0.77]	−0.13 [0.79]
Trauma Symptoms (Self-Report)	0.34 [1.1]	0.04 [1]	−0.03 [0.97]	−0.03 [1.03]	−0.29 [0.84]	−0.2 [0.84]

Note: Scale scores are log-transformed, centered at their mean, and scaled by their standard deviation.

Table 4.

Demographics of Peer Behaviors and Substance Use Analysis

	Assessment 1	Assessment 2	Assessment 3	Assessment 4	Assessment 5	Assessment 6
Population	51	56	53	48	52	35
Age	43 [0.84]	10 [0.18]	3 [0.06]	9 [0.19]	11 [0.21]	21 [0.6]
Male	19 (37%)	2 (4%)	1 (2%)	1 (2%)	4 (8%)	7 (20%)
Marijuana Use	14.37 (152%)	14.39 (149%)	14.51 (140%)	14.5 (149%)	14.44 (147%)	14.94 (130%)
Alcohol Use	43 (84%)	49 (88%)	46 (87%)	40 (83%)	43 (83%)	31 (89%)
Peer Delinquency (Self-Report)	0.37 [1.04]	−0.06 [0.99]	−0.14 [0.9]	−0.07 [0.94]	−0.24 [0.97]	0.26 [1.1]
Peer Delinquency (Parent Report)	0.45 [1.07]	−0.19 [0.85]	−0.04 [1.07]	−0.28 [0.81]	−0.11 [0.97]	0.29 [1.09]
Peer Conventional Activities (Self-Report)	0.01 [0.82]	0.09 [0.99]	−0.04 [0.97]	−0.05 [1.02]	0.03 [1.13]	−0.09 [1.07]
Peer Conventional Activities (Parent Report)	−0.16 [1.06]	0.12 [0.92]	−0.02 [1]	−0.08 [1.05]	0.14 [0.99]	−0.09 [0.98]

Note: Scale scores are log-transformed, centered at their mean, and scaled by their standard deviation.

Table 5.

Demographics of Mental Health Symptoms and Sexual Risk Behaviors Analysis

	Assessment 1	Assessment 2	Assessment 3	Assessment 4	Assessment 5	Assessment 6
Population	102	93	90	78	79	53
Age	14.86 [1.41]	14.88 [1.4]	14.94 [1.33]	14.9 [1.4]	14.85 [1.42]	15.23 [1.23]
Male	85 (83%)	79 (85%)	76 (84%)	65 (83%)	67 (85%)	47 (89%)
Any Sex	34 (33%)	33 (35%)	34 (38%)	30 (38%)	35 (44%)	27 (51%)
Condomless Sex	12 (12%)	10 (11%)	9 (10%)	12 (15%)	11 (14%)	12 (23%)
Internalizing Symptoms (Self-Report)	0.36 [1.04]	0.07 [1]	0 [1]	−0.07 [1.05]	−0.31 [0.85]	−0.25 [0.86]
Depressive Symptoms (Self-Report)	0.25 [1.08]	0.07 [0.98]	0.1 [1.06]	0 [1.03]	−0.27 [0.82]	−0.37 [0.81]
Anxiety Symptoms (Self-Report)	0.36 [1.13]	0.07 [1.05]	−0.09 [0.93]	−0.09 [1.06]	−0.27 [0.77]	−0.13 [0.79]
Trauma Symptoms (Self-Report)	0.35 [1.11]	0.04 [1]	−0.03 [0.97]	−0.03 [1.03]	−0.29 [0.84]	−0.2 [0.84]

Note: Scale scores are log-transformed, centered at their mean, and scaled by their standard deviation.

Table 6.

Demographics of Peer Behaviors and Sexual Risk Behaviors Analysis

	Assessment 1	Assessment 2	Assessment 3	Assessment 4	Assessment 5	Assessment 6
Population	51	56	53	48	52	35
Age	14.37 [1.52]	14.39 [1.49]	14.51 [1.4]	14.5 [1.49]	14.44 [1.47]	14.94 [1.3]
Male	43 (84%)	49 (88%)	46 (87%)	40 (83%)	43 (83%)	31 (89%)
Any Sex	14 (27%)	12 (21%)	11 (21%)	11 (23%)	18 (35%)	16 (46%)
Condomless Sex	4 (8%)	1 (2%)	3 (6%)	5 (10%)	6 (12%)	7 (20%)
Peer Delinquency (Self-Report)	0.37 [1.04]	−0.06 [0.99]	−0.14 [0.9]	−0.07 [0.94]	−0.24 [0.97]	0.26 [1.1]
Peer Delinquency (Parent Report)	0.45 [1.07]	−0.19 [0.85]	−0.04 [1.07]	−0.28 [0.81]	−0.11 [0.97]	0.29 [1.09]
Peer Conventional Activities (Self-Report)	0.01 [0.82]	0.09 [0.99]	−0.04 [0.97]	−0.05 [1.02]	0.03 [1.13]	−0.09 [1.07]
Peer Conventional Activities (Parent Report)	−0.16 [1.06]	0.12 [0.92]	−0.02 [1]	−0.08 [1.05]	0.14 [0.99]	−0.09 [0.98]

Note: Scale scores are log-transformed, centered at their mean, and scaled by their standard deviation.