The Vancouver Injection Drug Users Study (VIDUS) and AIDS Care Cohort to Evaluate Access to Survival Services (ACCESS) are open prospective cohorts of illicit drug users in Vancouver. The recruitment and follow up procedures for the two studies are largely identical to allow for analyses of merged data, with the only key differences being that HIV positive individuals are followed in ACCESS whereas HIV-negative individuals are followed in VIDUS. In both studies the primary modes of enrollment were self-referral, word of mouth, and street outreach. Detailed sampling and recruitment procedures for these two cohorts have been described elsewhere [18,19].
To be eligible, participants were 18 years of age or older, had ever injected illicit drugs and resided in the greater Vancouver region. Participants with no completed follow up surveys were excluded. All participants provided written informed consent. Participants were given a $20 stipend at each study visit for their time and transportation. The study was approved by the University of British Columbia/Providence Healthcare Research Ethics Board.
At baseline and semianually thereafter, participants completed an interviewer-administered questionnaire that elicited a range of data, including demographic characteristics, housing status, injection and non-injection drug use, and sexual risk behaviors. In addition, venous blood samples were drawn at each visit and tested for HIV and hepatitis C virus (HCV) antibodies among those previously testing negative for these diseases. Venous blood samples taken from HIV positive individuals were assessed for HIV disease progression [18]. All participants had private interviews and were offered both pre- and post-test counseling with trained nurses. Referral for free healthcare was provided to those who tested HIV positive and these individuals were subsequently followed in ACCESS.
The present study included PWID who were recruited and completed at least one follow up visit between May 1996 and December 2012. To avoid potential bias relating to long durations between the last study visit where behavioural information was assessed and the date of death, individuals who were identified as deceased more than 24 months after their last follow up visit were censored on the date of the last follow up.
The primary endpoint in this analysis was all-cause mortality. Here, we ascertained all-cause mortality rates among participants through a confidential record linkage using personal health numbers with the British Columbia Vital Statistics Agency and through ongoing follow up with contacts provided by participants. The primary explanatory variable of interest was unstable housing in the previous six months. As previously, unstable housing was defined as living in a single room occupancy hotel, shelter or other transitional housing, or living on the street [20,21].
Potential confounders that were considered included gender (male vs. female); age (per year older); ancestry (Caucasian vs. non-Caucasian); HIV serostatus (positive vs. negative); and sex work involvement (yes vs. no). A number of substance use behaviors (in the previous six months) were also considered, including ≥ daily heroin injection (yes vs. no), ≥ daily cocaine injection (yes vs. no), ≥ daily crack cocaine smoking (yes vs. no), and current enrolment in a methadone program (yes vs. no). Other covariates that were considered included incarceration (yes vs. no) and HCV serostatus (positive vs. negative). With the exception of age, gender and ancestry, all variables were in reference to the prior six months and measured at each semiannual follow up visit and were treated as time-updated.
As a first step, we used Chi-square test and Wilcoxon rank sum test to compare the baseline characteristics of the participants who did and did not report unstable housing at baseline. Those who did not report unstable housing at baseline were maintained as the reference group. All-cause mortality rate and 95% confidence interval [CI] were calculated using the Poisson distribution. Survival probabilities from all-cause mortality were estimated using the Kaplan-Meier product limit method, and compared using the two-sample log-rank test.
Next, we used bivariate Cox proportional hazards regression to examine the associations between each explanatory variable and time to all-cause mortality. To fit the multivariate model, we employed a conservative stepwise backward selection approach which considered the magnitude of change in the coefficient of unstable housing [22]. Specifically, we included all variables found to be associated with time to all-cause mortality in bivariate analyses at p < 0.10 in a multivariate model and used a stepwise approach to fit a series of reduced models. After comparing the value of the coefficient associated with unstable housing in the full model to the value of the coefficient in each of the reduced models, we dropped the secondary variable associated with the smallest relative change. We continued this iterative process until the minimum change exceeded 5%. Remaining variables were considered as potential confounders in a final multivariate model. All statistical analyses were performed using SAS software version 9.3 (SAS, Cary, NC). All p-values were two-sided.