Abstract
Introduction
Substance use and misuse among adolescents in U.S. high schools (typically, Grades 9-12) remains a major public health concern in the United States. Alcohol and tobacco are the most common substances used during the adolescent period in the United States, and they have been considered to be the gateway substance for many adolescents (Nkansah-Amankra & Minelli, 2016). The Monitoring the Future Study (Johnston et al., 2018) reveals that approximately 62% of American 12th-grade students report consuming alcohol, and 49% report experimenting with an illicit drug prior to reaching adulthood. Moreover, one in five American 12th-grade students use marijuana or synthetic marijuana, and 6% of these students report daily use for at least 1 month (Johnston et al., 2018). Indeed, since 2005, marijuana use has increased among American adolescents who concurrently use alcohol and tobacco products (Miech, Johnston, & O’Malley, 2017). In contrast, cigarette use is on the decline, with only 7.8% of American adolescents endorsing past or current cigarette smoking (Wang et al., 2018). Nonetheless, the prevalence of the use of tobacco products overall has remained constant with the rise in popularity of electronic cigarettes (e-cigarettes) and vaporing products, and nearly 15% of 12th graders in the United States endorse the use of traditional or nontraditional tobacco products (Barrington-Trimis et al., 2016).
Adolescents typically exercise poor judgment and impulsivity compared with their adult counterparts (Romer, 2010; Steinberg, 2008, 2010). A neural maturation gap in cognitive control over risk-taking behaviors renders them more likely to engage in risk-taking behaviors such as substance use, sexual experimentation, or driving under the influence. In addition, social influence such as peer pressure contributes to adolescent substance use (Mason et al., 2017; Romer & Hennessy, 2007).
Certain risks for adolescent drug experimentation have been correlated with the emotional state and well-being of adolescents; moreover, mental health diagnoses such as mood and/or anxiety disorder can function as an additional risk factor. According to the National Comorbidity Survey-Adolescent supplement (NCS-A), rates of substance use are three times higher among those with a co-occurring mental illness (e.g., mood and/or anxiety disorders) (Merikangas et al., 2010). Indeed, the onset of bipolar disorders is often preceded by substance use (Duffy et al., 2012); moreover, substance abuse signals an increased risk of suicidal behavior (Wong, Zhou, Goebert, & Hishinuma, 2013). Furthermore, Conway, Swendsen, Husky, He, and Merikangas (2016) report that adolescents with a history of mood, phobia, anxiety, conduct, and/or eating disorders report alcohol use (10%) and illicit drug use (15%). Certainly, the intersections of mood status and substance use are complicated, and mood symptoms are hypothesized to precede substance use/abuse and vice versa (Tolliver & Anton, 2015).
Identifying substance use patterns and their predictors is necessary for developing and executing prevention strategies to intervene in cases of adolescent substance use and experimentation. In addition, identifying these patterns and predictors is important because of (1) the newly emerging substance use patterns and behaviors among adolescents (e.g., using vaping products) and (2) the paucity of literature on how these behaviors and their combinations are presented. Against this backdrop, the study presented herein examined current substance use behaviors and patterns among U.S. high school students to examine whether or not individual factors (i.e., gender, grade in school, and depressive mood) can predict substance use patterns. The data source was the Centers for Disease Control and Prevention (CDC), 2015 National Youth Risk Behavior Survey (YRBS). The objective of the study was to better understand not only risk factors predisposing adolescents to substance use but also the patterns that they follow concerning the simultaneous experimentation of more than one type of drug. The results of this study (i.e., the risk factors of substance use and associated drug-use patterns) can inform primary prevention screening tools, psychoeducation, treatment modalities, and public policy to assist adolescents in the prevention of drug-use/abuse and reduce harm.
Method
This study involved a secondary data analysis of the CDC, YRBS, a national school-based survey of 9th- to 12th-grade students in private and public schools in the United States to monitor six categories of health risk behaviors (i.e., unintentional injuries and violence, sexual behaviors, alcohol and drug use, tobacco use, dietary behaviors, and physical activity), based on a stratified, multistage probability sample design. YRBS data are collected every 2 years, and the data set collected in 2015 was used in this study. Detailed information about this survey and its data are available online (i.e., https://www.cdc.gov/healthyyouth/data). Institutional review board approval requirements of the authors’ respective universities were waived because data from YRBS are publicly available, and the responses of its participants are deidentified.
Measures
Demographic Information
Self-reported demographic information comprised gender (i.e., female, male), race (i.e., White, others), and grade in school. Grade in school elicited a response from 9th grade to 12th grade. Some students responded with either no grade or another grade (i.e., unweighted; n = 35), and these participants were not included in the logistic regression analysis.
Substance Use
A total 13 substance usages were assessed: cigarettes, electronic vapor products, alcohol, marijuana, cocaine, inhalant, heroin, methamphetamines, ecstasy, synthetic marijuana, steroids, prescription drugs without a prescription, and injected illegal. Students were asked whether or not they either currently use cigarettes, electronic vapor products, alcohol, or marijuana or had used any of them at least once during the 30 days before the survey. Use of cigarettes included cigars and/or smokeless tobacco. Examples of electronic vaping included e-cigarettes, e-pipes, vaping pens, e-hookahs, and hookah pens or Starbuzz, based on 30-day use. Moreover, students were asked whether or not they have ever used cocaine, inhalant, heroin, methamphetamines, ecstasy, synthetic marijuana, steroids, prescription drugs without a prescription (e.g., opioid or stimulants), or injected illegal drugs one or more times during their life. If they used one of these substances at least once, their responses were recorded as 1 for yes and 2 for no.
Depressive Mood
Depressive mood was measured by the following question: Have you felt sad or hopeless for 2 weeks or longer, which has affected your activities during the past 12 months? Responses were recorded as 1 for yes and 2 for no.
Statistical Analysis
Latent class analysis (LCA) is a method to identify subgroups or classes of participants from set of information, and LCA was used to investigate the underlying substance use behavioral patterns that were measured by the 13 substance use indicators. Like factor analysis, LCA is one of most-used latent variable methods to identify the probability of a particular participant belonging to an unobserved latent class, by using observed survey responses. However, unlike factor analysis, LCA utilizes categorical variables, not continuous variables (Bartholomew, Steele, Moustaki, & Calbraith, 2008; Collins & Lanza, 2010). Therefore, we used potential combinations of binary responses from the 13 indicators, which comprised a 213 set from 15,624 respondents (i.e., 8,192 possibilities), and this complex data was allocated to latent classes of substance-use behavioral patterns.
LCA was performed using SAS 9.4 (SAS Institute Inc., Cary, NC) Specifically, the PROC LCA program, a SAS procedure of LCA, developed by the Penn State Methodology Center (Lanza, Dziak, Huang, Wagner, & Collins, 2015), was used for the latent class model to estimate parameters (i.e., latent class membership probabilities and item-response probabilities) (Lanza, Collins, Lemmon, & Schafer, 2007; Lanza et al., 2015). The PROC LCA procedure relies on the expectation–maximization (EM) algorithm and produces maximum likelihood (ML) estimates for parameters (i.e., probability of class membership and item-response probabilities). Missing data is assumed to be missing at random using the EM algorithm (Collins & Lanza, 2010; Lanza et al., 2007; Lanza et al., 2015). To find a final model solution, a one-class to seven-class model was fit to the data and compared to goodness-of-fit statistics. The G2, Akaike information criterion (AIC), and Bayesian information criterion (BIC) were used for model selection (Collins & Lanza, 2010; Lanza et al., 2007). To examine the predictors of class membership, multinomial logistic regressions were used. Covariates included gender (male as reference), race (White as reference), depressive mood, and grade in school (transformed though z score). Sampling weights and clusters were taken into account (Lanza et al., 2015).
Results
The YRBS for 2015 featured 15,624 student respondents, of which 51.3% were male, and 48.7% were female; 54.5% were White, and 13.6% were African American. Concerning grade level, 27.2% were in 9th grade, 25.7 were in 10th grade, 23.9% were in 11th grade, and 23.1% were in 12th grade. Germane to this study, 29.9% of the respondents reported experiencing sadness or hopelessness almost every day for 2 or more weeks. Detailed demographic information about the participants is publicly available (Kann et al., 2016).
LCA with one to seven-class model was performed. Table 1 displays the goodness-of-fit statistics. The G2 from the four-class to five-class model exhibited a large decrease. Although the AIC and BIC become smaller as the number of class increased, indicating better-fit models, the solution with more classes was difficult to interpret. The values of AIC and BIC tended to level off in the six- and seven-class models. Therefore, we selected the five-class model because it was parsimonious and exhibited better class separation and homogeneity compared to the solutions generated by the six- and seven-class model (Collins & Lanza, 2010; Lanza et al., 2007).
Criteria to Assess Model Fit for Substance-Use Behavior LCA Models.
Note. AIC = Akaike information criterion; BIC = Bayesian information criterion; df = degree of freedom.
Table 2 shows the five latent classes of substance use, the probabilities of membership (i.e., latent class prevalence) with assigned labels, and item-response probabilities for each item. Item response probabilities greater than or equal to 0.5 were considered to be high probabilities (Collins & Lanza, 2010). The largest class, abstinent, which consisted of approximately 64% of the respondents, is distinguished by low or zero probabilities of using assessed substances. In contrast, about 2% of respondents belonged to the smallest class, full experimenter. Adolescents in this class exhibited higher estimated probabilities (i.e., 67%-93%) of using all assessed substances. The class first-step social experimenter, which comprised 25% of the respondents, exhibited markedly high probabilities of using substances that included alcohol (76%), electronic vapor products (57%), and marijuana (51%). The class second-step social experimenter and the class pill experimenter comprised 6% and 4% of respondents, respectively. Respondents in the second-step social experimenter class exhibited high probabilities of using substances that included alcohol (96%), marijuana (87%), cigarettes (82%), electronic vapor products (81%), prescription pills (82%), and synthetic marijuana (60%). The class pill experimenter accounted for 2% of the sample and comprised adolescents who exhibited a high probability (57%) of using prescription drugs.
Five-Class Model of Substance Use Behaviors in 2015 (N = 15,624).
Table 3 presents latent class as a function of gender, race, grade in school, and depressive mood. The class abstinent was used a reference group and was compared with each of the other classes. There were statistically significant gender differences between abstinent and first-step social experimenter; abstinent and second-step social experimenter; and abstinent and full experimenter. Compared to the class abstinent, the odds of belonging to the class first-step social experimenter, second-step social experimenter, or full experimenter was significantly lower for female (odds ratio [OR] = 0.66; 95% confidence interval [CI] [0.52, 0.85]; OR = 0.47; 95% CI [0.37, 0.60]); OR = 0.28; 95% CI [0.18, 0.42]), respectively. No gender difference was found for the class pill experimenter compared to the class abstinent.
Depressive Mood as Predictor of Membership in Latent Class of Substance Use Behaviors (N = 15,589).
Note. OR = odds ratio; CI = confidence interval. Ref = reference latent class.
The analysis shows statistically significant race differences between not only abstinent and pill experimenter, but also abstinent and full experimenter. Compared with abstinent, the other race group increased the probability of membership in either pill experimenter (OR = 1.27; 95% CI [1.07,1.51]) or full experimenter (OR = 1.28; 95% CI [1.11, 1.48]), respectively.
Regarding grade in school (i.e., from 9th to 12th grade), compared to those in the abstinent class, adolescents in both the first-step social experimenter class (OR = 1.56; 95% CI [1.42, 1.71]) and the second-step social experimenter class (OR = 1.51; 95% CI [1.27, 1.81]) were more likely to be older.
Depressed mood increased the probability of class membership. For example, compared to adolescents in the class abstinent, adolescents in all other classes reported a more depressive mood. Indeed, all four covariates predicted class membership in a statistically significant fashion (i.e., ps < .001).
Discussion
Our results, generated through the use of latent class analysis, provide not only a behavioral profile of U.S. adolescents concerning substance use but also the specific predictors of using these substances. Five classes of substance use/abuse were identified (i.e., abstinent, first-step social experimenter, second-step social experimenter, pill experimenter, and full experimenter).
Adolescents in the class first-step social experimenter exhibited a low probability of using cigarettes yet a high probability of using alcohol, marijuana, and vapor products. Indeed, this class reflects the current trend of using drugs among adolescents in the United States: a decrease in the smoking of cigarettes (Jamal et al., 2017) and an increase in the use of vapor products (Arrazola et al., 2015; Havens, Young, & Havens, 2011; Jamal et al., 2017), while alcohol and marijuana are a primary and secondary substance of abuse among American youth aged 15 to 17 years (Substance Abuse and Mental Health Services Administration, Center for Behavioral Health Statistics and Quality, 2014).
Adolescents in the second-step social experimenter class used cigarettes, synthetic marijuana, and nonmedical prescription medication in addition to substances used by those in the first-step social experimenter class. Although our study did not examine the progression of first-step social experimenters to second-step social experimenters (because of the nature of cross-sectional analysis), participants in the first-step social experimenters are more likely to experiment with drugs. Previous studies have not only reported that vaping is a risk factor of cigarette smoking among nonsmoking adolescents (Barrington-Trimis et al., 2016; Miech, Patrick, O’Malley, & Johnston, 2017) but also suggest a relationship between the use of electronic vape products and subsequent cigarette smoking among adolescents (Watkins, Glantz, & Chaffee, 2018). Moreover, past research (Leventhal et al., 2016; McCarthy, 2014) has demonstrated that vaping may be the new introductory pathway to smoking tobacco among adolescents. Furthermore, marijuana and alcohol use are longitudinal predictors of using synthetic marijuana among high school students (Ninnemann, Jeong Choi, Stuart, & Temple, 2017). From these results, a possible progression from first-step social experimenter to second-step social experimenter certainly can be speculated; therefore, subsequent studies should examine the nature of such a relationship.
An interesting finding of our analysis is that nonmedical use of prescription medication (e.g., opioids and stimulants) among adolescents in the pill experimenter class exhibited a small estimated probability of using other substances—except for nonmedical prescription medication. Simultaneously, we observed high probabilities of using nonmedical prescription medication in the second-step social experimenter class, as was observed in a prior study (Moss, Chen, & Yi, 2014) that reveals that adolescent use stimulants along with other substances (e.g., alcohol, marijuana, and cigarettes). Indeed, adolescents exhibiting chronic medical or psychiatric illnesses may be predisposed to such use because access to prescription medications, such as opioids or stimulants, is readily available; therefore, misuse can easily ensue. Moreover, performance enhancement and recreational misuse account for nonmedical prescription medication use (McCabe, West, & Boyd, 2013; Wilens et al., 2008).
Of particular concern is the full experimenter class. Adolescents in this class exhibit a high probability of using all 13 substances queried in the YRBS compared to those in any other class. Although only 2% of the total participant sample belonged this class, this membership could pose serious problems for their health and life—adolescents of this type are likely to maintain their behaviors over time and increase potential risk related to multiple use of different substances (including illicit drugs) simultaneously (Merrin, Thompson, & Leadbeater, 2018).
Consistent with findings reported in the literature, the correlates of class membership revealed in our analysis comprised gender, race, grade in school, and depressive mood. In particular, being female revealed a protective factor for all classes compared to the class abstinent—with the exception of pill experimenter. As such, experimentation with more active forms of drug use (i.e., inhalants and drinking) appears to be less attractive to females, while pill use was comparable to risks seen among males. This finding should be taken into consideration when (1) addressing the severity of nonmedical pill use among young women and (2) discussing the risks of using these drugs compared to more active forms of drugs.
Race is associated with adolescent substance use behaviors. Prior studies (Wu, Swartz, Brady, Blazer, & Hoyle, 2014; Wu, Woody, Yang, Pan, & Blazer, 2011) have indicated that the use of nonmedical opioid analgesics have been prevalent in non-White adolescents (i.e., Native Americans and multiple-race individuals), while the use of nonmedical stimulants have been prevalent among White and Native American adolescents. Adolescents who are not White have demonstrated increased risk of using not only methamphetamines but also any illicit substances (McDermott et al., 2013). Although individuals should not be singled out based on race or ethnicity when screening adolescents, clinicians must remain cognizant that these individual risk factors may require additional screening attention and intervention support when discussions of substance use occur.
The pattern across grades in school (i.e., from 9th to 12th grade) suggests change over time, especially for the first-step social experimenter and second-step social experimenter classes, compared to the abstinent class. A possible explanation for this involves (1) the relatively easy access to the substances in questions, (2) the sharing of these substances among adolescents, and (3) the presence of peer pressure.
Depressed mood was related to membership in all classes. Conway et al. (2016) claim mental illness increased the risk of not only transition from nonuser to user but also developing substance-related problems among adolescents. Moreover, one of our findings (i.e., the increased odds of use from abstinent to full experimenter) suggests that depressed mood can exacerbate the severity of experimentation. These ascending risks magnify the urgency of screening for not only use but also depressive symptoms. The reason is that a given individual is at higher risk for extensive use, rather than casual experimentation, when depressive symptoms are present.
Although we observed individuals on the extreme ends of the spectrum from abstinent to full experimenter, prevention may be possible among adolescents who experiment with pills (beginning as a first-step social experimenter and/or second-step social experimenter); moreover, treatment is possible for those who have experimented with most drugs. Knowledge of these substance use patterns can help health care providers allocate time during an already brief visit with adolescents regarding drug and alcohol use. Because existing adolescent screening tools for smoking cigarettes simply may not be effective among adolescents (i.e., introductory use has changed with technology), identifying possible patterns that incorporate vaping is now necessary for not only the prevention of tobacco use but also concurrent alcohol and marijuana use.
Moreover, knowing that pill experimentation may present without additional indication or concurrent use must be understood. A main source of diversion of prescription medications is peers and/or family (Festinger et al., 2016; McCabe & West, 2014); therefore, health care providers should deliver education to adolescents and their parents regarding (1) the safe storage and disposal of medications and (2) the health consequences of any nonmedical use of prescription drugs. Identifying risky pill use and preventing an adolescent from progressing to another class of substance experimentation can—and should—be achieved by health care providers such as mental health nurses. Indeed, the development of a drug and alcohol screening tool for adolescents—or the revision of current screening tools, based on new latent patterns—could increase the effectiveness of prevention and treatment interventions initiated by mental health nurses. Continuing health education in schools regarding substance and mental illness can increase student awareness and, therefore, increase the access to relevant resources.
Moreover, we contend that drawing on a neurodevelopmental perspective, rather than merely relying on only student education (i.e., providing psychiatric interventions such as counseling or peer support for students instead of merely indoctrinating them to abstain from using substances), would yield better results in stemming substance abuse among adolescents through the development and implementation of policy and interventions targeting the access to said substances. The nature of the partnerships between K-12 institutions and psychiatric services necessary to leverage the neurodevelopmental perspective among the adolescents in question has been examined (i.e., exchanges between K-12 administrators and psychiatry teaching faculty resulted in more comprehensive intervention strategies for the adolescent students) (Brown & Astman, 2007), and mental health nurses should take advantage of this perspective in reducing substance abuse among adolescents. Furthermore, the U.S. Food and Drug Administration (US FDA) recently has announced a plan to limit access to flavored cigarettes (US FDA, 2018). As such, mental health nurses and other relevant stakeholders should observe how this plan will change the use of adolescent vapor products and subsequent traditional cigarette smoking.
Of course, the findings of this study are limited by certain constraints. First, the substance use behavior reflected in the YRBS was collected via self-report, which confounds results through problems of subjectivity and memory. Second, the cross-sectional study design of the YRBS places constraints on inferring the causal relationships between covariates and behavioral patterns. Furthermore, the time frame of using substances captured by the YRBS was inconsistent (i.e., current use and lifetime use may have been used interchangeably). Indeed, depressive mood was screened with only one item, which may not detect other behavioral changes in depressive mood, such as insomnia or social isolation. Not only individual characteristics but also environmental characteristics, such as family, peer, or school, would provide a more comprehensive picture of adolescent substance use behaviors. Nonetheless, our results, based on a secondary analysis of the YRBS data, patently provide needed insight into adolescent behavioral patterns and their profiles regarding substance use, as demonstrated in the literature (Evans-Polce, Lanza, & Maggs, 2016)—that is, college students exhibit various patterns of drug use (e.g., nonuser and polysubstance user) and the increased use of vapor products.
Conclusion
This study identified five latent patterns of substance use among U.S. adolescents. Based on our latent class analysis of data drawn from the CDC YRBS, gender, grade in school, and depressive mood were associated with membership in particular behavioral classes related to substance use/abuse. Clinicians, nurses, teachers, and parents can leverage our results in screening and identifying substance use and relevant factors among adolescents. Furthermore, following our results, nurses can help adolescents reduce substance use behavior through appropriate intervention strategies. For example, strategies in the clinical setting would be to either use the LCA grouping patterns that identify substances that adolescents typically use together or stage the patterns into different risk-level categories. These strategies could allow for more insightful engagement to not only elucidate the breadth of a given adolescent’s substance use but also determine the level of intervention needed for the adolescent (e.g., in-office motivational interviewing or referral for specialized substance use treatment). For subsequent study, we recommend a longitudinal study of not only substance use patterns but also the progression through substance use disorders among U.S. adolescents.
Footnotes
Author Roles
HL: Concept and design; analysis and interpretation; drafting the manuscript; final approval of the article; agrees to be accountable for all aspects of work ensuring integrity and accuracy. KY: Analysis and interpretation; revision of the article; final approval of the article; agree to be. JP: Concept; drafting the manuscript; final approval of the article; agrees to be accountable for all aspects of work ensuring integrity and accuracy. BK: Concept; drafting the manuscript; final approval of the article; agrees to be accountable for all aspects of work ensuring integrity and accuracy. LL: Concept; drafting the manuscript; final approval of the article; agrees to be accountable for all aspects of work ensuring integrity and accuracy. BG: Acquisition; critical revision of the article; final approval of the article; agrees to be accountable for all aspects of work ensuring integrity and accuracy.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
