The At Risk Youth Study (ARYS) is an ongoing prospective cohort of homeless and street-involved youth in Vancouver, Canada. The study has been described in detail previously (Wood et al., 2006
). Briefly, the study consists of a convenience sample recruited through extensive street-based outreach and snowball sampling in which street youth were encouraged to recruit their peers. Outreach teams consisting of street nurses and study staff worked in close collaboration with local youth service organizations and agencies to refine the recruitment process and to ensure that specific hidden subpopulations (e.g., street youth who identify as lesbian, gay, bisexual, transsexual, or transgendered [LGBTT]) were represented. Furthermore, these organizations were encouraged to refer eligible youth to the study. Although these recruitment techniques are inferior to a random probability-based sampling, extensive efforts were taken to maximize the representativeness of the sample. For example, outreach was systematically conducted in a range of city neighbourhoods in areas where street youth are known to congregate, including shelters, youth services, and on the street. Outreach was also conducted at different times during the day and at night.
The term “street-involved youth” has been defined as an adolescent who spends a substantial amount of time on the street or is heavily engaged in the street economy (Marlatt, 2002
), and may include youth who are absolutely, periodically, or at imminent risk of being homeless (Daly, 1996
). As in previous studies of street-involved youth (Roy et al., 2003a
; DeMatteo et al., 1999
), housing status was not included as an eligibility criterion in order to capture as much as possible the diverse spectrum of street youth. Specific eligibility criteria included being between the age of 14 and 26 years of age at baseline, illicit drug use other than or in addition to marijuana in the past 30 days, and the provision of informed consent. At baseline and semi-annually, participants complete an interviewer-administered questionnaire and provide blood samples for HIV and hepatitis C (HCV) serology. All participants receive a monetary stipend of $20 CDN after each visit. The study has been approved by the University of British Columbia/Providence Health Care Research Ethics Board.
All participants who were recruited and completed a baseline survey between September 1, 2005 and October 31, 2006 were included in this cross-sectional analysis. The primary outcomes of interest in this study were number of sexual partners in the past 6 months and condom use during vaginal and anal intercourse during the past 6 months. Participants were asked to report how many male and female partners they had engaged in sexual activities with, excluding those with whom they had engaged in sex for money, drugs, shelter, or gifts. Since sex work is known to be associated with a profile of sexual risk behaviour distinct from that in the context of voluntary sex among both male and female street youth (Haley et al., 2004
; Weber et al., 2004
), we chose to omit this activity from our analyses. For both same and opposite sex partnerships, participants were also asked to report how often a condom was used during vaginal and anal intercourse. Possible responses included: always, regularly, occasionally, and never. To be consistent with previous studies of condom use among street-involved youth (de Carvalho et al., 2006
; Moon et al., 2001
), this variable was dichotomized into “consistent” (i.e., always) and “inconsistent” (i.e., regularly, occasionally, or never) condom use. Participants who reported more than one type of sexual activity in the past six months and who reported discordant condom use patterns were coded as inconsistent condom users.
The primary explanatory variable of interest in this study was current place of residence. For analytical purposes, we created a categorical variable with the following three levels: 1) stably housed – living in an apartment, house or single room occupancy hotel (SRO); 2) unstably housed – living in a shelter or hostel; and 3) homeless – living on the street or having no fixed address (NFA). Other independent variables included a range of sociodemographic, individual level, drug-related, and social factors, and were chosen based on their known or a priori
hypothesized relationship with both housing status and one or both of the sexual risk behaviours listed above. Sociodemographic variables that were examined included: age (per year older), sex (female vs. male), Aboriginal ethnicity (yes vs. no) and sexual orientation (LGBTT vs. heterosexual). Other individual level factors that were examined included engaging in anal intercourse in the past six months (yes vs. no), sexual abuse (yes vs. no), depression, and self-efficacy for limiting HIV risk behaviours (LHRB). Sexual abuse was defined as answering “yes” to the question, “Have you ever been sexually abused?”, while the latter two variables were defined using validated, published scales. The Center for Epidemiologic Studies Depression Scale (CES-D) is designed to measure current levels of depressive symptoms and has been shown to have high levels of internal consistency and reliability among groups of adolescents (Radloff, 1977
; Rushton et al., 2002
). The presence of depressive symptoms was evaluated using a well-defined cut-off (CES-D ≥ 16 [yes] versus CES-D < 16 [no]). The self-efficacy for LHRB Scale is a validated instrument found to have high levels of consistency among at-risk youth (Smith et al., 1996
; Kang et al., 2004
). Self-efficacy refers to the belief in one’s capability of performing actions to attain certain goals (Bandura, 1986
); therefore, the LHRB Scale consists of questions that require a respondent to record how sure they are that they could perform sexual and injection-related risk reduction behaviours, such as talking about safe sex with a partner, using a condom correctly during sex, and refusing to share needles with friends. Responses were dichotomized into “high” versus “low” self-efficacy for LHRB based on their value with respect to the sample median. Social and drug-related factors that were examined included: relationship status (single or casually dating vs. regular partner or married), drug dealing (yes vs. no), alcohol dependence (yes vs. no), crack use (yes vs. no), cocaine use (yes vs. no), heroin use (yes vs. no), crystal methamphetamine use (yes vs. no), injection drug use (yes vs. no), and sharing syringes (yes vs. no). Alcohol dependence was measured using the Perceived-Benefit-of-Drinking Scale (PBDS), a 5-item true/false validated instrument that is used to assess drinking behaviours among adolescents (Petchers and Singer, 1987
; Petchers et al., 1988
). A standard a priori
defined cut-off of three or higher was used to define alcohol dependence. All other drug-use variables refer to behaviours occurring in the past six months: cocaine, heroin, and crystal methamphetamine use refer to both injection and inhalation routes of consumption. To be consistent with our previous work, syringe sharing was defined as lending or borrowing a used syringe in the past six months (Kerr et al., 2005
). Since sexual risk behaviours are known to be interdependent (Aral, 2004
), each dependent variable was included as a potential explanatory variable when not used as the primary dependent variable of interest.
Initially, we examined bivariate associations between the individual level, social, drug-related and environmental variables and each sexual risk behaviour outcome. To analyse the sexual partnership data, a Poisson-type regression was used to estimate the unadjusted incidence rate ratio (IRR) and 95% confidence intervals (95% CI) associated with each independent variable. The distribution of recent sex partners was highly skewed; therefore, in order to account for overdispersion, we used a log-linear quasi-Poisson regression. This approach relaxes the assumption that the variance of the underlying distribution is equal to the mean by allowing the variance to be modeled as a linear function of the mean (McCullagh and Nelder, 1999
; Ver Hoef and Boveng, 2007
). To analyse the bivariate associations with consistent condom use, dichotomous and categorical variables were examined using the Pearson χ2
Since the objective of this study was to determine whether environmental conditions, independent of established sociodemographic and individual level factors, are associated with sexual risk behaviour, we fit a series of confounding models based on an approach described by Rothman and Greenland (Rothman and Greenland, 1998
; Maldonado and Greenland, 1993
). For a variable to be considered a confounder in the relationship between place of residence and sexual risk behaviour, it must be associated with both the behaviour and with housing status. Therefore, we conducted initial bivariate screenings based on a conservative p
-value of 0.20 to determine which explanatory variables were associated with the outcome and thus may be potential confounders. We then used these variables to create two multivariate “full” models. These models were then subjected to a manual stepwise approach to select significant confounders. Starting with the “full” models, variables that did not alter the relative change in the coefficient of the primary variable of interest by more than 10% were removed in a sequential fashion. Since the primary variable of interest was categorical with two levels, variables were considered significant confounders if their removal altered one or both of the coefficients by more than 10%. This technique has been described and used successfully by several authors (Lima et al., 2007
; Lima and Kopec, 2005
; Maldonado and Greenland, 1993
). To account for the well-established confounding effects of age, sex, ethnicity, and sexual orientation, these variables were forced into the “full” models and were not subjected to the stepwise procedure. All statistical modelling was conducted using S-PLUS software version 8.0. All reported p
-values are two-sided.