Using polytomous latent class analysis to compare patterns of substance use and co-occurring health-risk behaviors between students in alternative and mainstream high schools

Abstract

Introduction:

Alternative high school (AHS) students, an understudied and underserved population, experience educational, social, and health disparities relative to students in mainstream high schools. Disparities in single types of substance use are particularly high, yet no known studies have compared patterns of substance use or relationships between these patterns and other health-risk behaviors between AHS and mainstream high schools.

Methods:

Using data from the Texas Alternative School Health Survey (n =515; mean age 17.1 years, 49% male, 59% Hispanic, 23% White, 15% Black) and the Texas Youth Risk Behavior Survey (n =2,113; mean age 16 years, 47% male, 64% Hispanic, 22% White, 7% Black), we used latent class analyses to compare patterns of substance use in AHSs and mainstream high schools. We used latent class regression to examine relationships between patterns of substance use and involvement in other health-risk behaviors in each school setting.

Results:

Students in AHSs and mainstream high schools had similar patterns of substance use, and youth in higher risk categories engaged in higher levels of other health-risk behaviors. A substantially greater proportion of AHS students, however, fell into the moderate and high use categories, in support of continuing disparities for AHS students.

Conclusions:

Additional support is needed in AHSs to address the prevalence of high-risk patterns of substance use and associated health-risk behaviors. For example, ongoing public health surveillance is needed in AHSs, just as is done in mainstream high schools, to monitor trends in substance use and impact of policies and interventions.

Substance use encompasses a wide range of behaviors that independently and collectively pose a threat to adolescent health and undermine lifelong health and well-being (National Center on Addiction and Substance Abuse, 2011). Ample evidence demonstrates that in adolescence, substance use and other health-risk behaviors often cluster together (Bernat & Resnick, 2006; Institute of Medicine and the National Research Council [IOM], 2011); specifically, adolescent substance use has been linked to risky sexual behaviors, violence, delinquency, school dropout, and mental health problems (IOM, 2011, p. 91; Jackson, Seth, DiClemente, & Lin, 2015). In the U.S., the prevalence of adolescent substance use and other health-risk behaviors is therefore closely monitored by several surveys, such as the Youth Risk Behavior Survey (YRBS) and Monitoring the Future. Yet the school-based surveys on which we rely for nationally representative prevalence estimates (e.g., the YRBS) to guide our decisions in program and policy development draw samples from mainstream high school settings only.

Alternative high schools (AHSs) serve a growing but understudied population of youth who are at risk for school dropout. Yet AHS students are rarely included in local, state, or national prevalence estimates of substance use and other health-risk behaviors (Brener et al., 2013; Demissie et al., 2013). The last—and only—time that nationally representative prevalence estimates of health-risk behaviors among AHS students were obtained was in 1998 when the Centers for Disease Control and Prevention (CDC) conducted the YRBS in AHSs. The results suggested that AHS students were significantly more likely than students in mainstream schools to engage in substance use and associated risk behaviors. For example, AHS students were significantly more likely than students in mainstream high schools to report using marijuana in the past month (54% vs. 26%; Grunbaum, Lowry, & Kann, 2001).

This exclusion of AHS students is problematic for several reasons. The AHS population has unique characteristics in comparison with the population of youth enrolled in mainstream high schools that may result in different patterns of how substance use emerges during adolescence, differences in relationships with co-occurring risk behaviors, and differences in the legal implications of illicit drug use. As a group, AHS students experience higher levels of social risk factors for substance use and co-occurring risk behaviors, such as adverse childhood experiences (e.g., parental incarceration, abuse, unstable home environments; Carver, Lewis, & Tice, 2010; Johnson, Morris, Rew, & Simonton, 2016). Compared with students in mainstream schools, AHS students have been found more likely to report having four or more sexual partners during their lifetime (48% vs. 16%), attempting suicide in the 12 months preceding the survey (18% vs. 8%), and being in a physical fight in the last 12 months (62% vs. 37%; Grunbaum et al., 2001). Additionally, AHS students are disproportionately low income, Hispanic, and Black (Johnson et al., 2017; Johnson, McMorris, & Kubik, 2013). Institutional factors have contributed to these racial and socioeconomic disparities, including unequal enforcement of disciplinary policies (e.g., zero tolerance policies) that lead to suspension or expulsion of students of color at higher rates, and high stakes testing that drives underperforming students—often in underperforming schools—out of mainstream high schools and into AHSs (Heitzeg, 2009). Given these forms of institutional racism that already influence AHS students, in addition to the fact that people of color, as a group, face harsher punishment than white people for the same drug offenses (Bunting, Garcia, & Edwards, 2013; Golub, Johnson, & Dunlap, 2007), AHS students who use illicit drugs are at particularly dire risk for unequal treatment in the justice system, which will further disrupt their development and perpetuate health and social disparities.

Understanding how substance use and co-occurring risk behaviors operate in this unique population is therefore crucial to informing primary, secondary, and tertiary prevention efforts to reduce health and social disparities. There may be important differences between AHS and mainstream high school settings that could result in different patterns of substance use and co-occurring risk behaviors among adolescents in those settings. For example, the AHS setting is typically characterized by smaller classes and more individualized learning environments that may result in teachers and peers’ being more aware of which students within the school are using substances (Lehr, Tan, & Ysseldyke, 2009). In addition, AHSs have historically been stigmatized as a place for “bad” or “broken” youth (Becker, 2010; McNulty & Roseboro, 2009), resulting in students’ receiving less attention, concern, and resources from school districts and society at large. Without adequate resources, administrators and teachers within the AHS setting may not have the capacity to address upstream and downstream causes of substance use, such as negative peer influence, stressors or adverse childhood experiences faced at home and in the community, and systematic inequalities that place AHS students at greater disadvantage.

The focus of health-related research conducted in AHSs has generally been narrow. For example, examining single substances in isolation can mask the heterogeneity of substance use behaviors (Salas-Wright, Vaughn, & Ugalde, 2016)—specifically patterns of polysubstance use (Parker & Bradshaw, 2015; Tomczyk, Hanewinkel, & Isensee, 2015). Most research on AHSs to date has done just that, so we know little about how polysubstance use operates among AHS students as opposed to mainstream high school students. Latent class analysis (LCA) is a useful analytic tool to help identify different profiles of substance use among adolescents, one that has been sporadically employed in studies of adolescent substance use but not in looking at the AHS subset of the adolescent population (Parker & Bradshaw, 2015; Tomczyk et al., 2015).

Among studies employing analytic techniques such as LCA to identify patterns of substance use among young people in general, researchers have commonly identified three to five latent classes of substance users. Youth categorized in these latent classes generally range from those who primarily abstain from using any substances to those who use a wide variety of substances in moderate or frequent intervals. Often LCA models suggest that there is a group of users who primarily use alcohol, at least one group that combines two or more “soft drugs” considered to be more socially acceptable (i.e., alcohol, marijuana, tobacco) in varying frequencies (Kulis, Jager, Ayers, Lateef, & Kiehne, 2016; Morean et al., 2016; Parker & Bradshaw, 2015; Riehman, Stephens, & Schurig, 2009; Salas-Wright et al., 2016), and a polysubstance use group that uses “soft drugs” along with additional illegal drugs (e.g., inhalants, prescription drugs, cocaine; Bohnert et al., 2014). Findings from LCAs conducted with non-AHS samples indicate that different patterns of substance use are related to various patterns of other risky behaviors. Generally, findings suggest that the latent classes categorized by riskier patterns of use, including more frequent use and/or use of multiple substances, have higher risk for other adverse outcomes, such as higher levels of psychological distress or mood disorders (Bohnert et al., 2014; Riehman et al., 2009), risky sexual behaviors such as multiple sex partners (Bohnert et al., 2014; Connell, Gilreath, & Hansen, 2009), dating violence (Parker & Bradshaw, 2015), and poorer grades and academic outcomes (Bohnert et al., 2014).

To summarize, gaps in the literature concerning substance use among students in AHSs include the lack of (1) current public health surveillance data, (2) analyses of patterns of substance use and associated health-risk behaviors, and (3) comparisons between AHSs and mainstream schools. The purpose of this study was to use YRBS data to identify the most common configurations of substance use behaviors among students in AHSs serving students at risk for school failure in Texas, to explore how those configurations co-occur with other health risks (e.g., mental health, sexual risk-taking), and to compare these findings with results from mainstream high schools in Texas.

1. Methods

1.1. Sample and procedures

We used data from two sources for this cross-sectional secondary analysis. The primary source was the 2015 Texas Alternative School Health (TXASH) Survey, in which an expanded YRBS was administered to students in Central Texas AHSs. For comparison, the secondary source was the 2017 YRBS conducted in traditional high schools in Texas. We used the 2017 dataset because the 2015 statewide YRBS data was unavailable, and the 2013 YRBS variables did not align with those in our AHS sample.

2015 Texas Alternative School Health (TXASH) Student Survey.

The TXASH Survey was administered in the spring of 2015 to a convenience sample of students enrolled in one of two types of AHSs in Central Texas: (1) alternative schools of choice serving at-risk students who have enrolled voluntarily, or (2) disciplinary alternative education programs (DAEPs) where students have been placed mandatorily. The survey consisted of 129 questions, including all items from the core YRBS and additional questions about risk and protective factors derived from other public use surveys (e.g., the Minnesota Student Survey). At the school level, we identified 38 AHSs across 26 districts that fit one of these two classifications. Of these, 14 schools participated in the TXASH Student Survey. At the student level, we used active parental consent and youth assent for students under the age of 18 years, and active consent from students who were older than 18. Students were informed that participation was voluntary and anonymous, and they received a $15 incentive for participating. We distributed 1,261 consent forms, and 515 students completed the survey for a co-operation rate of 41% (American Association for Public Opinion Research, 2016). A detailed discussion of our methods and co-operation rate has been published elsewhere (Johnson et al., 2017). This study was approved by the Institutional Review Board at our university.

2017 Texas Youth Risk Behavior Survey (YRBS).

The Texas YRBS has been administered by the Texas Department of State Health Services (DSHS) every other year since 1991 to students in grades 9–12 in traditional public high schools across the state. Schools were selected for the survey using probability proportional to size; then classrooms within the schools were randomly selected. All students within the selected classrooms were eligible to participate. To be representative of the population and released for public use, the survey requires an overall participation rate of 60%, which is calculated by multiplying the school-level participation rate by the student-level participation rate. Texas did not obtain an overall participation rate of 60% on the 2015 YRBS, so data from that year were unavailable. The school-level participation rate for 2017, however, was 70% and the student-level participation rate was 88%, for an overall participation rate of 62%. DSHS used a mixture of active and passive parental permission procedures to administer the survey, allowing districts to decide which type of consent procedure to use. The majority of districts used passive parental permission. Students were informed that the survey was voluntary and anonymous.

The 2017 Texas YRBS consisted of 99 items asking about involvement in six types of health-risk behaviors that contribute to the leading causes of morbidity and mortality among young people: (1) tobacco use (2) alcohol and other drug use, (3) sexual behaviors, (4) unintentional injury and violence, (5) dietary behaviors, and (6) physical activity (CDC, 2017).

1.2. Measures

The survey items we used in our analysis were identical between the TXASH Student Survey and the Texas YRBS as described below.

Substance use profiles.

We estimated latent class membership probabilities for each student, based on self-reported frequencies of 13 types of substance use including more mainstream or “soft drugs” including alcohol, tobacco, and marijuana and less mainstream but highly addictive “hard drugs” including cocaine, inhalants, heroin, methamphetamine, ecstasy, synthetic marijuana, and prescription drugs. Depending on the item, respondents were asked to indicate how many days or times they had used the substance in the past 30 days or during their lifetime. We collapsed response options into four categories, ranging from never to heavy. See Table 1.

Table 1.

Definitions of frequency categories for each type of substance use in the survey.

Unit of Measurement	Period	Substances	Never	Low	Moderate	Heavy
Days used	Past 30 days	Tobacco (Smoke, Chew, Cigar, Vape) and Alcohol (Drink)	0	1–5	6–19	20+
Times used	Past 30 days	Marijuana	0	1–2	3–19	20+
Times used	Lifetime	Cocaine, Inhalant, Heroin, Methamphetimines, Ecstasy, Synthetic Marijuana, Prescription drugs	0	1–2	3–19	20+

Associated health-risk behaviors and mental health indicators.

We estimated the prevalence of five different health-risk behaviors and mental health indicators in each latent class of substance users: sadness/hopelessness, suicidal ideation, attempted suicide, bullying victimization, fighting, and sexual activity. Each variable is described in Table 2.

Table 2.

Potential correlates of substance use latent classes.

	Time Frame	Variable Type	Description
Sadnessa	12 months	binary	Experienced severe sadness ≥2 weeks
Suicidal ideationb	12 months	binary	Strongly considered suicide
Attempted suicidec	12 months	binary	Attempted suicide
Fightingd	12 months	numeric	# of physical fights
Bullying victimizatione,f	12 months	binary	Bullied in school and/or online
Sexual activityg	3 months	binary	Had sex with ≥1 partner

^aDuring the past 12 months, did you ever feel so sad or hopeless almost every day for two weeks or more in a row that you stopped doing some usual activities? (answer options: yes/no).

^bDuring the past 12 months, did you ever seriously consider attempting suicide? (answer options: yes/no).

^cDuring the past 12 months, how many times did you actually attempt suicide? (categorical answer options ranging from 0 times to 6 or more times).

^dDuring the past 12 months, how many times were you in a physical fight? (categorical answer options ranging from 0 times to 12 or more times).

^eDuring the past 12 months, have you ever been bullied on school property? (answer options: yes/no).

^fDuring the past 12 months, have you ever been electronically bullied? (answer options: yes/no).

^gDuring the past 3 months, with how many people did you have sexual intercourse (categorical answer options range from 1 person to 6 or more people and options for never had sex and had sex but not in last 3 months).

1.3. Analysis

Polytomous Latent Class Analysis.

Polytomous latent class analysis (PLCA; McCutcheon, 1987; Agresti, 2002) was used to identify patterns of substance use. PLCA classifies subjects into $K$ latent classes based on their responses to survey items. In contrast to the typical binary latent class model, PLCA can accommodate categorical survey items and thereby account for varying intensities of substance use. Its central assumption, called “local independence,” is that after accounting for class membership, each student’s responses will be mutually independent. Formally, say there are $I$ polytomous survey items, and each item $i=1\dots I$ admits $q_i,2\le q_i$ categories. In a PLCA model with $K$ classes, the probability that an individual student’s response vector $\text{Y}$ will take a particular set of values$\text{y}$, say $y_1=3$, $y_2=1$, $y_3=5,\dots$, $y_I=2$ is written as:

$$Pr(Y=y)=\sum _{k=1}^{K}Pr(class=k)\prod _{i=1}^{I}Pr\left(Y_i=y_i\mid class=k\right)$$

where $Pr\left(Y_i=y_i\text{|}class=k\right)$ is the probability of response $y_i$ on survey item $i$, for students in class $k$. These response probabilities characterize the $K$ latent classes.

The “prior probabilities” $Pr(\text{ class }=k)$, the proportion of students in class $k$, and the response probabilities $Pr\left(Y_i=y_i\mid class=k\right)$ are estimated from the data. The fitted PLCA model also estimates “posterior” class membership probabilities, using Bayes’ Theorem. Let $P_k=Pr(class =k\mid Y=y)$, the probability that a student is in class $k$, given his or her survey responses $\text{y}$. Then,

$$P_k=\frac{Pr(Y=y\mid class=k)Pr(class=k)}{Pr(Y=y)}$$

where $Pr(Y=y\mid class =k)=\prod _{i}Pr\left(Y_i=y_i\mid class=k\right)$, as in equation (1), and the denominator is the entire sum in equation (1). By substituting student $j$’s responses $y_j$ for the generic $\text{y}$ in equation (2), we can estimate the posterior probability student $j$ is in each class $k$, given his or her survey responses, $P_{jk}$, with the $K$ -vector of class posterior probabilities denoted as $P_j$. Thus, we may classify students probabilistically, given their survey responses. This equation can also be used to estimate class membership probabilities for students who were not in the sample within which the PLCA model was fit, so long as their survey responses $\text{y}$ are available.

The number of classes $K$ must be chosen in advance. The LCA literature proposes a number of model statistics to choose $K$ based on the data; we relied on the BIC, CAIC and aBIC (e.g. Finch, 2015), which attempt to balance goodness of fit (measured by the log likelihood) against model complexity, and also report the AIC (Akaike, 1973).

We fit PLCA models in R statistical environment, using the poLCA package [R Core Team 2018; Linzer & Lewis, 2011]. We assumed that missing student responses were missing at random given latent class membership. Under this assumption, we fit PLCA models using full information maximum likelihood–i.e., by maximizing likelihood equations using whatever data were available for each subject, rather than deleting cases with missing data. The process is described in [Linzer & Lewis, 2011].

1.4. Correlates of substance use

We estimated the prevalence or extent of psychosocial indicators in each latent substance use class, in alternative and mainstream high schools. To do so, we followed the “Pseudo-class” approach of Bandeen-Roche et al. (1997); also see Lanza, et al., 2013), which multiply-imputes students’ unknown latent class memberships based on a fitted PLCA model. We classified students in both samples and across analyses using the initial PLCA model fit to the AHS sample, in order to facilitate comparisons between school types within a particular latent class. That is, we calculated probabilities of mainstream high school students’ class membership using the response probabilities $Pr(\text{Y}=\text{y}\mid class=k)$ estimated from the AHS sample, and “prior” probabilities $Pr(class=k)$ from the model fit to the mainstream high school sample. We adjusted standard errors for the clustering of students within high schools using the small-sample-corrected cluster-robust method of Pustejovsky & Tipton (2018).

In a parallel analysis, we regressed each psychosocial indicator S on class membership, school type, and their interaction, along with control variables age, gender, and race/ethnicity (i.e., Black/African American Hispanic/Latinx). These regressions measure differences in psychosocial indicators between classes and school-types, after accounting for student race, gender, and age. For our measurement of fighting, which was numeric, we used ordinary least squares linear regression; for the other indicators, which were binary, we used logistic regression. In either case, we estimated regression parameters using the pseudo-class approach, imputing mainstream students’ class memberships using the AHS model, and estimating standard errors with the Pustejovsky & Tipton (2018) cluster-robust method.

We assumed missing psychosocial indicator data were missing at random and used the mice package in R to multiply-impute their values conditional on estimated class membership probabilities, age, race and gender. We created 50 imputed datasets and analyzed them in parallel with pseudo-class draws described below.

Replication code for all analyses is available in an online appendix.

2. Results

A total of 515 AHS students took the TXASH Student Survey (mean age: 17.13 years; 48.7% male; 59.1% Hispanic, 23% White, 15% Black, 2.6% other; 63.8% free/reduced lunch). A total of 2,113 students in mainstream high schools took the 2017 Texas YRBS (mean age: 16 years; 46.7% male; 64.1% Hispanic, 22.1% White, 7.1% Black, 6.7% other).

2.1. Substance use in Central Texas alternative high schools

Choosing the Number of Classes $K$.

Table 3 gives a number of model fit criteria sometimes used to select $K$, the number of classes in a PLCA. We chose the three-class model, which minimizes the BIC, aBIC, and CAIC. On the other hand, the AIC recommends a model with 7 classes. In the three-class model, 81% of students were assigned to a class with probability of at least 0.9. (Our analyses used the entire sample, and probabilistic assignment.)

Table 3.

Model selection statistics for PLCA models, fit to the AHS data, with 1–8 latent classes. For each of the information criteria, the lowest (preferred) values is in bold.

	1	2	3	4	5	6	7	8
BIC	10,846	9,891	9,804	9,935	10,063	10,214	10,380	10,555
aBIC	10,722	9,640	9,427	9,430	9,432	9,455	9,494	9,543
AIC	10,681	9,556	9,299	9,260	9,219	9,200	9,196	9,201
CAIC	10,885	9,970	9,923	10,094	10,262	10,453	10,659	10,874

Characterizing Three Groups of AHS Students.

Fig. 1 characterizes the three latent classes among AHS students and provides estimated proportions of the sample belonging to each group. Each vertical bar in the figure represents the estimated probability of a student in a given latent class giving a particular response to a survey question about frequency of a particular type of substance use. For instance, the vertical bar on the top left of the figure represents the probability that a student in the top group (labeled “Low Users”) responded that they never smoke, which is approximately 93%. The bar on the bottom left reports the proportion of students in the “Heavy Users” group who reported never smoking as 27%. Table 4 gives the estimated percentage of each group reporting any use on each of the substance use questions. In all three groups, marijuana was easily the most used substance.

Fig. 1.

The probability of an alternative high school student in each latent class reporting each possible level of frequency for each type of substance use.

Table 4.

Percentage of alternative high school (AHS) and mainstream high school students in each latent class reporting any use, for each type of substance abuse.

	Low Users (%)		Moderate Users (%)		Heavy Users (%)
	AHS	Mainstream	AHS	Mainstream	AHS	Mainstream
Smoke tobaccoa	7	0	58	29	73	68
Chewing tobaccoa	3	0	6	15	47	49
Cigara	1	0	57	25	59	65
Vapea	21	1	67	33	70	58
Alcohola	19	15	68	79	86	95
Marijuanaa	26	6	76	58	100	72
Cocaineb	1	1	25	15	92	82
Inhalantb	2	3	10	14	69	63
Heroinb	0	0	2	3	55	54
Methamphetamineb	0	0	9	4	58	72
Ecstasyb	1	0	22	13	76	84
Synthetic Marijuanab	7	1	46	23	77	83
Prescription drugsb	12	6	60	40	94	76

^aPast month.

^bLifetime.

Approximately 47% of students in the AHS sample were Low Users. On each substance use question, the vast majority of Low Users responded ‘never.’ Specifically, over 90% of the students in this group reported “never” regarding use of tobacco use, cocaine, inhalants, heroin, methamphetamine, and synthetic marijuana; 88% reported never using prescription drugs. However, a fifth to a quarter of Low Users reporting some vaping, drinking, and marijuana use. The substance use that did take place in this group was typically infrequent. The estimated probability of a Low User reporting frequent use on any of the substances was below 3%, with the exception of marijuana use (10%).

Moderate users–roughly 41% of the sample–reported some types of substance use, but not others. In general, this group is characterized by occasional use of “light” drugs, and avoidance or infrequent experimentation with “heavy” drugs. Most students in this group reported at least some smoking of cigarettes and of cigars, vaping, drinking, and use of marijuana and prescription drugs; slightly less than half reported using synthetic marijuana. Roughly a quarter of moderate users reported some use of cocaine and ecstasy, yet few reported chewing tobacco, or using inhalants, heroin, or methamphetamines. The frequency of substance use in this group was also—for the most part—moderate. Seventy percent of moderate users who reported some drinking over the past month, and roughly half of those who reported some tobacco use, reported doing so on five or fewer days in the past month, consistent with weekend use. On the other hand, a plurality of moderate users who reported some marijuana use used it over 20 times in the past month. Otherwise, relatively small proportions of moderate users reported high frequency use of any of the substances. Roughly 15% reported smoking cigarettes, cigars, or electronic cigarettes all or most days in the past month or using prescription drugs 20 or more times in their lives.

Heavy using students–roughly 13%–exhibited all measured types of substance use at high levels; nearly every type of substance use was reported by a majority. Notably, the model estimates that 100% of heavy users used marijuana in the past 30 days, with 72% using marijuana at least 20 times in that period. Roughly 86% drank in the past 30 days, 50% of whom did so more than just on the weekend. Almost half reported abusing prescription drugs 20 or more times in their lives, and 74% used cocaine multiple times in their lives. Although a greater number of heavy users than moderate users smoked during the past 30 days, the proportions smoking 20 days or more were roughly equal between the two groups (16% in the moderate group and 17% in the heavy group).

2.2. Comparison: substance use in mainstream high schools

By way of comparison, we fit the same model to analogous data from mainstream high school students who took the 2017 YRBS in Texas. Because our goal is comparison between alternative and mainstream high schools, versus studying mainstream high schools themselves, we fit the model with the same specifications (i.e., a three-class model). Using the entire sample and probabilistic assignment, the PLCA model fit assigned 89% of students in the mainstream sample to a class with probability of at least 0.9.

The results, in the form of response probabilities by latent class, are displayed in Fig. 2. Overall, the pattern of substance use in mainstream high schools is quite similar to the pattern in alternative high schools. Specifically, the three groups can be roughly characterized as low-, moderate-, and heavy-users. Low-users barely report any substance use at all, while heavy users regularly use or experiment with a wide variety of substances. Moderate users tend to use tobacco, marijuana, or alcohol, and may have experimented with some other substances. The biggest difference between mainstream and alternative high schools is the distribution of the three classes: Whereas roughly 47% of AHS students were low-users, the low-use group comprises 79% of the mainstream high school students. On the other end, 13% of AHS students are heavy-users, but only 4% of mainstream high school students are heavy-users.

Fig. 2.

The probability of a mainstream high school student in each latent class reporting each possible level of frequency for each type of substance use.

Though patterns of substance use within the three latent classes are similar in the two types of schools, they are not identical. That is, the latent classes are defined differently in the two school types. For instance, AHS moderate users are about 30 percentage points more likely to report some use of each type of tobacco (cigarettes, e-cigarettes, cigars) than moderate users at mainstream high schools (p < 0.001).¹

Comparing class definitions between AHS and mainstream high school samples.

To compare the class definitions more broadly, we may compare students’ classifications under the PLCA model fit to data from their school type, to what their classifications would have been according to the model fit to the other type (see Tables 5 and 6). The results show that 78% of AHS students and 90% of mainstream students would have been classified in the same latent class in both fitted models. Of those AHS students whose classifications disagree, 98% would have been classified in the next higher class according to the mainstream classification. On the other hand, 87% of the mainstream students whose classifications disagree would have been classified in the next lower class according to the alternative classification. In other words, the two classifications largely agree; in cases where they disagree, the AHS latent classes tend to encompass more substance use than their mainstream counterparts.

Table 5.

Re-classification of each alternative high school (AHS)latent class, according to the PLCA model fit to mainstream schools.

AHS	Mainstream Classification
Classification	Low Users (%)	Moderate Users (%)	Heavy Users (%)
Low Users (%)	65	35	0
Moderate Users (%)	0	87	13
Heavy Users (%)	0	3	97

Table 6.

Re-classification of each mainstream latent class, according to the PLCA model fit to alternative high schools (AHS).

Mainstream	AHS Classification
Classification	Low Users (%)	Moderate Users (%)	Heavy Users (%)
Low Users (%)	99	1	0
Moderate Users (%)	45	54	1
Heavy Users (%)	0	28	72

2.3. Correlates between health-risk behaviors/mental health and latent class membership

We estimated the extent to which students’ substance use class membership predicted their responses to other health-risk behaviors and mental health indicators. As previously stated, students in both types of schools are classified as they would have been under the model fit to the AHS students to facilitate comparisons between students in AHSs and mainstream high schools.

The results are displayed in Table 7 and Fig. 3. The relationships between substance use class membership and involvement in the other health-risk behaviors followed roughly the same pattern in both the AHS sample and the mainstream high school sample. In both samples, suicidal ideation and attempts, fighting, and bullying all increase in severity with increasing levels of substance use. In both school types, moderate and heavy users were similarly likely to have experienced sadness, with both classes being more likely than low users to have experienced sadness. Across the board, AHS students were more likely to be sexually active than mainstream students in the same substance use class. The difference was most pronounced among low users: half of the low-users in the AHS sample were sexually active, compared with only 18% in mainstream high schools. In the AHS sample moderate users were more likely to be sexually active than heavy users, while in the mainstream sample moderate and heavy users were sexually active at approximately the same rates (see Table 7).

Table 7.

Percentages or means of psychosocial indicators by latent class in alternative high schools (AHS) and mainstream schools, according to the PLCA model fit in AHS.

		Low Users	Moderate Users	Heavy Users
Sadness (%, SE)	AHS	31.9 (4.8)	46.8 (3.1)	49.9 (5.5)
	Mainstream	32.6 (1.3)	53.5 (1.9)	51.7 (9.2)
Suicidal ideation (%, SE)	AHS	16.8 (4.2)	29.9 (4.9)	39.2 (6.1)
	Mainstream	15.3 (0.4)	33.6 (1.9)	46.8 (11.4)
Attempted Suicide (%, SE)	AHS	9 (2)	16.8 (3.5)	38.8 (6.7)
	Mainstream	8.5 (0.4)	25.3 (1.8)	56.7 (12.6)
Fighting (mean, SE)	AHS	0.8 (0.1)	1.9 (0.4)	5 (0.7)
	Mainstream	0.4 (0.1)	1.6 (0.3)	3.8 (0.3)
Bullying victimization (%, SE)	AHS	16.1 (3.5)	29 (4.8)	45 (5.8)
	Mainstream	23 (0.8)	30.1 (4.6)	38.1 (5.1)
Sexually Active (%, SE)	AHS	50.4 (1.7)	67.2 (3.2)	46.2 (6.7)
	Mainstream	18.5 (3.2)	42.2 (1.7)	42.9 (3.5)

Note. AHS =alternative high school. Standard errors (SE) in parentheses. Note that for suicidal ideation, suicide attempt, bullying victimization, and sexual activity, which are dichotomous, averages are proportions.

Fig. 3.

Estimate of how substance use class membership predicted student responses to other health-risk behaviors and mental health indicators, by school type (alternative high school [AHS] vs. mainstream high school). The stars above the bars give statistical significance for differences between students in the same school type but in different latent classes (horizontal comparisons on the graph). Specifically, the stars above the bar for low users indicates significance of the difference between low and medium users, the stars on top of the bar for moderate users gives significance for the difference between moderate and heavy users, and the stars on top of the bar for heavy users gives the significance of the difference between heavy and low users in the same type of school. The stars below the AHS bars refer to differences between students at different types of schools, within the same latent substance use classes (vertical comparisons within the graph).

In a follow-up analysis, we used regression to account for the possibility that some differences between substance use classes and between school types are probably driven by differences in demographic variables (see Table 8). The results, available in an online appendix, largely replicate the patterns in Fig. 3 and Table 7. Within the AHS sample, all variables other than sexual activity increased in severity with increasing substance use. As in the previous analysis, without adjusting for demographic covariates, sexual activity was higher among AHS moderate users than low users, and roughly similar (perhaps a bit lower) among heavy users than low users. Sexual activity was also the variable in which the differences between AHS and mainstream high school students was most pronounced: a significantly lower percentage of mainstream low users was sexually active than AHS low users, and a significantly higher percentage of mainstream heavy users was sexually active than AHS heavy users.

Table 8.

Coefficients and standard errors for regressions predicting psychosocial indicators as a function of latent class membership.

	Sadness	Suicidal ideation	Attempted suicide	Fighting	Bullying victimization	Sexual activity
Model Type	Logit	Logit	Logit	OLS	Logit	Logit
(Intercept)	0.38	0.43	−0.91	2.48***	0.76.	−7.86***
	(0.7)	(0.66)	(1.06)	(0.58)	(0.4)	(0.31)
Moderate Users	0.7*	0.78.	0.83*	1.09**	0.76**	0.89***
	(0.28)	(0.43)	(0.36)	(0.42)	(0.29)	(0.16)
Heavy Users	1.03*	1.31*	2.08***	4.11***	1.61***	−0.11
	(0.42)	(0.55)	(0.29)	(0.66)	(0.25)	(0.32)
Mainstream HS	−0.04	−0.24	−0.07	−0.43*	0.29	−0.99***
	(0.26)	(0.27)	(0.23)	(0.22)	(0.22)	(0.16)
Mod. Users: Mainstream	0.39	0.41	0.52	0.02	−0.24	0.22
	(0.31)	(0.4)	(0.38)	(0.44)	(0.31)	(0.32)
Heavy Users: Mainstream	0.16	0.51	0.66	−0.96.	−0.55	1.43***
	(0.48)	(0.56)	(0.73)	(0.56)	(0.4)	(0.41)
age	−0.04	−0.09.	−0.08	−0.12***	−0.1***	0.46***
	(0.03)	(0.05)	(0.06)	(0.03)	(0.03)	(0.02)
Male	−1.03***	−0.69***	−0.56***	0.51***	−0.6***	−0.25*
	(0.05)	(0.12)	(0.16)	(0.1)	(0.15)	(0.12)
Latinx	−0.03	−0.23**	0.02	0.01	−0.59***	0.14
	(0.05)	(0.08)	(0.13)	(0.12)	(0.05)	(0.14)
Black	−0.21.	−0.15	0.15	0.39.	−0.47***	0.06
	(0.12)	(0.19)	(0.23)	(0.22)	(0.09)	(0.1)

Note. The reference categories for the latent class and school type variables were, respectively, low users and alternative high schools. Therefore, the coefficients on “Moderate Users” and “Heavy Users” represent the differences (mean difference for violence, log odds ratio otherwise) between the low users latent class and the other two latent classes, among AHS students, after accounting for age, sex, and race. The coefficient on “Mainstream HS” gives the difference between AHS low users and mainstream low users. The coefficients on “Mod. Users: Mainstream” and “Heavy Users: Mainstream” give the differences between AHS and mainstream moderate and heavy users, respectively.

3. Discussion

AHS students are at high risk for substance use but are excluded from most school-based epidemiological studies, resulting in missed opportunities to understand how substance use operates in this higher risk group. To compare substance use and related behavioral and mental health outcomes between AHS students and students in mainstream schools, in this study we used latent class analysis to provide a more nuanced conceptualization of substance use. To date, previous studies comparing substance use between AHS and mainstream students have overwhelmingly used single variables (Grunbaum et al., 2001; Johnson et al., 2013), which can mask the heterogeneity of substance use (Salas-Wright et al., 2016) and the identification of groups with riskier patterns of use that are related to other risky behaviors and mental health problems (e.g., Bohnert et al., 2014). While AHS students are, on average, more likely than students in mainstream high schools to use tobacco, alcohol, and other drugs (Grunbaum et al., 2001; Johnson et al., 2013), the current study provides new insights into their specific patterns of use, relationship to other behavioral and mental health outcomes, and how AHS students compare to those in mainstream high schools in Texas.

Consistent with other latent class analyses that identified three to five classes of users who vary in frequency and type of substance use (e.g., Kulis et al., 2016; Parker & Bradshaw, 2015), in our analysis a three-class model with low, moderate, and heavy users was the best fit. The latent classes of substance users were similar between AHSs and mainstream schools in their patterns of drug use. However, some differences were noted between moderate users in each setting, with AHS students in this group reporting more cigarette, e-cigarette, and cigar use than their counterparts in traditional schools. This is particularly notable given the general decrease in cigarette use among adolescents overall and the drastic increase in e-cigarette use (CDC, 2017). Given that cigarette use remains high among AHS students, programs and policies geared towards addressing tobacco use among AHS students should address both traditional and electronic cigarettes. Future studies to assess differences and similarities between school policies and programs related to tobacco use in AHSs and traditional high schools are warranted to better understand how appropriately these school- and district-level factors are tailored to the needs of students in each setting.

Although patterns among low, moderate, and heavy substance users were relatively similar between AHSs and mainstream high schools, the distribution of students across the three classes was drastically different, with more AHS students than mainstream high school students being classified as moderate (41% vs. 18%) or heavy (13% vs. 4%) users. This supports previous studies finding higher levels of substance use among AHS students, measured individually as single substances (Grunbaum et al., 2001; Johnson et al., 2013), and advances our knowledge of substance use among AHS students by providing novel information about patterns of substance use in this population. Altogether, our study contributes to a relatively small body of literature regarding substance use disparities in AHS students, showing that, as compared to students in traditional high schools, where nearly 80% of students were classified as low users, over half (53%) of AHS students can be considered moderate or heavy users. The high levels of prescription drug misuse in both the moderate (60%) and heavy (94%) user groups from AHSs is particularly concerning, given the present opioid epidemic and warrants further study to understand what measures are being taken by AHSs to address it. Given that use of “lighter” drugs such as cigarettes, e-cigarettes, alcohol, and marijuana was commonly reported by both the moderate and heavy use groups, known recent use of these substances on a weekly basis or more warrants further assessment for prescription drug misuse.

These findings have important implications for addressing substance use and co-occurring health-risk behaviors among AHS students. Ultimately, there were more students in the heavy use category in the AHS sample. As compared with those in the moderate use group, who may have been using a variety of “soft” drugs to cope with stress or other adversities, those in the heavy use category may have been exhibiting the greatest risk for addiction. Secondary and tertiary prevention efforts are needed to reach moderate and heavy users and will ultimately reach a greater proportion of students in the AHS setting than in mainstream high schools. The higher use of illicit substances such as marijuana also places AHS students—who are already in the “school-to-prison pipeline” and have experienced the impact of unequally enforced educational policies—at greater risk for involvement in the criminal justice system. Once in the criminal justice system, as youth of color, they face greater risk for harsher punishments than those received by their white peers who use the same illicit substances (Bunting et al., 2013; Golub et al., 2007). Therefore, in addition to secondary and tertiary interventions that target student behaviors and the influence of peers and families, AHS students may need more legal resources and support to ensure that they are treated fairly in the criminal justice system, should they become involved in it.

Our analysis also provided novel insights into how different patterns of substance use were correlated to other behavioral and mental health outcomes among AHS students relative to peers in mainstream high schools. Not surprisingly, given the results of previous LCAs (Kulis et al., 2016; Morean et al., 2016; Parker & Bradshaw, 2015; Riehman et al., 2009; Salas-Wright et al., 2016) and what is known about the clustering of health-risk behaviors among adolescents (Bernat & Resnick, 2006), the prevalence of other health-risk behaviors and mental health challenges generally increased in a stepwise fashion across low, moderate, and heavy substance use groups in both samples. The one exception was for sexual activity in the AHS sample, with those in the moderate use category (67%) reporting being more likely than the low-use (50%) and heavy-use groups (46%) to report being sexually active. In the mainstream sample, moderate and heavy users were equally likely to report being sexually active. This was a curious finding that warrants further investigation to understand the nature of students’ reported sexual activity across substance use groups; for example, there may be differences between low, moderate, and heavy users in the number of sex partners they have, whether they are engaged in casual sex or sex in the context of a relationship, and whether they are using substances in the context of sexual encounters (none of which was analyzed in the present study).

For the most part, comparisons between AHS and mainstream high school students who were categorized in the same latent classes were not significantly different between samples. Again, sexual activity was an exception, with AHS students in the low (54%) and moderate (67%) use groups being more likely to report sexual activity than mainstream students in the low (19%) and moderate (42%) use groups. This disparity may be attributable, in part, to certain characteristics of AHS students that make them more likely to initiate and sustain sexual activity; for example, AHS students tend to be older and therefore more likely to be sexually active. AHSs also serve a high proportion of adolescent parents, and a disproportional amount of students have experienced adverse childhood experiences that increase the likelihood of being sexual active at a younger age. Yet, even in our analyses that controlled for age, sexual activity was still strikingly higher among the low and moderate AHS groups. Interestingly, there was no statistical difference in sexual activity between heavy users in AHSs (46%) and mainstream (43%) high schools. Further investigation is warranted to understand how the interplay between different patterns of substance use and sexual activity differs between students in AHSs and in mainstream high schools in order to design and implement appropriate policies, programs, and services in each setting.

Finally, moderate users in mainstream high schools were more likely to report attempting suicide than moderate users in AHSs (25% vs. 17%). This may suggest that students in mainstream high schools who use moderate to high levels of alcohol, marijuana, and tobacco are more likely to be using it to cope with stress and mental health problems. Although this is certainly a plausible explanation for AHS students, there may also be other distinct groups of moderate users that are doing so recreationally as opposed to using as a coping mechanism.

3.1. Limitations

The findings of this study must be interpreted within the context of several limitations. First, given that no data were available from Texas mainstream high schools for 2015, we had to use 2017 YRBS data from mainstream schools as a comparison for our 2015 data from AHSs. Second, the Texas YRBS data are representative of the entire state and cannot be analyzed at the level of the county or school district, whereas our AHS data were drawn from the Central Texas region. Third, while the 2017 YRBS data are representative of mainstream high schools in Texas, our 2015 data for AHSs were not representative of AHS students in our region, because we used a convenience sample with a low participation rate. Actual disparities between AHS students and those in mainstream schools are likely greater than what is reported here as our data likely underestimate the prevalence of substance use and co-occurring risk behaviors in AHSs, given that students involved in more high-risk behaviors are less likely to be represented in survey research (Courser, Shamblen, Lavrakas, Collins, & Ditterline, 2009; Tigges, 2003). Additionally, substance use and other associated behaviors may not overlap entirely, given that variables were measured over different time periods (e.g., 12 months, 3 months, 1 month). There was also a change in the question about prescription drug use between the 2015 and 2017 surveys; however, when we did a sensitivity analysis in which we removed the prescription drug question, the results remained essentially the same. Finally, our use of the terms “hard” and “soft” drugs was somewhat problematic, given varying uses of these terms in the literature (Janik, Kosticova, Prof, & Turcek, 2017). We did, however, define each term for the purposes of this study.

Despite these limitations in comparing the two groups, the sample of mainstream high school students provided an important context for interpreting the findings from AHS settings. Currently, no representative data from any state or region of the U.S. are available for AHS students. In fact, the only known source of representative public health surveillance data for AHSs is two decades old, from the 1997 YRBS conducted nationally in AHSs. In our analysis, therefore, we used the data that were available in order to compare AHS and mainstream high school settings. We strongly recommend that resources be devoted to establishing consistent public health surveillance systems for AHSs that can yield representative data over time that can be compared with data from mainstream high schools. Such data will allow for more scientifically sound comparisons between these two settings and allow scientists and practitioners to draw stronger conclusions about the data as they develop programs and policies to address disparities in adolescent substance use and associated health-risk behaviors. Until then, we must continue to use what little public health surveillance data are available to highlight disparities between AHSs and mainstream high schools to inform action.