PWID who experience homelessness are subject to additional structural and environmental barriers—such as poverty and exposure to violence—that amplify the IDU-related harms they face . Further, among PWID, experiencing homelessness is associated with an additional elevated risk of acquiring HIV and hepatitis C . While preventing injection-naïve individuals from transitioning into IDU has long been a public health goal , better characterizing the role of homelessness in transitions into IDU could directly inform strategies to respond to some of the upstream drivers of IDU-related morbidity and mortality. Recent research has highlighted the key role of experienced PWID in assisting injection naïve individuals initiating IDU . Across study samples, between 75%–95% of PWID reported that their IDU initiation was assisted by established PWID . While research has demonstrated that experiencing an episode of homelessness in the past six months increases the risk that injection-naïve individuals initiate IDU , there is a lack of research concerning the relationship between recent homelessness and the provision of IDU initiation assistance among PWID. In fact, prior studies of IDU initiation assistance have operationalized recent homelessness or housing status as a covariate to be controlled for in subsequent analyses, rather than as a critical factor in and of itself . In the present study, we therefore assessed the association between recent homelessness and providing IDU initiation assistance among PWID from two cities in North America .Preventing Injecting by Modifying Existing Responses is a multi-cohort, multicountry, plant benches mixed-methods study with a primary aim of identifying socio-structural factors that influence the likelihood that PWID help injection-naïve individuals inject for the first time . For this study, data were drawn from four PRIMER-affiliated longitudinal cohort studies in Tijuana, Mexico and Vancouver, Canada. In Tijuana, PRIMER was conducted within the Proyecto El Cuete cohort study .

For ECIV, at baseline, all participants were at least 18 years old, had reported IDU in the prior month, spoke at least Spanish or English, were residing in Tijuana with no plans to relocate, and were not participating in any other intervention studies . In Vancouver, data were collected within three ongoing cohort studies: the At-Risk Youth Study ; the AIDS Care Cohort to Evaluate exposure to Survival Services study; and the Vancouver Injection Drug Users Study . For ARYS, recruited participants were between the ages of 14 and 26, reported illicit drug use in the past month, and reported recently being homeless or accessing services intended for homeless youth at baseline . For ACCESS, recruited participants were at least 18 years old, HIV seropositive, and reported illicit drug use at baseline . For VIDUS, recruited participants were at least 18 years old, HIV seronegative, and reported IDU on at least one occasion in the past month at enrolment. At recruitment and semiannually thereafter, all participants of these PRIMER-affiliated cohort studies completed interviewer-administered questionnaires that capture participant-reported information on socio-demographic characteristics and drug use behaviors. Starting in late 2014, corresponding cohort questionnaires were amended under PRIMER to add survey items concerning participants’ experiences with providing injection initiation assistance to others. The first interview completed by a participant involving the PRIMER items on injection initiation assistance is referred to as that participant’s PRIMER baseline interview . The present study includes data collected on ECIV and ARYS/ ACCESS/VIDUS participants from 2014 to 2017. The PRIMER study was approved by the Institutional Review Board of the University of California San Diego . It is also important to highlight that the dynamics of homelessness and IDU are different across these two sites. While there are challenges in estimating the number of people who experience homelessness, estimates indicate that, at minimum, several thousand individuals experience homelessness each year in both Tijuana and Vancouver .

In Vancouver, homelessness and IDU are concentrated and highly visible in the Downtown Eastside neighborhood . This centralization reduces barriers to recruiting and providing resources to PWID. Whereas, in Tijuana, homelessness is more dispersed and encampments that do arise are frequently subject to law enforcement interaction . As such, our study reflects on the relationship between homelessness and IDU initiation assistance provision across two heterogenous settings, expanding the potential generalizability of our findings. Our study was restricted to members of the ECIV and ARYS/ACCESS/VIDUS cohorts who: 1) completed a PRIMER baseline interview within the study window; 2) reported a history of IDU at baseline; and 3) completed at least one follow-up visit six months after baseline. Eligible participants contributed a minimum of 1 and a maximum of 5 follow-up visits. If a participant had missing baseline data for any time-varying measure , then baseline was redefined to be that participant’s first subsequent visit with complete data. All subsequent PRIMER follow-up visits within the study period with complete data for a participant were included. If a participant had missing data for a follow-up visit, then that follow-up visit and all subsequent follow-up visits for that participant were excluded from the analysis. The outcome of interest was recent provision of IDU initiation assistance . To operationalize this measure, participants were asked if they had helped an injection-naïve individual inject for the first time in the past six months. This question is intended to capture participants’ recent experiences with direct assistance and/or indirect assistance . The exposure of interest was recent homelessness , defined via self-report as experiencing an episode of homelessness in the past six months. Due to differences in the cohort questionnaires by setting, the self-reported exposure was measured differently for participants from Tijuana and Vancouver. In Tijuana, participants were given a set of locations and asked to mark all the places they have lived or slept in the past six months.

Participants that reported having lived or slept in their workplace, in a vehicle, in an abandoned building, in a shelter, on the streets, or in a shooting gallery in the past six months were deemed to have recently experienced homelessness. In Vancouver, participants were asked a single yes/no question: “Have you been homeless in the last six months?” with those responding “yes” deemed to have recently experienced homelessness. Both the exposure and outcome were repeatedly assessed at each visit over follow-up. We identified a set of covariates a priori that might confound the relationship between our exposure and outcome of interest based on prior literature. The set consists of both baseline-fixed and time-varying covariates. Baseline-fixed covariates included: age , gender , and cohort . Time-varying covariates included: whether participants reported being stopped by law enforcement in the past six months ; whether participants reported being incarcerated in the past six months ; whether participants reported IDU in the past six months ; and for Vancouver participants only, neighborhood of residence . Excluding the baseline-fixed covariates, values of all other variables were allowed to vary over time to reduce misclassification bias. Prior to any analyses, time-varying covariate values at a given followup visit t were recoded to their corresponding value at visit t-1 occurring six months earlier. This lagging was done to ensure that covariate measurement always preceded both exposure and outcome measurement at the same visit. Due both to differences in underlying study design and how the exposure was defined between the Tijuana and Vancouver cohorts ,gardening rack all analyses described herein were undertaken separately by setting. Given our interest in estimating the effect of recent homelessness on recent provision of injection initiation assistance , it is important to note that traditional regression-based approaches to control for measured confounding may yield a biased effect estimate when the set of covariates includes time-varying variables that are caused by prior exposure and also influence subsequent exposure and outcome values . This consideration is relevant to our study, as we have measured several time-varying covariates that may satisfy these criteria; for example, recent IDU may be both a consequence of prior homelessness and a confounder of subsequent homelessness and providing injection initiation assistance). Alternatively, an unbiased estimate of the effect of interest can be obtained from an inverse-probability-of-treatment-weighted marginal structural model, which accounts for baseline-fixed and time-varying confounding via weights . Estimation of this marginal structural model requires two steps: first, we calculate stabilized inverse probability-of-treatment weights for each person-visit occurring after baseline to account for confounding; and, second, we fit a generalized estimating equations logistic regression model to the weighted sample to estimate the parameters of a marginal structural model. IPTWs are calculated in order to evenly distribute potential confounders across the different treatment groups – the application of the weights to the study sample generates an artificially balanced pseudo-sample in which recent homelessness status is independent of all measured confounders . The IPTW approach is particularly appropriate given that it can effectively account for confounding caused by time-varying measures in longitudinal analyses . See the Supplemental Materials for full details on the calculation of IPTWs. Next, we fit a GEE logistic model to the inverse-probability-of-treatment-weighted sample with our repeated measures outcome regressed on terms for exposure , time , and all baseline-fixed covariates . A first-order autoregressive working correlation matrix was specified to account for repeated measures within participants – meaning that the model assumed that a participant’s outcome response at follow-up visit t was correlated with their outcome response at visit t-1 .

Assuming the absence of model misspecification, unmeasured confounding, and informative censoring, the inverse-probability-of-treatment-weighted GEE coefficient estimates estimate the corresponding causal parameters of a marginal structural model . In other words, under these assumptions, the exponentiated exposure coefficient estimate from our weighted model – which is an adjusted odds ratio – may be interpreted as the relative effect of recent homelessness on a participant’s odds of providing injection initiation assistance over the same six-month period . Corresponding 95% confidence intervals were calculated for effect estimates using robust sandwich-type standard errors with clustering by participant. We performed two sensitivity analyses to assess the influence of measured confounding on our estimates of the association between recently experiencing homelessness and recently providing IDU initiation assistance: first, to assess the influence of measured time-varying confounding, we ran the GEE logistic model as described above without the IPTWs; second, to assess the influence of measured time varying and baseline-fixed confounding, we ran the GEE logistic model as described above without the IPTWs and without adjusting for baseline-fixed covariates. We identified 703 eligible participants in Tijuana and 1551 eligible participants in Vancouver . At baseline, 12.5% of participants in Tijuana and 23.3% of participants in Vancouver reported experiencing homelessness in the past six months. Individuals in Vancouver who reported recent homelessness at baseline were younger on average than those who had not . In both Tijuana and Vancouver, individuals who reported recent homelessness at baseline reported higher prevalence of being stopped by police in the past six months. In Vancouver, 21.0% of those who recently experienced homelessness at baseline reported recent incarceration versus just 3.0% of those who did not report recent homelessness. In Tijuana, past six-month IDU was more prevalent among those reporting recent homelessness at baseline than those who did not . The same was true in Vancouver, where 80.9% of recently homelessness participants reported IDU in the past six months compared to 58.9% of participants who did not report recent homelessness. A higher proportion of those reporting recent homelessness at baseline also reported recently providing IDU initiation assistance in both Tijuana and Vancouver . The median number of follow-up visits was 5 in Tijuana and 4 in Vancouver . At a given follow-up visit, between 11.6% and 16.5% of participants in Tijuana and between 9.4% and 18.9% of participants in Vancouver reported experiencing homelessness in the past six months. Between 3.3% and 5.4% of participants in Tijuana and 2.5% and 4.1% of participants in Vancouver reported recently assisting an IDU initiation at each follow-up visit. In Tijuana and Vancouver, respectively, 79 and 150 participants reported assisting at least one IDU initiation across the study period, with 19 and 28 of these participants reporting recent injection initiation assistance provision at multiple follow-up visits. In Tijuana, at a given follow-up visit, between 12.5% and 30.4% of participants who reported recently assisting a first-time injection also reported recent homelessness during the same six-month period. In Vancouver, through the first 4 follow-up visits between 18.4% and 36.5% of participants reporting recently assisting IDU initiation also reported recent homelessness, though this fell to 5.6% at the 5th follow-up.