Homelessness

Psychology of Addictive Behaviors Understanding Pathways Between PTSD, Homelessness, and Substance Use Among Adolescents Jordan P. Davis, Graham Diguiseppi, Jessenia De Leon, John Prindle, Angeles Sedano, Dean Rivera, Benjamin Henwood, and Eric Rice Online First Publication, July 25, 2019. http://dx.doi.org/10.1037/adb0000488

CITATION Davis, J. P., Diguiseppi, G., De Leon, J., Prindle, J., Sedano, A., Rivera, D., Henwood, B., & Rice, E. (2019, July 25). Understanding Pathways Between PTSD, Homelessness, and Substance Use Among Adolescents. Psychology of Addictive Behaviors. Advance online publication. http://dx.doi.org/10.1037/adb0000488

Understanding Pathways Between PTSD, Homelessness, and Substance Use Among Adolescents

Jordan P. Davis, Graham Diguiseppi, Jessenia De Leon, John Prindle, Angeles Sedano, Dean Rivera, Benjamin Henwood, and Eric Rice

University of Southern California

Youth experiencing homelessness have been shown to experience high levels of both trauma and substance use. However, prior work has yet to consider how substance use, posttraumatic stress disorder (PTSD) symptoms, and homelessness are temporally, or reciprocally, associated over time. The current study uses symptom-driven and experience-driven models to examine the reciprocal relationships between substance use, PTSD symptoms, and homelessness among a large sample of adolescents receiving substance use treatment in the United States. Adolescents (n � 20,069; Mage � 15.6; 74% male) completed baseline, 3-, 6-, and 12-month assessments. Autoregressive latent trajectory with structured residual (ALT-SR) models were used to examine within- and between-person relationships. We found continued support for prior work at the between-person level of analysis. At the within-person level, during the treatment phase, PTSD emerged as a key mechanism predicting both return to use and increased days of homelessness posttreatment. Further, greater substance use at treatment completion was associated with greater PTSD symptoms and homelessness, prospectively. The current study extends the previous work to consider individual level processes in conjunction with overarching event level predictors of homelessness. We found that PTSD symptomology is a driving factor that influences, both directly and indirectly, experiences of homelessness posttreatment. Interventions may wish to incorporate trauma informed approaches for youth entering treatment as this may mitigate long-term experiences of homelessness and return to substance use.

Keywords: addiction treatment, substance use, posttraumatic stress disorder, homelessness, adolescents

Current estimates indicate that, in the United States, approxi- mately 184,000 (0.7%) adolescents aged 14 to 17 in the United States received substance use treatment in the past year (Substance Abuse and Mental Health Services Administration, 2018). A num- ber of studies have reported lifetime incidence of posttraumatic stress disorder (PTSD) among (mostly adult) treatment samples to be between 26% and 52% (Driessen et al., 2008; Reynolds, Hinchliffe, Asamoah, & Kouimtsidis, 2011; Schäfer et al., 2010).

Of great concern are subpopulations of adolescents who have experienced, or are at risk of experiencing, homelessness (Davies & Allen, 2017). Youth who experience homelessness are a highly vulnerable population with high rates of abuse, trauma, violence, and substance use (Davies & Allen, 2017).

Previous developmental research examining intersections be- tween homelessness, posttraumatic stress and substance use has involved samples of youth drawn from drop-in centers or shelters (Milburn et al., 2009; Milburn et al., 2019; Tyler, Johnson, & Brownridge, 2008; Whitbeck, Hoyt, & Yoder, 1999). However, there is a dearth of evidence documenting risks for experiencing homelessness among adolescents receiving substance use treat- ment. Further, among the few longitudinal studies involving run- away and homeless youth, few have examined the complex, re- ciprocal relationships between PTSD symptoms, substance use, and homelessness episodes over time. Therefore, the current study seeks to address these gaps by examining bidirectional relation- ships between PTSD symptoms, substance use, and episodes of homelessness over time among adolescents receiving substance use treatment.

Theoretical and Empirical Considerations

Many adolescents participating in substance use treatment, also have histories of homelessness (Folsom et al., 2005) and traumatic stress (Berry & Sellman, 2001; Cottler, Compton, Mager, Spitzna- gel, & Janca, 1992) and, unfortunately, continue to experience these maladies after exiting substance use treatment. There are two

X Jordan P. Davis, Centers for Artificial Intelligence in Society and Mindfulness Science, University of Southern California; Graham Diguiseppi and Jessenia De Leon, Suzanne Dworak-Peck School of Social Work, University of Southern California; John Prindle, Children’s Data Network, University of Southern California; Angeles Sedano, Dean Rivera, Benjamin Henwood, and Eric Rice, Suzanne Dworak-Peck School of Social Work, University of Southern California.

The data reported in this article were obtained from publicly available data via the Global Appraisal of Individual Needs website (gaincc.org). A bibliography of journal articles, working papers, conference presentations, and dissertations using Global Appraisal of Individual Needs data is available at http://gaincc.org/publications/. Data from this article have not been published or presented elsewhere.

Correspondence concerning this article should be addressed to Jordan P. Davis, Centers for Artificial Intelligence in Society and Mindfulness Sci- ence, University of Southern California, 669 West 34th Street, Los Ange- les, CA 90089. E-mail: [email protected]

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

Psychology of Addictive Behaviors © 2019 American Psychological Association 2019, Vol. 1, No. 999, 000 0893-164X/19/$12.00 http://dx.doi.org/10.1037/adb0000488

1

theoretical models that may help aid our understanding of how trauma symptomology may influence experiences of homelessness and eventual return to substance use following treatment dis- charge: symptom-driven models and experience-driven models.

Symptom-driven models posit that psychological symptoms (e.g., symptoms of PTSD, depression, or other psychological dis- order) precede long-term cascades of negative behavioral health outcomes (Agoston & Rudolph, 2013; Davis et al., 2018; Lewin- sohn, Steinmetz, Larson, & Franklin, 1981). Originally used as a theoretical orientation for depression, symptom-driven models have since been used to understand how symptomology of other disorders can put individuals at greater risk of maladaptive inter- personal outcomes. Evidence of symptom-driven models is abun- dant. Prior research has shown that, among homeless individuals, severity of PTSD symptoms (e.g., no symptom improvement fol- lowing intervention) predicted increased substance use following treatment discharge (Burns, Lehman, Milby, Wallace, & Schum- acher, 2010). Others have noted that the onset of PTSD precedes the onset of homelessness (Hodgson, Shelton, van den Bree, & Los, 2013; North & Smith, 1992). Further, in a small, cross- sectional study, PTSD symptoms were directly related to increased alcohol use among a sample of homeless mothers (Yeater, Austin, Green, & Smith, 2010). Studies of homeless youth have also found that for every additional experience of victimization, the odds of meeting criteria for a substance use disorder nearly double (Bender, Brown, Thompson, Ferguson, & Langenderfer, 2015), and meeting criteria for PTSD was associated with increased odds of having a substance use disorder (Bender, Ferguson, Thompson, Komlo, & Pollio, 2010). Finally, among an analog population (military vet- erans), trauma symptomology and alcohol use were associated with time to homelessness for women (Benda, 2006). Thus, PTSD symptomology may serve as a catalyst for a larger pattern of reciprocal problems following treatment discharge. As such, using a symptom-driven perspective can aid in determining if PTSD symptomology precedes experiences of homelessness or increases in substance use following adolescents’ discharge from substance use treatment.

We next set out to explore how experiences of homelessness may drive both short and long-term psychopathology and maladaptive behaviors. We conceptualize this pathway as an experience-driven risk model. This framework is borrowed from long-standing con- ceptual and empirical work based on interpersonal risk models (Agoston & Rudolph, 2013). Interpersonal risk models were born out of the bullying (peer)-victimization literature and propose that experiencing bullying victimization initiates a cascade of problems such as increased psychopathology and problematic substance use (Davis et al., 2018). Prior research involving homeless youth lends support for experience-driven risk models. For example, cross- sectional work has shown that homeless youth who are more transient likely have higher rates of PTSD and alcohol dependence (Bender et al., 2010). Others have found that, once youth become homeless, there is an increase in psychological (e.g., PTSD, de- pression) and substance use diagnoses (Martijn & Sharpe, 2006). In the current study, traumatic events experienced while currently home- less may increase trauma symptomology and substance use, either through social learning mechanisms while on the street and/or as a means to cope with trauma symptoms.

Disaggregating Between- and Within-Person Effects

Despite previous longitudinal work investigating the associa- tions between substance use, PTSD symptoms, and homelessness (either together or separately), many of these studies are limited in their scope and ability to accurately interpret estimated paths. Even among the few studies that have incorporated directionality in their models (Tyler & Schmitz, 2018; Whitbeck, Hoyt, & Bao, 2000), many have not assessed bidirectionality (e.g., transactional pro- cesses) or appropriately disaggregated variance into their two components: between-person and within-person. Typically, when researchers attempt to understand reciprocal relationships, they use autoregressive cross-lagged models. These models yield estimates that are a combination of both between-person and within-person variance which are difficult (if not impossible) to interpret, be- cause they are a combination of between-and within-person vari- ance (Berry & Willoughby, 2017). This can have profound effects on the interpretability of the bidirectional (or unidirectional) asso- ciations between two or more variables. We use recent advances in modeling longitudinal relationships that allow for the disaggrega- tion of both within- and between-person effects. Recent models for longitudinal data, such as the autoregressive latent trajectory model with structured residuals (ALT-SR) introduced by Curran, Howard, Bainter, Lane, and McGinley (2014), improve our ability to understand cross-lagged or reciprocal relationships over time (Berry & Willoughby, 2017; Davis et al., 2018; Merrin, Davis, Berry, D’Amico, & Dumas, 2016). Specifically, the ALT-SR model allows one to simultaneously consider between-person re- lations among more systematic— or trait-like—aspects of sub- stance use and, say, homelessness (e.g., mean levels, growth rates), while simultaneously modeling reciprocal relations between these variables as they manifest within individuals over time. This modeling approach invites two advantages. First, it anchors the reciprocal processes at an arguably more meaningful and relevant level of analysis—within-person. Second, the internal validity of the reciprocal effects is strengthened as each individual serves as his or her own control (and therefore, all time-invariant confounds are controlled).

Summary and Hypotheses

While prior studies have attempted to ascertain mental health correlates of homelessness, many of the studies are cross-sectional in nature and, thus, cannot determine temporal order. Thus, while there has been some longitudinal work both predicting homeless- ness (e.g., among a non-homeless sample) and predicting sub- stance use and psychopathology among homeless samples, no longitudinal study has attempted to discern the temporal order of events related to PTSD symptomology, substance use, and expe- riences of homelessness. Further, most prior work has only con- sidered trauma related events and not symptomology (e.g., PTSD) that may be a unique factor predicting risk of homelessness and future problems (e.g., increased substance use). In the current study, we test symptom-driven and experience drive risk models. Specifically, we sought to examine the overall between-person associations among our variables of interest including the initial levels (i.e., intercepts) and change (i.e., slopes). Thus, we hypoth- esize (Hypothesis 1) small to moderate between-person associa- tions among PTSD symptoms, substance use, and homelessness episodes.

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

2 DAVIS ET AL.

Our hypotheses for within-person associations will examine temporal associations from two theoretical perspectives: symptom- driven and experience-driven risk model. Given that we only have two time points during the treatment phase (baseline to 3-month follow-up) our hypotheses are directed at the posttreatment phase (3-, 6-, and 12-month follow up). We will, however, report any emergent findings during the treatment phase as these can inform clinical practice and theory. Hypothesis 2 reflects a symptom- driven model. After controlling for lifetime experiences of victim- ization, we sought to understand whether symptoms of PTSD are driving pathways to increased experiences of homelessness and return to substance use following completion of substance use disorder treatment. Hypothesis 3 reflects an experience-driven risk model. Here we seek to understand if experiences of homelessness precede increased PTSD symptomology and subsequent return to substance use following treatment discharge. Based on prior liter- ature we hypothesize that experiences of homelessness will predict increased PTSD symptomology as well as increased substance use.

Method

Participants and Procedure

Participants were 20,069 adolescents aged 15.6 years on average (SD � 1.21), receiving substance use treatment in treatment set- tings throughout the United States that utilized the Global Ap- praisal of Individual Needs (GAIN) set of assessments. Data were collected with the informed consent of individuals receiving treat- ment as part of general clinical practice or in the context of a research study at each site, which were approved by each site’s respective Institutional Review Board. The majority of participants were male (74%; n � 14,811) and the sample was diverse with 36.1% identifying as White (n � 7,252), 29.9% Hispanic (n � 6,005), 15.4% African American (n � 3,086), 15.4% multiple race/ethnicity (n � 3,084), and 3.2% as other race/ethnicity (n � 634). In terms of substance use, 52.2% (n � 10,480) had lifetime substance dependence with 77.9% (n � 15,647) having any past year substance use diagnosis. On average, at intake, participants reported using cannabis on 22.5 (SD � 28.8) days, alcohol on 5.4 (SD � 11.6) days, and heavy drinking on 3.27 (SD � 8.38) days out of the past 90 days. Further, nearly 9% of the sample (n � 1,779) reported experiencing homelessness at some time between intake and their 12 month follow-up. Further, youth reported, on average, 2.12 (SD � 3.34) PTSD symptoms at baseline. See Tables 1 and 2 for more details on participant characteristics.

Data were obtained from 137 sites associated with the Center for Substance Abuse Treatment Substance Abuse and Mental Health Services Administration–funded programs that provide commu- nity outpatient substance use treatment and whose records feed a national data set managed by the GAIN Coordinating Center (Dennis, Titus, White, Unsicker, & Hodgkins, 2003). The GAIN originated as a collaborative effort between clinicians, researchers, and policymakers to create a standardized assessment tool for individuals receiving substance use and mental health services. The GAIN provides tools for initial screenings, brief interventions and referrals, clinical assessments, placement recommendations, and program/evaluation services. The GAIN assessment tool (GAIN-I) is a comprehensive, structured interview with over 100 scales situated within eight main sections including background,

substance use, physical health, risk behaviors, mental health, en- vironment, legal, and vocational. Each site that administered the GAIN went through extensive training (2–3 months for certifica- tion) that included practice sessions, observations, and supervised administration. Typically, the GAIN takes between 90 and 120 min per patient to complete. All assessments are completed on a computer adaptive system that includes skip patterns and prompts when discrepancies occur to ensure accurate data collection.

Youth entering treatment came from a variety of referral sources, including self-referral (1%), family or friends (7%), criminal jus- tice system (23%), school or work (4%), social services agency (3%), or a behavioral health/state treatment facility (4%). Not all participants had data on referral source. At treatment entry, each person completed the initial GAIN assessment (GAIN-I). After the initial assessment, participants were referred to receive treatment (with treatment type varying by site) and completed GAIN follow-up assessments at 3, 6, 9, and 12 months. All participants were given the same GAIN follow-up assessments which were completed with the assistance of GAIN trained staff.

In general, adolescents received a variety of interventions in- cluding Adolescent-Community Reinforcement Approach (n � 7,461, 37.1%), motivational enhancement therapy/cognitive– behavioral therapy (n � 7,213, 35.9%), other evidence based treatments (such as cognitive– behavioral therapy or motivational interviewing; n � 1,537, 7.6%), multidimensional family therapy (n � 392, 2.0%), specific manualized programs (e.g., site specific programming; n � 548, 2.7%), seven challenges (n � 220, 1.1%), and what would be considered “treatment as usual” (e.g., group therapy, 12-step facilitation; n � 1,970, 9.8%). Finally, the ma- jority of participants entered treatment as early intervention, out- patient, or intensive outpatient (88.2%), residential facility (7.7%), or posttreatment continuing care (4.0%).

Measures

Demographic control variables and lifetime victimization. Gender was coded with female as the reference group. Race/ ethnicity was dichotomized into two separate categories: nonwhite

Table 1 Baseline Characteristics

Characteristic M (SD) or n (%) (N � 20,069)

Demographics Age, in years 15.6 (1.21) Female, n (%) 5,258 (26.2%)

Race/Ethnicity, n (%) White 7,252 (36.1%) African American 3,086 (15.4%) Hispanic 6,005 (29.9%) Multiple race/ethnicity 3,084 (15.4%) Other 634 (3.2%)

Psychiatric disorders Posttraumatic stress disorder 2.12 (3.34) General Victimization Scale 2.98 (3.09)

Substance use Substance Frequency Scale 11.6 (13.6) Prior treatment episodes, n (%) 6,319 (31.5%)

Homelessness Days of homelessness 3.86 (6.55)

Note. Data are presented as M (SD) except where otherwise indicated.

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

3PTSD, HOMELESSNESS, AND SUBSTANCE USE IN ADOLESCENTS

and white, with nonwhite participants as the reference group. Number of prior treatment episodes (count), age (continuous), and ever been homeless (yes as reference) were all entered as covari- ates. Lifetime experience of trauma was measured using the Gen- eral Victimization Scale (GVS). The GVS is a combination of early life traumatic experiences (e.g., physical abuse, sexual abuse, emotional abuse), exposure to direct victimization (e.g., being beat up, attacked by a knife or weapon), and trauma related character- istics (e.g., known perpetrator, happened multiple times). The GVS has good internal consistency (� � .89).

Substance use. Substance use was assessed using the GAIN Substance Frequency Scale (� � .82) which is the average pro- portion of past 90-day alcohol, heavy alcohol, cannabis, illicit drug, and problems associated with substance use. Values on the substance frequency scale are expressed as an index (from 0 to 1) and then multiplied by 100, with higher scores on this scale representing increasing frequency of substance using days and days causing problems (Dennis et al., 2003). The substance fre- quency scale has been validated with adolescent samples (� � .79).

PTSD symptomology. The GAIN traumatic stress scale is a count of 13 items representing past 90-day traumatic stress symptoms that correspond to Diagnostic and Statistical Manual of Mental Disorders–IV PTSD symptomology. The items were chosen based on item response theory of the Mississippi Scale for Civilian PTSD (Vreven, Gudanowski, King, & King, 1995), and has shown good reliability in adolescent samples (0.92; Dennis, Chan, & Funk, 2006). Example items include “When something reminds you of the past, you become very distressed and upset,” “You were frightened by your urges,” and “You had a hard time expressing your feelings, even to the people you care about.” Participants responded with a “yes” or “no” to each of the items. Higher scores represent endorsing more symptoms of PTSD.

Homelessness. We assessed participant’s experiences of homelessness by using a single item indicator. Each participant was asked, “How many days, in the past 90, have you been homeless?” in which participants indicate a value from 0 (no days homeless) to 90 (everyday homeless). Thus, higher scores indicate more days of homelessness at each time point.

Analytic Plan

To address our hypotheses regarding the temporal associations between PTSD symptomology, substance use, and homelessness we fit a taxonomy of ALT-SR models (Curran et al., 2014). One advantage of the ALT-SR over traditional autoregressive cross- lagged models is that we are able to capture variance that does not change (intercept), variance that changes over the course of the study (slope), and the remaining residual variance captured by the residual errors representing the within-person cross-lagged asso- ciations. We specified the respective between-person trajectories as bilinear spline functions. Bilinear spline models are useful when data can be separated into discrete phases and when simple growth models cannot fit the functional form of the data. Thus, each discrete phase is a simple growth model (e.g., linear or quadratic) and the segments that connect the phases of growth are knot or transition points. In the current study we used bilinear spline models to explain change during the treatment phase (baseline to 3-month follow-up) as well as differential change during the post treatment phase (3-month follow-up to 12-month follow-up) with the intercept centered at the pretreatment assessment. All models tested for the presence of a quadratic effect during the posttreat- ment phase using changes in model fit (�2 log likelihood [�2LL] ratio test). When assessing the functional form of each variable, independently, we found a bilinear spline model did not fit the data better than a simple linear latent growth model for homelessness (�2��2LL � 43.5, �df � 1, p � .001). Both substance use and PTSD symptomology growth models fit best using bilinear spline models. Results of our model building process suggested that random intercepts, random linear treatment phase slopes, and fixed linear posttreatment slopes fit the data best. The latent intercepts represent the estimated population mean level and (residual) between-person variance of the given variable (i.e., baseline). The mean of the latent slope factors (both treatment and posttreatment slopes) represent the between-person variance of the change or growth of the given variable. Thus, our between person effects are captured by correlating our random latent growth factors (repre- sented by �standardized). Doing this allows our remaining within- person variance to be “pushed” into the residual autoregressive and cross-lagged portions of the model. Here, we create latent variables from the residual (e.g., error) variance from our

Table 2 Correlations Among Variables of Interest at Each Time Point

Variable (follow-up period) 1 2 3 4 5 6 7 8 9 10 11 12

1. Substance use (baseline) 1 2. Substance use (3 months) .315 1 3. Substance use (6 months) .250 .507 1 4. Substance use (12 months) .217 .385 .472 1 5. PTSD (baseline) .181 .066 .080 .060 1 6. PTSD (3 months) .111 .102 .073 .068 .370 1 7. PTSD (6 months) .079 .106 .135 .063 .306 .518 1 8. PTSD (12 months) .070 .057 .097 .128 .262 .397 .464 1 9. Homeless (baseline) .097 .01 .018 .01 .093 .102 .079 .076 1

10. Homeless (3 months) .023 .088 .032 .029 .051 .091 .092 .066 .269 1 11. Homeless (6 months) .032 .055 .123 .036 .047 .102 .113 .123 .157 .312 1 12. Homeless (12 months) .045 .052 .068 .141 .073 .119 .122 .163 .115 .145 .206 1 M (SD) 11.57 (13.6) 6.05 (10.4) 6.11 (10.9) 7.06 (12.3) 2.12 (3.34) .67 (2.01) .51 (1.77) .46 (1.67) .86 (6.54) .75 (6.14) .75 (6.42) .92 (7.24)

Note. PTSD � posttraumatic stress disorder.

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

4 DAVIS ET AL.

observed variables. These newly created latent variables are used as predictors and outcomes (e.g., each variable is a pre- dictor and outcome in a cross-lagged model) in our within- person portion of the model.

Our taxonomy of models first established basic within-person autoregressive associations among our variables of interest (Model 1). Second, we freed the within-person cross-lagged associations between substance use, PTSD symptomology, and homelessness (Model 2). Because we are interested in associations between our variables of interest after youth complete treatment, we also esti- mated our within-person cross-lagged effects as two discrete phases (e.g., treatment phase and posttreatment phase). We used model constraint tests to determine if cross-lagged effects, autore- gressive parameters, and within-time correlations could be con- strained to be equal over time. Results of our model building process revealed improved model fit when autoregressive compo- nents of both PTSD and substance use were allowed to be freely estimated between baseline and the initial 3-month follow-up and constrained to be equal during the posttreatment phase. Within- time correlations were allowed to be freely estimated at baseline and constrained to be equal over time. Further, we found our model fit better when cross-lagged estimates were allowed to be freely estimated during the posttreatment phase (vs. constrained to be equal over time; �2��2LL � 20.7, �df � 6, p � .02). We used standard fit statistics to assess improvement in model specification. We used comparative fit index of .95 or greater, root mean square error of approximation of .05 or less, and standardized root mean square residual of less than .08 to indicate excellent model fit.

The GAIN data maintain an unbalanced study design. That is, some individuals have missing data simply due to the amount of time elapsed between baseline and study design (e.g., typical attrition). The majority of our missing data can be explained by censoring, or individuals who did not have an opportunity to provide data. Of the participants who could have provided data, approximately 16% and 23% of the participants showed a missing- data pattern consistent with attrition between 3 and 6 or 6 and 12 months after the start of treatment, respectively. In our missing data analysis we found, among our covariates, male, �2 � 20.4(1), p � .01, those youth who identified as racial/ethnic minority, �2 � 24.8(1), p � .01, and those who had a lifetime experience of homelessness, �2 � 8.45(1), p � .01, had more missing data. Prior substance use treatment, �2 � 1.24(1), p � .27, and lifetime victimization scores, t � 0.32(20,057), p � 0.74, were not asso- ciated with missingness. For individuals who were able to provide data we utilized full-information maximum likelihood estimator (Mplus Version 8; Muthén & Muthén, 1998 –2017), treating all observed predictors as single-item latent variables. As such, each individual contributes whatever the data they have to the likeli- hood function (i.e., both X and Y variables). If we assume our data are missing at random (assumption of full information maximum likelihood) our estimates would be considered unbiased (Enders, 2011). Given that prior values on X and Y variables are often reasonable predictors of missingness in longitudinal data, this lends support to the plausibility of this assumption. To adjust for nonnormality all standard errors were bootstrapped (iterations � 10,000).

Results

Below we report unstandardized estimates (b) as well as stan- dardized estimates ( ; which are not found in the figures or tables). Between-person correlations are represented by �standardized.

Overall Mean Trajectories

In our unconditional models, substance use ( � �5.25, SE � 0.09, p � .01) and PTSD symptomology ( � �1.45, SE � 0.03, p � .01) showed steep declines from baseline to 3 months (treat- ment phase linear growth). Growth following the treatment phase (shift parameter) slightly increased for substance use ( � 0.22, SE � 0.04, p � .01); however, PTSD symptomology ( � �0.08, SE � 0.01, p � .01) showed slight decreases through 12-month follow up during the posttreatment phase. Homelessness did not require a bilinear spline model, with participants showing small, and nonsignificant, increases in days of homelessness throughout the study period ( � 0.03, SE � 0.01, p � .40).

Hypothesis 1

Between-person associations. The intercept and slope factors indicated moderate to strong associations for between-person sub- stance use, PTSD symptoms, and days of homelessness. Between- persons, an individual who reported higher initial levels of sub- stance use also reported more PTSD symptoms (�standardized � 0.36) than those who engaged in less substance use at treatment intake. Thus, a standard deviation increase in substance use at treatment entry is associated with a 1.20-symptom increase in PTSD. Interestingly, we did not find a significant association between initial levels of homelessness with substance use (�standardized � 0.09) or PTSD symptoms (�standardized � 0.03). In terms of change (e.g., slope to slope correlations) during treatment, we found individuals who reported change in substance use also reported positive change trajectories in PTSD symptomology (�standardized � 0.29). This correlation can be interpreted as such: a standard deviation increase in changes in substance use during treatment is associated with a 0.70 increase in PTSD symptom change. That is, given both slopes are negative during treatment, when youth report increases in substance use the rate of change in PTSD symptomology decreases more slowly. We did not find that changes in PTSD (�standardized � 0.03) or substance use (�standardized � 0.05) were associated with changes in homeless- ness during the treatment phase.

Within-person cross-lagged associations. The final within- person cross-lagged portion of our models are presented in Figure 1. All significant pathways are represented in Figure 1, which demonstrates the lagged effects of each variable over time. All effects are presented in Table 3.

Hypothesis 2: Symptom-Driven Model

We sought to examine the presence of a symptom-driven model, where, after controlling for lifetime victimization, long-term prob- lems (e.g., continued substance use and days of homelessness) will be driven by PTSD symptomology. Our final model evidenced excellent model fit (comparative fit index � 0.96, Tucker–Lewis index � 0.91, root mean square error of approximation � 0.03, standardized root mean square residual � 0.02). We found partial

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

5PTSD, HOMELESSNESS, AND SUBSTANCE USE IN ADOLESCENTS

support for our hypothesis of a symptom-driven model (see Figure 1). That is, during the posttreatment phase, when tracing pathways starting from PTSD (from Time 2), we found that reporting higher PTSD symptomology than one’s typical average was directly associated with reporting more days of homelessness than one’s typical average at the next time point (b � 0.15, 95% confidence interval [CI: 0.02, 0.28]; � 0.04). Further, we can see that youth who reported higher PTSD symptomology than their own average at 3 months reported higher frequency of substance use at 6 months (b � 0.36, 95% CI [0.15, 0.57]; � 0.09), and this increased substance use was associated with experiencing more days of homelessness at 12 months (b � 0.03, 95% CI [0.01, 0.05]; � 0.06).

Alternatively, we also found evidence of a cascade of negative problems stemming from substance use. Specifically, those who reported a higher prevalence of substance using days than one’s typical average at 3 months (e.g., directly following treatment completion) reported higher PTSD symptoms than one’s typical level at 6 months (b � 0.03, 95% CI [0.02, 0.04]; � 0.20). Subsequently, this increased PTSD symptomology at 6 months was associated with reporting more days of homelessness than one’s typical average at 12 months (b � 0.35, 95% CI [0.08, 0.62]; � 0.08).

Hypothesis 3: Experience-Driven Risk Model

Here, we sought to explore how experiences of homelessness may drive long-term negative consequences such as increased PTSD symptomology or return to substance use posttreatment. Looking at Figure 1, we did not find support for any experience- driven risk during the posttreatment phase. That is, experiences of homelessness did not influence PTSD symptomology or substance use posttreatment. However, if we look at the treatment phase we did find that youth who reported more days of homelessness than their typical average at baseline reported higher prevalence rates of substance use (b � 0.08, 95% CI [0.03, 0.13]; � 0.05) and

higher PTSD symptomology (b � 0.06, 95% CI [0.02, 0.10]; � 0.22) than one’s typical average immediately following treatment (3 months). If we continue to follow one of these pathways, we see that this greater substance use at 3 months is associated with reporting higher PTSD symptoms than one’s typical average at 6 months, and subsequently this increased PTSD symptomology is associated with reporting more experiences of homelessness than one’s typical average at 12 months.

Discussion

The current study provides the first longitudinal evidence, in populations who are not exclusively homeless, of temporal order of events linking PTSD symptomology, experiences of homeless- ness, and substance use following substance use treatment com- pletion. Overall, we found support for both symptom-driven and experience-driven risk models. That is, even after controlling for lifetime experiences of victimization, we found PTSD symptom- ology to be a key mechanism leading to both homelessness and substance use following treatment (symptom-driven models). Fur- ther, we found that, while experiences of homelessness did not emerge as an important mechanism during the posttreatment phase, early experiences of homelessness (at baseline) were a catalyst that triggered a cascade of heightened PTSD symptomol- ogy and greater levels of substance use that eventually led to more days of homelessness. Results from the current study highlight the importance of untreated trauma symptomology, posttreatment, and a better understanding how youth homelessness (when entering treatment) can influence long-term problems. Our results suggest directions for initial assessment of homelessness among youth entering treatment as well as a focus on targeting PTSD sympto- mology among at risk youth entering substance use treatment.

Many youth who become homeless are either forcibly removed from their homes or leave in response to abuse by their caregivers. In general, results from the current study provide continued sup- port for prior work on youth experiencing homelessness. For

Treatment Phase Post-Treatment Phase

.03 (.01) .03 (.01)

.36(.11)

.06 (.02)

.03 (.004)

.18 (.07)

.16 (.06)

.63 (.32)

.15 (.07)

0.40 (0.02)

.35 (.14)

.08 (.03)

.67 (.23)

.02 (.004).21 (.09)

0.40 (0.02) 0.40 (0.02)

1.02 (.12) .16 (.02) .16 (.02)

.04 (.01) .04 (.01) .04 (.01)

0 3 6 12

0 3 6 12

0 3 6 12

Figure 1. Autoregressive latent trajectory with structured residual final model. Bold lines indicate a significant path; gray dash lines indicate a nonsignificant path. All estimates can be found in Table 3. sub � substance use; PTSD � posttraumatic stress disorder; Hmls � days of homelessness.

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

6 DAVIS ET AL.

example, early work has found that a large majority of homeless and runaway youth have experienced abuse by a parent or adult caretaker (Whitbeck, Hoyt, & Ackley, 1997). Others have found that youth with more experiences of victimization report more time homeless, more substance use and more posttraumatic stress rel- ative to those youth who experience less victimization (Hsu et al., 2018; Radu, 2017; Rice, Milburn, Rotheram-Borus, Mallett, & Rosenthal, 2005; Whitbeck, Hoyt, Johnson, & Chen, 2007; Whit- beck et al., 1999; Yoshioka-Maxwell & Rice, 2017). Prior studies have found that many homeless youth were abused before the age of four, and generally have higher rates of child abuse and neglect

than the general population (McManus & Thompson, 2008). Not surprisingly, experiences of abuse in the home or community is one of the most robust risk factors for substance using youth (Slesnick, Bartle-Haring, & Gangamma, 2006). In fact, youth who are current substance users or are in treatment were upward of 12 times more likely to report experiencing physical abuse and 18 –21 times more likely to have a history of sexual abuse, compared to peers without substance use problems (Simpson & Miller, 2002).

The current study examines the within-person associations be- tween substance use, traumatic symptoms, and days of homeless- ness. We found partial support for our symptom-driven hypothesis that PTSD symptoms would initiate long-term problems (e.g., return to substance use, homelessness) posttreatment. Specifically, we found (after controlling for lifetime victimization) a direct association between PTSD symptomology and increased experi- ences of homelessness at the within-person level of analysis. Interestingly, we also found a direct association between height- ened levels of substance use and experiences of homelessness. These findings are in line with our hypotheses and prior literature. That is, research has found that exposure to violence and trauma related symptomology may drive experiences of homelessness among youth. Among longitudinal studies, prior research has found that somatic symptoms (e.g., physical health), symptoms of depression, and exposure to violence during adolescence are sig- nificant predictors of experiencing homelessness in young adult- hood (van Den Bree et al., 2009). Other studies have found that experiences of victimization (e.g., witnessing or direct trauma expo- sure), and symptoms of depression or substance use disorder all predicted a higher probability of youth running away from home (Slesnick, Guo, Brakenhoff, & Feng, 2013; Tucker, Edelen, El- lickson, & Klein, 2011; Tyler & Bersani, 2008). While we did find a direct association between PTSD symptomology and heightened experiences of homelessness, we also found several ‘pathways’ that may lead to homelessness and return to substance use. Spe- cifically, we found reporting more PTSD symptoms at treatment completion was associated with higher frequency of substance use (at the next time point). This higher rate of substance use was subsequently associated with increased experiences of homeless- ness 1-year posttreatment. While prior studies have posited that homelessness experiences should increase both posttraumatic stress and substance use (Hsu et al., 2018; Radu, 2017; Rice et al., 2005; Tyler et al., 2008; Whitbeck et al., 2007; Yoshioka-Maxwell & Rice, 2017), our results provide a more nuanced cascade of events in which homeless experiences are linked to posttraumatic stress and, subsequently, increased substance use. Theoretically, a symptom-driven perspective may aid in understanding what is hap- pening at the individual level. Not only are traumatic events a driver of within-person differences in experiences of homelessness, but individual increases in traumatic symptoms relative to personal norms can lead to increased time spent homeless, or indirectly lead to increased time spent homeless through increased substance use.

Our results also provide some insight into clinical implications. In addition to finding PTSD as a driving factor predicting home- lessness, we also found a full cross-lagged association between PTSD symptomology and substance use, which eventually leads to higher rates of homelessness experiences posttreatment. Thus, it may be that youth leaving substance use disorder treatment have residual, or unattended, trauma symptomology leading to a cas- cade of negative events. A substantial body of research has shown

Table 3 Final Autoregressive Latent Trajectory With Structured Residual Model: Associations Between Posttraumatic Stress Disorder (PTSD) Symptomology, Substance (Sub) Use, and Homelessness (Hmls)

Path estimates Final model, parameter

estimate [95% CI]

Within-person cross-lags Treatment phase

Sub Use3m on PTSDt .76 [�.09, 1.61] PTSD3m on Sub Uset .21 [.04, .38] Hmls3m on PTSDt .22 [�.01, .45] PTSD3m on Hmlst .06 [.02, .10] Hmls3m on Sub Uset .02 [�.05, .09] Sub Use3m on Hmlst .08 [.03, .13]

Posttreatment phase Sub Use6m on PTSD3m .36 [.15, .57] PTSD6m on Sub Use3m .03 [.02, .04] Hmls6m on PTSD3m .15 [.02, .28] PTSD6m on Hmls3m .01 [�.001, .02] Hmls6m on Sub Use3m .03 [.01, .05] Sub Use6m on Hmls3m .04 [�.10, .07] Sub Use12m on PTSD6m .67 [.21, 1.11] PTSD12m on Sub Use6m .02 [.01, .04] Hmls12m on PTSD6m .35 [.08, .62] PTSD12m on Hmls6m �.001 [�.03, .02] Hmls12m on Sub Use6m .03 [.01, .05] Sub Use12m on Hmls6m �.01 [�.06, .04]

Autoregressive PTSD3m on PTSDb 1.03 [.77, 1.27] Sub Use3m on Sub Useb .40 [.36, .42] Hmls3m on Hmlsb .04 [.01, .06] PTSDt�1 on PTSDt .16 [.12, .19] Sub Uset�1 on Sub Uset .40 [.36, .42] Hmlst�1 on Hmlst .04 [.01, .06]

Residual (co)variances Sub Use�it0��it12 87.6 [84.9, 90.3] PTSD�it0��it12 1.76 [1.68, 1.83] Hmls�it0��it12 34.9 [33.9, 35.9]

Fit statistics �2LL 1,096,177.8 RMSEA .03 SRMR .02 CFI .96

Note. CI � confidence interval; t � baseline; 3m � 3 months; 6m � 6 months; 12m � 12 months; ε � residual variance measured from baseline to 12 months; �2LL � �2 log likelihood; RMSEA � root mean square error of approximation; SRMR � standardized root mean square residual; CFI � comparative fit index. Boldface indicates confidence intervals that do not include 0.RMSEA �.05 are considered to be representative of good model fit; SRMR �.08 are considered to be representative of good model fit; CFI �.90 are indicative of good model fit.T

hi s

do cu

m en

t is

co py

ri gh

te d

by th

e A

m er

ic an

P sy

ch ol

og ic

al A

ss oc

ia ti

on or

on e

of it

s al

li ed

pu bl

is he

rs .

T hi

s ar

ti cl

e is

in te

nd ed

so le

ly fo

r th

e pe

rs on

al us

e of

th e

in di

vi du

al us

er an

d is

no t

to be

di ss

em in

at ed

br oa

dl y.

7PTSD, HOMELESSNESS, AND SUBSTANCE USE IN ADOLESCENTS

high comorbidity among PTSD and substance use disorders (Clark, Lesnick, & Hegedus, 1997; Cottler et al., 1992; Driessen et al., 2008). In an effort to manage internalizing symptomology (e.g., PTSD), youth may increase their use of substances, espe- cially if the symptomology (in this case PSTD) is not addressed. For example, theoretical and empirical evidence notes a functional relationship between substance use and trauma symptomology, where substances are used in response to symptoms of internaliz- ing symptomology (Turner, Mota, Bolton, & Sareen, 2018). Re- cent research supports this notion such that, weekly fluctuations in PTSD symptomology is associated with increased dependence symp- toms (e.g., alcohol, cocaine, opiate), but not vice versa (Ouimette, Read, Wade, & Tirone, 2010). Others have found that, among a sample of adolescents in high school, the presence of PTSD sympto- mology prospectively predicted onset of substance use disorders, but again, the reverse association was not found (Wolitzky-Taylor, Bobova, Zinbarg, Mineka, & Craske, 2012). Thus, it may be that when chronic strain (e.g., experiences of homelessness and/or trauma symptoms) produces internalizing psychopathology (e.g., PTSD), this operates as an additional strain that compounds maladaptive function- ing, eventually leading to increased homelessness experiences.

Our results have implications for adolescents entering substance use treatment, especially for those with a history of homelessness. That is, we found partial support for an experience-driven risk model such that reporting more days of homelessness at baseline was a catalyst to multiple pathways of increased PTSD symptom- ology, return to substance use, and greater homelessness experi- ences posttreatment. This indicates that, as practitioners, it may be important to focus first on youth’s homelessness experience. Spe- cifically, during the treatment intake process, screening for youths’ experience of homelessness may inform providers on immediate actions to take, such as receiving housing services, reunification, or a further investigation into why youth are leaving home. In fact, our results point to homelessness as the driving factor that predicts increases in posttreatment PTSD symptoms. That is, it appears that when youth enter treatment and report heightened experiences of homelessness, it begins a cycle of problematic psychopathology that eventually lead youth back to homelessness.

In summary, our results point to a desperate need to include trauma-informed interventions for youth who have experienced early trauma and are reporting PTSD symptomology. This is especially important in the context of youth’s prior experience of homelessness, as our results point to a cascade of events leading to increased risk of homelessness stemming from trauma symptom- ology immediately posttreatment. Prior research supports this no- tion with several studies noting, among those in treatment for substance use disorders, improvements in PTSD symptomology are associated with sustained and continued improvements in substance use severity, but improvements in substance use were not found to improve PTSD symptomology (Back, Brady, Sonne, & Verduin, 2006; Hien et al., 2010). Further, in a recent review on the treatment of comorbid PTSD and substance use disorders, Roberts, Roberts, Jones, and Bisson (2016) found that trauma- focused interventions that were delivered in conjunction with substance use disorder interventions were successful in reducing both PTSD symptomology and frequency of substance use. Thus, long-term experiences of homelessness and returning to substance use following treatment completion may be mitigated by address-

ing trauma symptomology among adolescents entering treatment for substance use problems.

Limitations and Conclusion

As with any study, there are limitations to the current article. First, the pathways elucidated here may not be generalizable to other groups of youth with experiences of homelessness and are not causal. This sample is drawn from youth in substance use treatment and the pathways linking homelessness, substance use, and traumatic stress may be different in populations drawn from street-based samples or samples of foster youth. Second, much prior work is quite concerned with the impact of peer influence in these processes and the current data set does not allow us to rigorously examine peer influence. Third, substance use was as- sessed via self-report, which may bias estimates. Fourth, we ac- knowledge that some bias may have been introduced due to attrition from specific groups of participants (i.e., male, racial/ ethnic minorities, and those with prior lifetime experiences of homelessness). Finally, the majority of the current sample did not experience homelessness in the time between their intake and follow-up assessment(s). Strategies for modeling an outcome vari- able with excess zeros such as this have yet to be developed for more advanced models, and therefore our ability to accurately model this rather uncommon outcome may have been limited. However, we feel that the novelty and utility of this analytic strategy outweighs its limitations.

Future research would benefit from conducting similar within- person analyses with different populations of youth experiencing or at high risk for experiencing homelessness, such as street-based samples, samples of youth exiting juvenile justice programs, or samples drawn from the foster care system. Second, data sets that more rigorously track peer influence would add to a more com- plete understanding of the within-person connections over time, as prior work has hypothesized that substance use and mental health symptoms are impacted by longitudinal differences in peer en- gagement (Rice et al., 2005; Tyler et al., 2008; Whitbeck et al., 1999; Yoshioka-Maxwell & Rice, 2017). Third, it would be useful to examine a richer set of resilience factors as the risk amplifica- tion and abatement framework has suggested that successful en- gagement with positive peers, family, social services, and formal institutions can all impact the causal connections between home- lessness experiences and behavioral health outcomes (Milburn et al., 2009; Rice, Stein, & Milburn, 2008). In conclusion, the current study extends primarily event-driven theoretical work, to consider individual level processes in conjunction with overarching event level predictors of homelessness. We found that PTSD symptom- ology is a driving factor that both directly and indirectly influences experiences of homelessness posttreatment. Interventions may wish to incorporate trauma informed approaches for youth entering treatment as this may mitigate long-term experiences of homeless- ness and return to substance use.

References

Agoston, A. M., & Rudolph, K. D. (2013). Pathways from depressive symptoms to low social status. Journal of Abnormal Child Psychology, 41, 295–308. http://dx.doi.org/10.1007/s10802-012-9675-y

Back, S. E., Brady, K. T., Sonne, S. C., & Verduin, M. L. (2006). Symptom improvement in co-occurring PTSD and alcohol dependence. Journal of

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

8 DAVIS ET AL.

Nervous and Mental Disease, 194, 690 – 696. http://dx.doi.org/10.1097/ 01.nmd.0000235794.12794.8a

Benda, B. B. (2006). Survival analyses of social support and trauma among homeless male and female veterans who abuse substances. American Journal of Orthopsychiatry, 76, 70 –79. http://dx.doi.org/10.1037/0002- 9432.76.1.70

Bender, K., Brown, S. M., Thompson, S. J., Ferguson, K. M., & Langen- derfer, L. (2015). Multiple victimizations before and after leaving home associated with PTSD, depression, and substance use disorder among homeless youth. Child Maltreatment, 20, 115–124. http://dx.doi.org/10 .1177/1077559514562859

Bender, K., Ferguson, K., Thompson, S., Komlo, C., & Pollio, D. (2010). Factors associated with trauma and posttraumatic stress disorder among homeless youth in three U.S. cities: The importance of transience. Journal of Traumatic Stress, 23, 161–168. http://dx.doi.org/10.1002/jts .20501

Berry, D., & Willoughby, M. T. (2017). On the practical interpretability of cross-lagged panel models: rethinking a developmental workhorse. Child Development, 88, 1186 –1206. http://dx.doi.org/10.1111/cdev.12660

Berry, R., & Sellman, J. D. (2001). Childhood adversity in alcohol- and drug-dependent women presenting to out-patient treatment. Drug and Alcohol Review, 20, 361–367.

Burns, M. N., Lehman, K. A., Milby, J. B., Wallace, D., & Schumacher, J. E. (2010). Do PTSD symptoms and course predict continued sub- stance use for homeless individuals in contingency management for cocaine dependence? Behaviour Research and Therapy, 48, 588 –598. http://dx.doi.org/10.1016/j.brat.2010.03.010

Clark, D. B., Lesnick, L., & Hegedus, A. M. (1997). Traumas and other adverse life events in adolescents with alcohol abuse and dependence. Journal of the American Academy of Child & Adolescent Psychiatry, 36, 1744 –1751. http://dx.doi.org/10.1097/00004583-199712000-00023

Cottler, L. B., Compton, W. M., III, Mager, D., Spitznagel, E. L., & Janca, A. (1992). Posttraumatic stress disorder among substance users from the general population. The American Journal of Psychiatry, 149, 664 – 670. http://dx.doi.org/10.1176/ajp.149.5.664

Curran, P. J., Howard, A. L., Bainter, S. A., Lane, S. T., & McGinley, J. S. (2014). The separation of between-person and within-person compo- nents of individual change over time: A latent curve model with struc- tured residuals. Journal of Consulting and Clinical Psychology, 82, 879 – 894. http://dx.doi.org/10.1037/a0035297

Davies, B. R., & Allen, N. B. (2017). Trauma and homelessness in youth: Psychopathology and intervention. Clinical Psychology Review, 54, 17–28. http://dx.doi.org/10.1016/j.cpr.2017.03.005

Davis, J. P., Dumas, T. M., Merrin, G. J., Espelage, D. L., Tan, K., Madden, D., & Hong, J. S. (2018). Examining the pathways between bully victimization, depression, academic achievement, and problematic drinking in adolescence. Psychology of Addictive Behaviors, 32, 605– 616. http://dx.doi.org/10.1037/adb0000394

Dennis, M. L., Chan, Y. F., & Funk, R. R. (2006). Development and validation of the GAIN Short Screener (GSS) for internalizing, exter- nalizing and substance use disorders and crime/violence problems among adolescents and adults. The American Journal on Addictions, 15, 80 –91. http://dx.doi.org/10.1080/10550490601006055

Dennis, M. L., Titus, J. C., White, M., Unsicker, J., & Hodgkins, D. (2003). Global Appraisal of Individual Needs (GAIN): Administration guide for the GAIN and related measures. Bloomington, IL: Chestnut Health Systems.

Driessen, M., Schulte, S., Luedecke, C., Schaefer, I., Sutmann, F., Ohlmeier, M., . . . the TRAUMAB-Study Group. (2008). Trauma and PTSD in patients with alcohol, drug, or dual dependence: A multi-center study. Alcoholism, Clinical and Experimental Research, 32, 481– 488. http://dx.doi.org/10.1111/j.1530-0277.2007.00591.x

Enders, C. K. (2011). Analyzing longitudinal data with missing values. Rehabilitation Psychology, 56, 267–288. http://dx.doi.org/10.1037/ a0025579

Folsom, D., Hawthorne, W., Lindamer, L., Gilmer, G., Bailey, A., Gol- shan, S., . . . Jeste, D. V. (2005). Prevalence and risk factors for homelessness and utilization of mental health services among 10,340 patients with serious mental illness in a large public mental health system. The American Journal of Psychiatry, 162, 370 –376.

Hien, D. A., Jiang, H., Campbell, A. N. C., Hu, M.-C., Miele, G. M., Cohen, L. R., . . . Nunes, E. V. (2010). Do treatment improvements in PTSD severity affect substance use outcomes? A secondary analysis from a randomized clinical trial in NIDA’s clinical trials network. The American Journal of Psychiatry, 167, 95–101. http://dx.doi.org/10.1176/ appi.ajp.2009.09091261

Hodgson, K. J., Shelton, K. H., van den Bree, M. B., & Los, F. J. (2013). Psychopathology in young people experiencing homelessness: A sys- tematic review. American Journal of Public Health, 103, e24 – e37. http://dx.doi.org/10.2105/AJPH.2013.301318

Hsu, H. T., Fulginiti, A., Rice, E., Rhoades, H., Winetrobe, H., & Danforth, L. (2018). But everyone is doing it (sort of)! Perceived sexual risks in the social environment and the impact on homeless youth engagement in concurrent sexual relationships. AIDS and Behavior, 22, 3508 –3518. http://dx.doi.org/10.1007/s10461-018-2133-2

Lewinsohn, P. M., Steinmetz, J. L., Larson, D. W., & Franklin, J. (1981). Depression-related cognitions: Antecedent or consequence? Journal of Abnormal Psychology, 90, 213–219. http://dx.doi.org/10.1037/0021- 843X.90.3.213

Martijn, C., & Sharpe, L. (2006). Pathways to youth homelessness. Social Science & Medicine, 62, 1–12. http://dx.doi.org/10.1016/j.socscimed .2005.05.007

McManus, H. H., & Thompson, S. J. (2008). Trauma among unaccompa- nied homeless youth: The integration of street culture into a model of intervention. Journal of Aggression, Maltreatment & Trauma, 16, 92– 109. http://dx.doi.org/10.1080/10926770801920818

Merrin, G. J., Davis, J. P., Berry, D., D’Amico, E. J., & Dumas, T. M. (2016). The longitudinal associations between substance use, crime, and social risk among emerging adults: A longitudinal within and between- person latent variables analysis. Drug and Alcohol Dependence, 165, 71–78. http://dx.doi.org/10.1016/j.drugalcdep.2016.05.009

Milburn, N. G., Rice, E., Rotheram-Borus, M. J., Mallett, S., Rosenthal, D., Batterham, P., . . . Duan, N. (2009). Adolescents exiting homelessness over two years: The risk amplification and abatement model. Journal of Research on Adolescence, 19, 762–785. http://dx.doi.org/10.1111/j .1532-7795.2009.00610.x

Milburn, N. G., Stein, J. A., Lopez, S. A., Hilberg, A. M., Veprinsky, A., Arnold, E. M., . . . Comulada, W. S. (2019). Trauma, family factors and the mental health of homeless adolescents. Journal of Child & Adoles- cent Trauma, 12, 37– 47. http://dx.doi.org/10.1007/s40653-017-0157-9

Muthén, L. K., & Muthén, B. O. (1998 –2017). Mplus user’s guide (8th ed.). Los Angeles, CA: Author.

North, C. S., & Smith, E. M. (1992). Posttraumatic stress disorder among homeless men and women. Psychiatric Services, 43, 1010 –1016. http:// dx.doi.org/10.1176/ps.43.10.1010

Ouimette, P., Read, J. P., Wade, M., & Tirone, V. (2010). Modeling associations between posttraumatic stress symptoms and substance use. Addictive Behaviors, 35, 64 – 67. http://dx.doi.org/10.1016/j.addbeh .2009.08.009

Radu, M. B. (2017). Who runs away from home and why? How families, schools, and bullying influence youth runaways. Sociology Compass, 11, e12537. http://dx.doi.org/10.1111/soc4.12537

Reynolds, M., Hinchliffe, K., Asamoah, V., & Kouimtsidis, C. (2011). Trauma and post-traumatic stress disorder in a drug treatment commu- nity service. The Psychiatrist, 35, 256 –260. http://dx.doi.org/10.1192/ pb.bp.110.030379

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

9PTSD, HOMELESSNESS, AND SUBSTANCE USE IN ADOLESCENTS

Rice, E., Milburn, N. G., Rotheram-Borus, M. J., Mallett, S., & Rosenthal, D. (2005). The effects of peer group network properties on drug use among homeless youth. American Behavioral Scientist, 48, 1102–1123. http://dx.doi.org/10.1177/0002764204274194

Rice, E., Stein, J. A., & Milburn, N. (2008). Countervailing social network influences on problem behaviors among homeless youth. Journal of Adolescence, 31, 625– 639. http://dx.doi.org/10.1016/j.adolescence.2007 .10.008

Roberts, N. P., Roberts, P. A., Jones, N., & Bisson, J. I. (2016). Psycho- logical therapies for post-traumatic stress disorder and comorbid sub- stance use disorder. Cochrane Database of Systematic Reviews, 4, CD010204. http://dx.doi.org/10.1002/14651858.CD010204.pub2

Schäfer, I., Langeland, W., Hissbach, J., Luedecke, C., Ohlmeier, M. D., Chodzinski, C., . . . the TRAUMAB-Study group. (2010). Childhood trauma and dissociation in patients with alcohol dependence, drug de- pendence, or both-A multi-center study. Drug and Alcohol Dependence, 109, 84 – 89. http://dx.doi.org/10.1016/j.drugalcdep.2009.12.012

Simpson, T. L., & Miller, W. R. (2002). Concomitance between childhood sexual and physical abuse and substance use problems: A review. Clinical Psychology Review, 22, 27–77. http://dx.doi.org/10.1016/ S0272-7358(00)00088-X

Slesnick, N., Bartle-Haring, S., & Gangamma, R. (2006). Predictors of substance use and family therapy outcome among physically and sexu- ally abused runaway adolescents. Journal of Marital and Family Ther- apy, 32, 261–281. http://dx.doi.org/10.1111/j.1752-0606.2006 .tb01606.x

Slesnick, N., Guo, X., Brakenhoff, B., & Feng, X. (2013). Two-year predictors of runaway and homeless episodes following shelter services among substance abusing adolescents. Journal of Adolescence, 36, 787– 795. http://dx.doi.org/10.1016/j.adolescence.2013.06.007

Substance Abuse and Mental Health Services Administration. (2018). Key substance use and mental health indicators in the United States: Results from the 2017 National Survey on Drug Use and Health (HHS Publi- cation No. SMA 18 –5068, NSDUH Series H-53). Rockville, MD: Center for Behavioral Health Statistics and Quality, Substance Abuse and Mental Health Services Administration. Retrieved from https:// www.samhsa.gov/data/

Tucker, J. S., Edelen, M. O., Ellickson, P. L., & Klein, D. J. (2011). Running away from home: A longitudinal study of adolescent risk factors and young adult outcomes. Journal of Youth and Adolescence, 40, 507–518. http://dx.doi.org/10.1007/s10964-010-9571-0

Turner, S., Mota, N., Bolton, J., & Sareen, J. (2018). Self-medication with alcohol or drugs for mood and anxiety disorders: A narrative review of the epidemiological literature. Depression and Anxiety, 35, 851– 860. http://dx.doi.org/10.1002/da.22771

Tyler, K. A., & Bersani, B. E. (2008). A longitudinal study of early adolescent precursors to running away. The Journal of Early Adoles- cence, 28, 230 –251. http://dx.doi.org/10.1177/0272431607313592

Tyler, K. A., Johnson, K. A., & Brownridge, D. A. (2008). A longitudinal study of the effects of child maltreatment on later outcomes among

high-risk adolescents. Journal of Youth and Adolescence, 37, 506 –521. http://dx.doi.org/10.1007/s10964-007-9250-y

Tyler, K. A., & Schmitz, R. M. (2018). Childhood disadvantage, social and psychological stress, and substance use among homeless youth: A life stress framework. Youth & Society. Advance online publication. http:// dx.doi.org/10.1177/0044118X18767032

van den Bree, M. B. M., Shelton, K., Bonner, A., Moss, S., Thomas, H., & Taylor, P. J. (2009). A longitudinal population-based study of factors in adolescence predicting homelessness in young adulthood. The Journal of Adolescent Health, 45, 571–578. http://dx.doi.org/10.1016/j .jadohealth.2009.03.027

Vreven, D. L., Gudanowski, D. M., King, L. A., & King, D. W. (1995). The civilian version of the Mississippi PTSD Scale: A psychometric evaluation. Journal of Traumatic Stress, 8, 91–109. http://dx.doi.org/10 .1002/jts.2490080107

Whitbeck, L. B., Hoyt, D. R., & Ackley, K. A. (1997). Abusive family backgrounds and later victimization among runaway and homeless ad- olescents. Journal of Research on Adolescence, 7, 375–392. http://dx .doi.org/10.1207/s15327795jra0704_2

Whitbeck, L. B., Hoyt, D. R., & Bao, W. N. (2000). Depressive symptoms and co-occurring depressive symptoms, substance abuse, and conduct problems among runaway and homeless adolescents. Child Develop- ment, 71, 721–732. http://dx.doi.org/10.1111/1467-8624.00181

Whitbeck, L. B., Hoyt, D. R., Johnson, K. D., & Chen, X. (2007). Victimization and posttraumatic stress disorder among runaway and homeless adolescents. Violence and Victims, 22, 721–734. http://dx.doi .org/10.1891/088667007782793165

Whitbeck, L. B., Hoyt, D. R., & Yoder, K. A. (1999). A risk-amplification model of victimization and depressive symptoms among runaway and homeless adolescents. American Journal of Community Psychology, 27, 273–296. http://dx.doi.org/10.1023/A:1022891802943

Wolitzky-Taylor, K., Bobova, L., Zinbarg, R. E., Mineka, S., & Craske, M. G. (2012). Longitudinal investigation of the impact of anxiety and mood disorders in adolescence on subsequent substance use disorder onset and vice versa. Addictive Behaviors, 37, 982–985. http://dx.doi .org/10.1016/j.addbeh.2012.03.026

Yeater, E. A., Austin, J. L., Green, M. J., & Smith, J. E. (2010). Coping mediates the relationship between posttraumatic stress disorder (PTSD) symptoms and alcohol use in homeless, ethnically diverse women: A preliminary study. Psychological Trauma: Theory, Research, Practice, and Policy, 2, 307–310. http://dx.doi.org/10.1037/a0021779

Yoshioka-Maxwell, A., & Rice, E. (2017). Exploring the impact of net- work characteristics on substance use outcomes among homeless former foster youth. International Journal of Public Health, 62, 371–378. http://dx.doi.org/10.1007/s00038-016-0845-5

Received December 4, 2018 Revision received June 6, 2019

Accepted June 6, 2019 �T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

10 DAVIS ET AL.