Participants and procedures

Participants were part of an evaluation of AWARE, a brief substance use and sexual risk reduction program for 18 to 25 year olds experiencing homelessness. Findings from the intervention trial and description of the intervention can be found elsewhere.26, 27 Data for the present analyses come from the baseline survey and from four follow-up surveys (3, 6, 12, and 24 months post-baseline). Participants were recruited from three drop-in centers serving young adults experiencing homelessness in Los Angeles County. Drop-in centers provide services to address the basic needs of young people experiencing homelessness (food, clothing), but oftentimes offer higher level services such as case management and other programs to meet health and social service needs. The three drop-in centers included in this study were diverse in location (e.g., Hollywood, Venice/Santa Monica) and population served, with one of the two drop-in centers in Hollywood offering services specifically for sexual and gender minority youth experiencing homelessness.

To be eligible for the study, participants needed to (1) be between the ages of 18 and 25, (2) be currently seeking any services at one of the drop-in centers, (3) plan to be in the study area for the next month, (4) be willing to provide contact information for follow-up surveys, (5) be reachable by e-mail or phone for follow-up, (6) be English-speaking, and (7) display no evidence of cognitive impairment at screening. All procedures were approved by the institution’s Internal Review Board.

Three hundred and seventy-one young adults at the drop-in centers were approached for screening, resulting in a final sample of 276 participants (see Tucker et al.22 for more details). Five participants had missing data on baseline predictor variables, resulting in an analytic sample of 271. Table 1 contains a description of the sample, which was about 22 years old on average, mostly male sex at birth (72%), and non-White (84%), with 45% reporting sexual and gender minority identity (42.7% reported sexual minority identification and 13.4% reported gender minority identification, with 28% of sexual minorities and 86% of gender minorities also identifying with the other minority identification). Sample demographics were similar to the demographic profile of the population of young people experiencing homelessness in Los Angeles County.

Table 1 Participant characteristics at baseline. Note: 1Responses of gay, lesbian, bisexual, questioning, or asexual for the sexual orientation survey item. Four participants did not respond to the sexual orientation item. 2Responses of “gender neutral,” “other gender,” or “transgender” to the gender identification item. 3Use of housing services value corresponds with a response option of use of housing services “3 to 5 days” in the past 3 months

Baseline surveys were completed in person via paper-pencil survey, whereas follow-up surveys were generally complete via online survey or phone interview. Our team has extensive experience tracking young people who experience homelessness and have developed tracking and locator information to limit attrition.28 Thus, 87% of the sample was retained at the 24-month follow-up.

Measures Demographics and control variables

Participants’ age, birth sex (male or female), gender identification, sexual orientation, race, and ethnicity were assessed. A dichotomous variable indicated “Sexual / Gender minority” was created and set equal to 1 if participants reported a gender identity that was different from their birth sex, transgender identity, or non-heterosexual orientation (this variable was equal to 0 if participants’ gender identity matched their birth sex and if participants reported “straight/heterosexual” orientation). Age at first homelessness was assessed with the question “How old were you the first time you left home and were living on your own, apart from a parent or guardian [even if it was just a short period of time]?” Being currently enrolled in school full- or part-time (vs. not in school) and currently employed full- or part-time (vs. unemployed) were also assessed. Dummy variables for drop-in center location (with the Venice drop-in center as reference) and intervention group (1 = AWARE intervention, 0 = usual care at drop-in) were included as control variables and effects were not interpreted.

Substance use

Use of three types of substances in the past 30 days was assessed. Heavy alcohol use was assessed by first presenting participants with a definition and images of standard drinks (i.e., “one regular size can/bottle of beer or wine cooler, one 5 ounce glass of wine, one mixed drink, or one shot glass of 1.5 ounce liquor”). Heavy alcohol use, defined according to the National Institute on Alcohol Abuse and Alcoholism, was assessed as the number of days participants reported drinking five or more drinks of alcohol in a row “within a couple of hours.” Past 30-day use of 13 classes of drugs (e.g., cannabis, methamphetamine, prescription drug misuse) was also assessed. Number of cannabis use days was the number of days participants reported using “marijuana or hashish.” Number of illicit drug use days was assessed by asking “How many days did you use any of the drugs listed above, not including marijuana?”

Health and social functioning

Six variables were used to describe participants’ health and social functioning. Participants reported their general health, ranging from “Excellent” (1) to “Poor” (5). Depression symptoms in the past 2 weeks were assessed using the eight-item Patient Health Questionnaire [PHQ-8;29]. Probable depression diagnosis (1 = yes, 0 = no) was indicated by a PHQ-8 score of greater than or equal to 10. Friendship relationship quality was assessed using the PROMIS Pediatric Peer Relationships Scale30 which consists of the mean of three items (e.g., “I was able to count on my friends”) with response options ranging from “Never” (1) to “Almost Always” (5). A binary variable also indicated if participants had been pregnant or had impregnated someone else in the past three months (1 = yes, 0 = no).

Use of housing services

Participants were asked how often they used formal services “at a drop-in center or other agency/organization” in the past 3 months. Participants indicated the number of days they used services “to help you find housing,” with response options on a six-point ordinal scale ranging from “0 days” (1) to “more than 15 days” (6).

Housing

At each survey, participants’ housing situation was assessed with the item: “In the past 3 months, on average, how often have you spent the night in each of the following places?” This was followed by a list of 10 different housing options, with eight response options for each choice ranging from “Never” to “Every day.” From these items, two dichotomous outcome variables were created. Participants were considered stably housed in their own home if they selected “Every day” for the item “Your own house, apartment or room” and received a value of 1 on this outcome variable. Participants who chose any option less than “Every day” (e.g., “Never” to “4-5 times a week”) were considered not stably housed and received a value of zero on this outcome variable. Options for housing that were not considered stably housed in their own home were staying temporarily in someone else’s apartment or house, in an emergency shelter or transitional housing program, outdoors or on the street, in a car or vehicle, in an abandoned building, in a hotel or motel, or somewhere else temporarily. Secondly, being unsheltered was indicated (outcome variable = 1) if participants reported that they had spent at least one night (i.e., “Less than once a month” to “Every day”) in at least one of the following places in the past 3 months: “Outdoors, the street, or a park,” “Car or other private vehicle (e.g., van, camper),” or “Abandoned building.” Participants who reported “Never” staying in all three of these places received a value of zero on this outcome variable.

Analytic plan

Latent growth curve modeling was used to estimate trajectories of stable housing and being unsheltered. Separate growth models were fit for each outcome. Because both outcomes were dichotomous, the authors used a maximum likelihood estimator and logit link. Model specification for categorical outcomes followed a common method,31 in which means of the intercepts were fixed at zero, and slopes were freely estimated. Different models were specified for no growth, linear, and quadratic change trajectories to determine the best fitting and most appropriate unconditional model. After arriving at the best fitting unconditional model, growth factors (i.e., intercepts and slopes) were regressed on covariates (demographics, substance use, health and social functioning, and service use). The authors undertook a model building process in which covariates were added as predictors of slope factors one at a time in a series of separate (bivariate) models. Covariates that were associated with each slope factor at p ≤ 0.10 in bivariate models were then included in a final multivariate model. All regression coefficients are presented as standardized parameter estimates with accompanying standard errors. Missing data were minimal across baseline variables (n = 1 missing for race/ethnicity and age; n = 2 missing for health, employment, and use of housing services). Therefore, the default in Mplus for handling missing data was used, in which cases with missing data on predictor variables were excluded from multivariate models.