http://orcid.org/0000-0001-5738-7486
ABSTRACT
Background. Alcohol and drug use is associated with significant morbidity and mortality and is highly prevalent among homeless youth. Ecological Momentary Assessments (EMA) have been used to examine the effect of urges on drug use, though not among homeless youth. Objectives. We assessed the patterns of drug use and the correlation between real-time contextual factors and drug use using EMA collected daily. We identified predictors of drug use among a sample of homeless youth 18–25 years old in Houston, Texas. Methods. Homeless youth (n = 66, 62% male) were recruited from a drop-in center between September 2015 and May 2016. We used generalized linear mixed models and cross-validation methods to determine the best predictive model of drug use. Results. The overall drug use was high: 61% and 32% of participants reported using drugs or alcohol at least one day, respectively. Marijuana and synthetic marijuana use (i.e., Kush, K2, incense packs) were reported most frequently; 86% and 13% of the total drug use EMAs, respectfully. Drug use urge was reported on 26% of days and was the highest on drug use days. Drug use was predicted by discrimination, pornography use, alcohol use, and urges for drugs, alcohol, and to steal. Conclusions. EMA can be used to predict drug use among homeless youth. Drug use treatment among homeless youth should address the role of experiencing discrimination, pornography and alcohol use, and urge management strategies on drug use. Research is needed to determine if EMA informed just-in-time interventions targeting these predictors can reduce use.
Introduction
Alcohol and drug use are associated with significant morbidity and mortality (Citation1) and are highly prevalent among homeless youth. The mortality rate among homeless youth is 10 times higher than that in housed youth, with drug overdose being one of the leading causes of death (Citation2). One study found that up to 86% of homeless youth met the DSM-IV diagnostic criteria for drug dependence or abuse (Citation3). Over half of homeless youth from a drop-in center were found to have alcohol or drug dependence (61.1% and 54.6%, respectively) (Citation4). Comparatively, a national survey measuring drug use rates of 8th-, 10th-, and 12th-grade students found that lifetime prevalence rates were 33% for any illicit drug, 29% for marijuana, and 42% for alcohol (Citation5). Another national survey from the 2015 National Survey on Drug Use and Health reported past-month drug use rates that were half that of homeless youth (Citation6).
The high prevalence of drug use among homeless youth has been linked to a myriad of risk factors specific to homeless populations. For example, housing instability has been associated with higher rates of risky behaviors including drug use (Citation7–Citation9). The age at homelessness onset and the duration of homelessness are also related to increased drug use among homeless adults. The rate of drug use is higher in those who first experience homelessness before age 25 and increases as the duration of homelessness increases (Citation9,Citation10). Wenzel et al. (2010) suggested that a longer duration of homelessness provides added time to form social networks with a greater density of drug using peers (Citation11). In addition to having higher rates of drug use and drug use disorders, homeless populations may also experience added societal consequences due to their drug use. For example, among homeless individuals, those with drug use problems were shown to be less likely to achieve housing stability after two years, even when compared to homeless individuals with other physical and non-drug-related mental health disorders (Citation12). To this end, beyond the adverse health risks of drug use, drug use is a strong barrier to establishing independent housing and is associated with the age at homelessness onset and duration of homelessness (Citation12).
Although the relation between housing instability and drug use has been well established, few studies have assessed how drug use varies across sheltering types. Homeless youth commonly shift between various unstable living situations (i.e., staying with others, in a shelter, or on the streets) depending on their social network (Citation13). However, most studies exploring this association have been limited by cross-sectional designs and are based on behavior recall, which can be unreliable because drug use is highly stigmatized and contextualized within one’s environment and social network (Citation14).
In addition to the association between drug use and sheltering, a substantial body of literature supports the significant association between stress and drug use (Citation15). Homeless youth experience high levels of stress as they face ongoing struggles associated with homelessness, including lacking food and shelter, being physically and/or sexually victimized and being exposed to violence (Citation9,Citation13,Citation16). To cope with stress, they may turn to drug and alcohol use to cope with the harsh conditions of the streets and escape from significant life stressors (Citation17,Citation18). Other studies support the positive relations between various street related stressors and drug use among homeless youth (Citation19,Citation20). For example, Tyler and colleagues found that homeless youth who had experienced victimization or trade sex since becoming homeless were more likely to use drugs (Citation19).
Ecological Momentary Assessments (EMA) have been used to examine the effect of urges to use drugs on drug use. A systematic review conducted by Serre et al. (2015) found that the majority of studies (92%) reported a positive relation between drug use urges and drug use, regardless of the level of use. These findings suggest that urge is a strong predictor of use when assessed in near real-time to actual drug use and that EMA can be used to identify variations in daily drug use urges and how these urges influence drug use (Citation21). Although EMA have not been applied to assess the effect of urges on drug use among homeless youth, findings from a qualitative study demonstrate the powerful influence urges can have on drug use (Citation17). For some homeless youth, the urge to use drugs can be all consuming and can fuel addiction (Citation17).
A greater understanding of the daily circumstances and contextual factors related to drug use is needed, particularly among high risk populations such as homeless youth. EMA methods may further our understanding of contextual factors related to real-time behaviors and combat some of the limitations related to recall bias by using repeated assessments that allow the participant to provide data proximal to the time of the behavior and that can account for current contextual factors such as housing and urge (Citation14,Citation22). EMA is the most accurate way to measure real-time factors in natural settings, is considered the gold standard (Citation23,Citation24), has reached high youth compliance rates (78%) across 42 studies (Citation25), and has been used to assess drinking behaviors (Citation26) and drug use among youth. To date, no studies have used EMA to assess drug use in real-time among homeless youth or to determine the real-time predictors of drug use in this high risk population. Additionally, no studies have assessed whether real-time contextual factors can predict drug use among homeless youth. Therefore, we conducted this study to assess the predictors of drug use with particular interest in the relation between drug use, sheltering type, and drug use urge.
Methods
This study was guided by the Risk Amplification Model, which posits that risk behaviors among homeless youth are elevated by the circumstances experienced prior to and subsequent to becoming homeless (Citation27–Citation29). We aimed to assess the correlation between real-time contextual factors and drug use using EMA collected daily and to identify predictors of drug use among a sample of homeless youth 18–25 years old in Houston, TX. The findings of this study are reported according to the adapted STROBE Checklist for Reporting EMA Studies (Citation30).
Participant recruitment
Homeless youth were recruited from the largest homeless youth drop-in center in Houston, TX between Sept, 2015 and May, 2016. Recruitment was conducted at a drop-in center vs. a shelter to oversample unsheltered youth due to the potential shelter related restrictions on drug use. Respondents were provided with the study details and interested individuals were screened for eligibility. Participants were included if they were 18–25 years old, spoke English, and were homeless. Homelessness was defined as sleeping on the streets, in a place not meant for human habitation, in a shelter, in a hotel/motel, or with someone where they could not stay for more than 30 days (i.e., couch surfing). Participants were excluded if they had very low literacy, given the need to read and respond to EMA unassisted as part of the data collection protocols of the study (Citation31). The literacy cutoff was a score of less than 4, a 6th grade reading level, on the Rapid Estimate of Adult Literacy in Medicine-Short Form health literacy assessment (Citation31,Citation32) an interviewer-administered list of words that individuals are asked to pronounce.
Study procedures
Using an Institutional Review Board-approved informed consent document (*BLINDED*# HSC-SON-15–0133), youth provided written consent in the presence of study staff and completed an extensive participant tracking form to assist with follow-up. After the initial eligibility screening, participants completed an audio-assisted survey on an iPad using Qualtrics. Of the screened youth, 2 were excluded based on low literacy. Additionally, 5 youth who approached the study staff to learn more about the study decided not to participate, 2 indicated they would not be in town for the 1 month study period and 3 indicated that they were not interested in participating. The baseline survey took approximately 30 minutes to complete and assessed gender identity, age, race/ethnicity, sexual orientation, educational attainment, and adverse childhood experiences. Participants were then provided with a study-issued smartphone and instructions for completing the EMA.
Study staff contacted participants on their study phone to schedule the final study visit 21 days after the initial study visit. Participants met with staff at drop-in centers, shelters, local libraries, or restaurants to complete their exit survey, return the study smartphone, and receive their grocery store gift cards for participation. Study participants received a $20 gift card for completing the initial baseline survey. Upon return of the study smartphone at the final visit, participants received up to $95 in gift cards. The amount of compensation for completing EMAs depended on the percentage of random and daily EMAs completed. Specifically, participants who completed 50%-75% of EMAs received a $50 gift card, those who completed 76%-89% of EMAs received a $75 gift card, and those who completed 90% or more EMAs received a $95 gift card. Those who completed < 50% of EMAs received a $20 gift card for returning the phone. This incentive structure was explained to all participants during the informed consent process. Participants could access their current compensation level throughout the study period on the password protected, study-issued, smartphone interface.
EMA measures schedule
The EMA methodology used is similar to that developed by Shiffman and colleagues (Citation14,Citation33,Citation34) and has been used by our study team on multiple studies (Citation35–Citation37). Daily and random EMA were prompted by study smartphones 5 times a day during the participant’s indicated normal waking hours.
Daily sampling
Daily assessments assessed risk behaviors and were prompted once every day 30 minutes after the participant’s indicated normal waking time. Questions referred to the previous 24 hours and captured near real-time risk behaviors. Items assessed in the daily EMA included work and school days, sheltering, discrimination encountered, pornography viewing, sexual behaviors, drug use, and alcohol consumption with items such as “Did you use drugs yesterday?” and “Did you drink any alcoholic beverages yesterday?” On average, daily diary assessments took less than 5 minutes to complete. Affect questions in the daily EMA asked about current participant-rated affect.
Random sampling
Random assessments assessed real-time affect and were prompted four times a day, occurring randomly in 4 epochs during each participant’s normal waking hours, and took approximately 2 minutes to complete. The phone audibly and visually cued each random assessment for 30 seconds. If the participant did not respond after three prompts, the assessment was recorded as missed. Random assessments were used to assess participant- rated affect by indicating the extent to which they felt upset, angry, guilty, scared, irritable, ashamed, restless, nervous, afraid, stressed, depressed, sad, and bored. Urge was assessed by asking if the participant felt the urge to drink alcohol, use drugs, have sex, or steal something. Items were selected based upon their hypothesized relation to drug use and measured real-time stress, risk behaviors, and urges (i.e., right now) and many of the items used a Likert-like scale (e.g., 1–5 scale with the following anchors: 1 = strongly disagree, 3 = neutral, 5 = strongly agree) based on our previous work with homeless youth and that of Shiffman et al (Citation14,Citation33,Citation34).
Technology and hardware
We used Samsung Galaxy Light smartphones with the Android 4.2 operating system. Participants navigated the EMA program and entered data by touching the screen. Participants could call and receive calls and had access to the Internet and text messaging throughout the study. Prior to launching the study, 10 youth participated in beta testing of the app.
Data management
EMA data were automatically uploaded from the study phone to a secure server daily. If the phone was off or the battery was dead, the data were uploaded once the phone was turned on or recharged. No data were accessible from the study phone except for an interface that displayed the participant’s current compensation level based on the percentage of completed EMAs. If a phone was lost or stolen, a new phone was issued, the phone was remotely wiped and returned to the manufacturer settings, and the cellular plan was discontinued. Approximately 30% of phones were damaged, lost, or stolen during the study largely due to the tumultuous lived experience of homelessness.
Data analysis
The descriptive analysis was generated from the participants and the EMA. For fixed indicators such as age, gender, race, ethnicity, and sexual orientation, we counted the number of participants who used alcohol, marijuana, and/or other drugs. For the EMA description, we counted the surveys and calculated the percentages of total valid observations across the following categories: work days and school days, sheltering status, stress level, discrimination, and pornography use. The EMA compliance rates were calculated from the average daily EMA compliance rate per person and the random EMA compliance rate per participant over the 21 possible daily EMA and 84 possible random EMA.
Generalized linear mixed models (GLMM) were used to model the longitudinal data, with a logistic link used for adaptation to the binary outcome of drug use. The random intercept effect was used to account for correlated observations within subject. The identified predictors of drug use risk included EMA data that preceded the case behavior and assessed knowledge, intentions, social support, real-time drug use urge, alcohol use, and stress up to 24 hours prior to the drug use occurrence. For longitudinal modeling of drug use, mean values of random assessments were calculated on a daily basis to match the collection frequency of the drug use variable from the matched daily EMA. Daily and random assessments of affect, drug use, and alcohol use were combined into a single daily summary of each variable and matched with the same day daily EMA for efficient use of data while minimizing shared variance. Time-invariant predictors such as demographic data were also included in the model exploration and Akaike information criterion (AIC) (Citation38) was used to systematically eliminate variables with backward selection. The optimal parsimonious model provided the risk estimator for the drug use outcome. Model performance was assessed with receiver operating characteristic (ROC) curve and with cross-validated sensitivity and specificity. Cross-validation was accomplished over the course of 100 runs with randomly generated 80/20 splits into training and test sets (Citation39). For measuring sensitivity and specificity, the model estimates of probabilities of drug use events can be converted to binary predictions of drug use events (yes/no) with the choice of a decision threshold, e.g. drug use event is predicted when probability of drug use exceeds 0.5.
Results
Demographic characteristics
The mean age of the participants was 21 years (range: 18–25; ). Most participants were male, minority (65% African American, 12% Hispanic/Latino), and heterosexual. The median age at the onset of homelessness was 16.9 years. On the majority of days, participants were unstably housed (47%) or unsheltered (34%). A large proportion of the sample (35%) had a history of foster care.
Table 1. Participant demographics and drug use behaviors.
EMA response rates
We collected EMA data on 66 of 74 recruited participants indicating an 89% participation rate (). The remaining 8 participants did not provide sufficient EMA data to be included in the analysis and therefore were dropped. The mean number of total EMA observations (daily and random) completed by each participant was 45 (i.e., 13 daily and 37 random EMA). The average daily survey compliance rate was 62%. The compliance rates of daily EMAs among all demographic groups (e.g., race, gender, age, sex orientation) are higher than the compliance rates of random EMA. Participants who identified as other race, female, and older than 20 years had higher total EMA response rates. Participants who completed the fewest daily EMAs (i.e., <25 percentile) were more likely to report their race as “other”. Youth with the highest 25 percentile of daily EMA were similar across all demographic characteristics as the total sample. No significant demographic differences were found among youth who did not contribute EMA and those who did.
EMA descriptive data
In total, participants indicated that they worked (6.3%) or went to school (6.0%) on very few study days (). Perceived stress scores were higher on drug use days than the average score across all days. Participants also reported more discrimination experiences and pornography use on alcohol and drug use days. Drug use urge was reported on 26% of all days and was highest on drug use days.
Drug use patterns
Overall drug use was high: 61% and 32% of participants reported using drugs or alcohol on at least one day, respectively. Participants used drugs on 26.6% and alcohol on 5.1% of the 804 total sample study days included in the analysis. Days of drug use per participant ranged from 0 to 18 days of 21 possible days. Male, Black, and heterosexual youth reported higher rates of both alcohol and drug use than females, White/Other, and LGBT youth. Both alcohol and drug use were higher on days when participants also viewed pornography. Different classes of drug use were collected and reported among 214 valid drug use surveys, and some participants used several different drugs in one day (). Of the total positive drug use EMAs, 85.5% reported marijuana use, 12.6% reported synthetic marijuana use (i.e., Kush, K2, synthetic cannabinoids, incense packs), and 8.9% reported using bath salts. Of the 66 total participants, 40 reported at least one day of drug use during the EMA period. Among the 40 participants who reported drug use, 36 of them used Marijuana, with a range of use from 1–18 days out of 21 total possible days and an average of 5 days of use.
Table 2. Drug use frequency.
Table 3. EMA compliance data.
Table 4. GLMM coefficients and odds ratios for predictors of drug use.
Drug use urge was high among participants: 40 of the 66 youth reported at least one occurrence of high drug use urge among that 2183 random EMA included in the analysis. Drug use urge was highly correlated with drug use. Of the 40 participants who experienced drug use urge, 70% also reported using drugs. Of the 207 EMA that reported high drug use urges, 43% of these urges occurred on days that drug use occurred.
Drug use by shelter type
Drug and alcohol use rates were higher on unsheltered and unstably housed days (data not shown). Using chi-squared tests, we found a significant difference (p < 0.01) in reported daily drug use by current shelter status. The rates of drug use were highest when participants stayed with a friend or acquaintance (35.6%), on the streets (34.4%), or in the home of a romantic or sexual partner (33.7%). The rate of drug use was lowest on days participants were sheltered (7.8%). We also found a significant difference (p < 0.01) in daily alcohol use by shelter status. Days when youth stayed in a car had the highest rates of alcohol use (11.1%), followed by days staying in a hotel or motel (10.7%), and on the streets (9.4%). Shelter days had the lowest rates of alcohol use (1.9%), followed by days staying with relatives (2.8%). However, when shelter residence status was parsimoniously added to the GLMMs, it no longer predicted drug use.
Predictors of drug use
Odds ratios (OR) were estimated for each of the predictors that were considered for inclusion in the risk estimator for drug use (). Race, gender, and age (i.e., characteristics associated with compliance) were not sufficiently associated with the outcomes. The parsimonious GLMMs that minimized AIC had urge to steal (OR = 56.1, p < .001), urge to use drugs (OR = 47.2, p < .001), urge to use alcohol (OR = 5.2, p = .058), along with pornography use (OR = 6.4, p < .001), alcohol use (OR = 4.6, p = .001), and experience of discrimination (OR = 3.8, p = .029) as predictors. Participants with insufficient data for analysis were dropped from the GLMM. The risk estimator performed well, as indicated by the value of 0.82 for the area under the ROC curve (Citation40). The model predicts the probability of drug use on a given day. For a decision threshold of p = 0.5 for making binary predictions of drug use events, the mean cross-validated sensitivity and specificity were 0.552 and 0.905, respectively, and the likelihood ratio was 5.8 for the true positive rate compared to the false positive rate. When the decision threshold was lowered to p = 0.33 for making binary predictions of drug use, the mean cross-validated sensitivity of the risk estimator increased to 0.668 and the specificity decreased to 0.835.
Discussion
The findings presented here represent a predominantly unsheltered and unstably housed sample of homeless youth aged 18–25 years old with high incidences of drug use. Drug use was much more prevalent than alcohol use in this population. EMA data indicated that marijuana was the primary class of drug used. We found that nearly one-fifth of the participants used synthetic marijuana while in the study, indicating an emerging drug use pattern of major public health concern with significant health implications (Citation41–Citation43). Further studies are needed to determine who is at greatest risk of using synthetic marijuana to inform prevention intervention development.
Using EMA data, we determined real-time risk factors that predict drug use including urge to use drugs, alcohol, and to steal, pornography use, alcohol use, and experiencing discrimination. The findings of this study suggest that the EMA approach is an effective way to capture variance in real-time factors that influence risk behaviors such as drug use and can be used to predict drug use based on these patterns. Urge was the strongest predictor of drug use. It is possible that the urge to steal is fueled by the desire for money or items to trade for drugs. It may also be a function of how desperate the youth’s situation has become with the scarcity of resources driving the urge to steal and the urge to escape or cope by using drugs. However, further research is needed to deepen our understanding regarding the urge to steal and perceived reasons for drug use. In particular, while perceived stress scores were higher on drug use days, stress was not found to predict drug use in the final model. While alcohol use was relatively low, it was an important contributing factor to drug use risk. It may be that youth who use drugs do so in conjunction with alcohol use.
Other behaviors were also found that clustered with drug use including pornography use and experiences of discrimination. Drug use was higher on days when youth viewed pornography or experienced discrimination. Because pornography, discrimination, and drug use measures were asked in relation to yesterday, we cannot determine from this data which preceded or followed drug use. Only 18% of the sample viewed pornography during the study. Yet, this significantly increased the odds of also reporting drug use. Additionally, discrimination occurred at a relatively low rate, 65 cases (8% of days), but also significantly increased the odds of drug use. Further studies are needed to address the temporality issue and tease out the relation between drug use, pornography use, and experiences of discrimination.
An interesting finding was that alcohol and drug use were higher on the days homeless youth were unsheltered (e.g., staying on the street, in an abandoned apartment, or in a car) and unstably housed (e.g., staying at the home of their friend or sexual partner) in comparison to the days they were sheltered. This finding may be an indication of the increased stressors homeless youth face when they are unsheltered. Previous studies revealed homeless youth often turn to drugs to cope with the daily experiences of living on the streets and the struggle to survive (Citation17,Citation18). In light of the Risk Amplification Model, these data may also suggest that homeless youth networks with high-risk peers influence the accessibility to and norming of drug use. This may also be associated with shelter policies to reduce drug use such as early curfews and the forbiddance of drug use on the premise. Future studies are needed to further untangle the impact of shelter policies and housing-related stress on drug use to determine systems-level intervention targets.
Significance of the findings
To our knowledge, this is the first study to use EMA to predict the likelihood of drug use among homeless youth. We found high rates of drug use and modifiable predictors, highlighting the potential to develop just-in-time personalized interventions that could utilize smartphone technologies. As well, we had a high participation rate and reasonable compliance among this high-risk population, indicating the feasibility and acceptability of using smartphone-based EMA. These interventions could target the modifiable predictors of drug use when they occur to disrupt the progression to use by identifying skills to manage drug use urges, and curb alcohol use. Evidence suggests that “urge surfing,” a mindful approach to managing urges by observing the craving without over-reacting to it (Citation44) may help reduce drug use. EMA could improve self-monitoring of cravings and could promote ‘urge surfing’ strategies by identifying drug use urges and prompting behavioral motivation communications to facilitate urge surfing (Citation45).
Limitations
This study did not include all possible real-time predictors of drug use. We constructed a comprehensive EMA survey that included variables indicated in the literature to affect drug use as well as real-time cognitions and behaviors that are theoretically related. That said, using EMA methods to predict risk behaviors in real-time is a relatively novel scientific method that does not have defined guidelines for best practices for implementation or analysis, particularly among vulnerable populations such as homeless youth. EMA are based on self-report of illicit and stigmatized behaviors, which may be subject to social desirability bias. In so far as the EMA approach is an emerging science, the measures used to assess drug use and cognitions in real-time have not yet been psychometrically validated. Additionally, the average daily survey compliance rate was lower than found among other populations. Several participants disclosed to the study staff that they had difficulties in maintaining a charged battery on their phones related to being unhoused and having intermittent access to electricity. Other issues that affected compliance include shelter restrictions on cellphone use and work schedules that did not permit access to one’s phone. This would have affected the delivery of and ability to complete EMAs when prompted. Future studies should assess the actual receipt of EMAs vs. using the schedule of EMA delivery as the denominator for EMA compliance among homeless youth who struggle to access electricity and experience restrictions to phone access. Caution should be used in interpreting the findings of this study as this was a single site study and thus the results might not generalize to other locations and venues. Multi-site EMA studies are needed to enhance generalizability of the findings. Finally, participants were incentivized to participate in this study. While 81% said they would enroll in a similar EMA study again, it is possible that youth would be less engaged with an intervention that utilizes EMA to deliver just-in-time messaging if they are not incentivized. Further research is needed to determine how to make EMA a rewarding experience and minimize participant burden to facilitate equitable compliance.
Conclusions
EMA data that assesses real-time factors can be used to predict drug use among homeless youth and inform the design of just-in-time, adaptive, smartphone delivered interventions.
Disclosure of Interest
The authors report no conflict of interest
Acknowledgments
The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The authors would like to extent our gratitude to the youth who participated in this study.
Additional information
Funding
References
- Whiteford HA, Degenhardt L, Rehm J, Baxter AJ, Ferrari AJ, Erskine HE, Charlson FJ, Norman RE, Flaxman AD, Johns N, et al. Global burden of disease attributable to mental and drug use disorders: findings from the global burden of disease study 2010. The Lancet. 2013;382(9904):1575–86. doi:10.1016/S0140-6736(13)61611-6.
- Roy É, Haley N, Leclerc P, Sochanski B, Boudreau J-F, Boivin J-F. Mortality in a cohort of street youth in Montreal. Jama. 2004;292(5):569–74. doi:10.1001/jama.292.5.569.
- Ginzler JA, Garrett SB, Baer JS, Peterson PL. Measurement of negative consequences of drug use in street youth: an expanded use of the rutgers alcohol problem index. Addict Behav. 2007;32(7):1519–25. doi:10.1016/j.addbeh.2006.11.004.
- Gomez R, Thompson SJ, Barczyk AN. Factors associated with drug use among homeless young adults. Drug Abuse. 2010;31(1):24–34. doi:10.1080/08897070903442566.
- Johnston LD, O’Malley PM, Bachman JG, Schulenberg J, Miech R. Demographic subgroup trends among adolescents in the use of various licit and illicit drugs, 1975–2013. Ann Arbor, MI: Institute for Social research, University of Michigan; 2014.
- Bose J, Hedden SL, Lipari RN, Park-Lee E. Key drug use and mental health indicators in the United States: results from the 2015 national survey on drug use and health. In: Drug Abuse and Mental Health Services Administration (SAMHSA), editor Rockville, MD. 2016.
- Bucher CEC. Toward a needs-based typology of homeless youth. J Adolesc Health. 2008;42(6):549–54. doi:10.1016/j.jadohealth.2007.11.150.
- Kemp PA, Neale J, Robertson M. Homelessness among problem drug users: prevalence, risk factors and trigger events. Health Soc Care Community. 2006;14(4):319–28. doi:10.1111/hsc.2006.14.issue-4.
- Edidin JP, Ganim Z, Hunter SJ, Karnik NS. The mental and physical health of homeless youth: a literature review. Child Psychiatry Hum Dev. 2012;43(3):354–75. doi:10.1007/s10578-011-0270-1.
- Childress S, Reitzel LR, Santa Maria D, Kendzor DE, Moisiuc A, Businelle MS. Mental illness and drug use problems in relation to homelessness onset. Am J Health Behav. 2015 Jul 1;39(4):549–55. doi:10.5993/AJHB.39.4.11.
- Wenzel SL, Tucker JS, Golinelli D, Green HD, Zhou A. Personal network correlates of alcohol, cigarette, and marijuana use among homeless youth. Drug Alcohol Depend. 2010;112(1):140–49. doi:10.1016/j.drugalcdep.2010.06.004.
- Aubry T, Klodawsky F, Coulombe D. Comparing the housing trajectories of different classes within a diverse homeless population. Am J Community Psychol. 2012;49(1–2):142–55. doi:10.1007/s10464-011-9444-z.
- Coates J, McKenzie-Mohr S. Out of the frying pan, into the fire: trauma in the lives of homeless youth prior to and during homelessness. J Soc Soc Welfare. 2010;37:65.
- Shiffman S. Ecological momentary assessment (EMA) in studies of drug use. Psychol Assess. 2009;21(4):486. doi:10.1037/a0017074.
- Sinha R. Chronic stress, drug use, and vulnerability to addiction. Ann N Y Acad Sci. 2008;1141(1):105–30. doi:10.1196/nyas.2008.1141.issue-1.
- Narendorf SC, Santa Maria DM, Cooper J Youth Count 2.0! Full report of findings Houston, TX. 2015. http://www.uh.edu/socialwork/New_research/projects/Youth%20Count%202.0/index.php
- Thompson SJ, Barczyk AN, Gomez R, Dreyer L, Popham A. Homeless, street-involved emerging adults: attitudes toward drug use. J Adolesc Res. 2010;25(2):231–57. doi:10.1177/0743558409350502.
- Tyler KA, Johnson KA. Pathways in and out of drug use among homeless-emerging adults. J Adolesc Res. 2006;21(2):133–57. doi:10.1177/0743558405285494.
- Tyler KA, Gervais SJ, Davidson MM. The relationship between victimization and drug use among homeless and runaway female adolescents. J Interpers Violence. 2013;28(3):474–93. doi:10.1177/0886260512455517.
- Bender K, Brown SM, Thompson SJ, Ferguson KM, Langenderfer L. Multiple victimizations before and after leaving home associated with PTSD, depression, and drug use disorder among homeless youth. Child Maltreat. 2015;20(2):115–24. doi:10.1177/1077559514562859.
- Serre F, Fatseas M, Swendsen J, Auriacombe M. Ecological momentary assessment in the investigation of craving and drug use in daily life: a systematic review. Drug Alcohol Depend. 2015;148:1–20. doi:10.1016/j.drugalcdep.2014.12.024.
- Rowe C, Hern J, DeMartini A, Jennings D, Sommers M, Walker J, Santos G-M. Concordance of text message ecological momentary assessment and retrospective survey data among drug-using men who have sex with men: a secondary analysis of a randomized controlled trial. JMIR mHealth and uHealth. 2016;4(2):e44. doi:10.2196/mhealth.5368.
- Stone A, Shiffman S, Atienza A, Nebeling L. The science of real-time data capture: self-reports in health research. Oxford, UK: Oxford University Press. 2007.
- Stone AA, Shiffman S. Capturing momentary, self-report data: A proposal for reporting guidelines. Ann Behav Med. 2002;24(3):236–43. doi:10.1207/S15324796ABM2403_09.
- Wen C, Schneider S, Stone S, Spruijt-Metz D. Compliance with mobile ecological momentary assessment protocols in children and adolescents: asystematic review and meta-analysis. J Med Internet Res. 2017;19(4):e132. doi:10.2196/jmir.6641.
- Wright CJ, Dietze PM, Crockett B, Lim MS. Participatory development of MIDY (Mobile Intervention for Drinking in Young people). BMC Public Health. 2016;16(1):1. doi:10.1186/s12889-016-2876-5.
- Thrane LE, Yoder KA, Chen X. The influence of running away on the risk of female sexual assault in the subsequent year. Violence Vict. 2011;26(6):816–29. doi:10.1891/0886-6708.26.6.816.
- Tyler KA, Johnson KA. Trading sex: voluntary or coerced? The experiences of homeless youth. J Sex Res. 2006;43(3):208–16. doi:10.1080/00224490609552319.
- Whitbeck LB, Hoyt DR, Yoder KA. A risk-amplification model of victimization and depressive symptoms among runaway and homeless adolescents. Am J Community Psychol. 1999;27(2):273–96. doi:10.1023/A:1022891802943.
- Liao Y, Skelton K, Dunton G, Bruening M. A systematic review of methods and procedures used in ecological momentary assessments of diet and physical activity research in youth: an adapted STROBE checklist for reporting EMA studies (CREMAS). J Med Internet Res. 2016;18(6):e151. doi:10.2196/jmir.4954.
- Quistberg DA. Assessing health literacy in African American and Caucasian adults: disparities in rapid estimate of adult literacy in medicine (REALM) scores. Fam Med. 2004;36(8):575–81.
- Murphy PW, Davis TC, Long SW, Jackson RH, Decker BC. Rapid estimate of adult literacy in medicine (REALM): a quick reading test for patients. J Reading. 1993; 37(2):124–30.
- Shiffman S, Hickcox M, Paty JA, Gnys M, Richards T, Kassel JD. Individual differences in the context of smoking lapse episodes. Addict Behav. 1997;22(6):797–811. doi:10.1016/S0306-4603(97)00063-4.
- Stone AA, Schwartz JE, Neale JM, Shiffman S, Marco CA, Hickcox M, Paty J, Porter LS, Cruise LJ. A comparison of coping assessed by ecological momentary assessment and retrospective recall. J Pers Soc Psychol. 1998;74(6):1670. doi:10.1037/0022-3514.74.6.1670.
- Watkins KL, Regan SD, Nguyen N, Businelle MS, Kendzor DE, Lam C, Balis D, et al. Advancing cessation research by integrating EMA and geospatial methodologies: associations between tobacco retail outlets and real-time smoking urges during a quit attempt. Nicotine Tob Res. 2013Sep 21;16 (Suppl_2): S93-101.
- Businelle MS, Ma P, Kendzor DE, Reitzel LR, Chen M, Lam CY, Bernstein I, Wetter DW. Predicting quit attempts among homeless smokers seeking cessation treatment: an ecological momentary assessment study. Nicotine Tob Res. 2014;16(10):1371–78. doi:10.1093/ntr/ntu088.
- Reitzel LR, Kendzor DE, Nguyen N, Regan SD, Okuyemi KS, Castro Y, Wetter DW, Businelle MS. Shelter proximity and affect among homeless smokers making a quit attempt. Am J Health Behav. 2014;38(2):161–69. doi:10.5993/AJHB.38.2.1.
- Akaike H. A new look at the statistical model identification. IEEE Trans Automat Contr. 1974;19(6):716–23. doi:10.1109/TAC.1974.1100705.
- Venables WN, Ripley BD. Modern applied statistics with S (statistics and computing). 4th edition. New York City, NY: Springer Publishing Co., 2003..
- Šimundić A-M. Measures of diagnostic accuracy: basic definitions. Ejifcc. 2009;19(4):203.
- Castaneto MS, Gorelick DA, Desrosiers NA, Hartman RL, Pirard S, Huestis MA. Synthetic cannabinoids: epidemiology, pharmacodynamics, and clinical implications. Drug Alcohol Depend. 2014;144:12–41. doi:10.1016/j.drugalcdep.2014.08.005.
- Winstock AR, Barratt MJ. The 12‐month prevalence and nature of adverse experiences resulting in emergency medical presentations associated with the use of synthetic cannabinoid products. Hum Psychopharmacology: Clin Exp. 2013;28(4):390–93. doi:10.1002/hup.2292.
- Winstock AR, Barratt MJ. Synthetic cannabis: a comparison of patterns of use and effect profile with natural cannabis in a large global sample. Drug Alcohol Depend. 2013;131(1):106–11. doi:10.1016/j.drugalcdep.2012.12.011.
- Bowen S, Marlatt A. Surfing the urge: brief mindfulness-based intervention for college student smokers. Psychol Addict Behav. 2009;23(4):666. doi:10.1037/a0017127.
- Shiffman S., Commentary on McCarthy, et al. (2015): Ecological momentary assessment‐Reactivity Intervention? Addiction. 2015;110(10):1561–62.