Study procedures

The Vancouver Injection Drug Users Study (VIDUS) and the AIDS Care Cohort to evaluate Exposure to Survival Services (ACCESS) are ongoing open prospective cohorts of adult drug users recruited through word of mouth, street outreach, and referrals from community organizations in Vancouver, Canada. These studies have been described in detail previously [20]. Briefly, VIDUS enrolls HIV-negative persons who reported injecting an illicit drug at least once in the month preceding enrollment; ACCESS enrolls HIV-positive individuals who report using an illicit drug (other than, or in addition to, cannabis) in the previous month. For both cohorts, other eligibility criteria included being aged 18 years or older, residing in the greater Vancouver region and providing written informed consent. The study instruments and all other follow-up procedures for each study are essentially identical to allow for combined analyses. At baseline and semi-annually thereafter, participants complete an interviewer-administered questionnaire eliciting sociodemographic data as well as information pertaining to drug use patterns, risk behaviors, and health care utilization. Nurses collect blood samples for HIV and hepatitis C virus serology, provide basic medical care and arrange referrals to appropriate health care services if required. Participants receive a $30 (CDN) honorarium for each study visit. The University of British Columbia/Providence Healthcare Research Ethics Board provided ethical approval for both studies.

All participants who were enrolled in the cohorts between December 1, 2005 (the start date of the VIDUS and ACCESS cohorts) and November 30, 2014 (the most recent follow-up period available for the present analysis), and who reported ever injecting drugs preceding the baseline interview were included in the present analysis. Additionally, at each follow up, the sample was restricted to individuals who reported smoking crack cocaine in the previous 6 months because the analysis was focused on crack cocaine smoking.

Study variables

The primary outcome of interest was experiencing health problems associated with smoking crack in the previous 6 months. As in a previous study [21], this was defined as reporting at least one of the following health problems: “Burns”, “Mouth sores”, “Cut fingers / sores”, “Raw throat”, or “Coughing blood” to the question within the interviewer administered questionnaire:: “In the past 6 months, have you experienced any of the following health problems from smoking crack?”

The primary explanatory variable of interest was crack pipe acquisition source in the previous 6 months. This was defined as reporting health service points only (e.g. needle exchange programs, health clinics, temporary shelters) vs. a mix of health service points and other sources vs. other sources only (e.g. street, homemade, corner store), to the question: “In the past 6 months, where did you get your crack pipes?”

We also considered secondary explanatory variables that might confound the relationship between crack pipe acquisition sources and reporting health problems from smoking crack. These included sociodemographic characteristics, including: age (per year older); biological sex at birth (female vs. male); ancestry (white vs. non-white); residing in the DTES in the previous 6 months (yes vs. no); homelessness in the previous 6 months, defined as having no fixed address, sleeping on the street, or staying in a shelter or hostel (yes vs. no); involvement in drug dealing in the previous 6 months (yes vs. no); involvement in sex work in the previous 6 months (yes vs. no); educational attainment (less than high school vs. high school completion or higher). Drug-use variables referred to behaviours in the previous 6 months, and included: ≥ daily crack smoking (yes vs. no); ≥ daily non-injection crystal methamphetamine use (yes vs. no); binge non-injection drug use, defined as compulsive high-intensity non-injection drug use that exceeds normal patterns of consumption (yes vs. no) [22]; shared crack pipe (yes vs. no); and rushed crack smoking while in public (yes vs. no). Other exposures and health status included: being a victim of violence, defined as having been attacked, assaulted, or suffered violence in the previous 6 months (yes vs. no); being HIV infected (yes vs. no); and incarceration in the previous 6 months (yes vs. no). All variable definitions are consistent with previous studies [23-25].

Statistical analysis

As a first step, we examined the baseline sample characteristics stratified by reports of experiencing health problems from smoking crack, using the Pearson’s Chi-squared test (for binary variables) and Wilcoxon Rank Sum test (for continuous variables). Fisher’s exact test was used when one or more of the cells contained expected values less than or equal to five. First, we examined the temporal trends of crack pipe acquisition source and health problems, respectively, using univariable GEE models including the calendar dates of 6-month follow-up periods (per period later) as the independent variable.

Since the analyses of experiencing health problems included serial measures for each participant, we used generalized estimating equations (GEE) with logit link, which provided standard errors adjusted by multiple observations per person using an exchangeable correlation structure. We first used bivariable GEE analyses to examine the association between each explanatory variable and experiencing health problems associated with smoking crack. To examine the relationship between crack pipe acquisition source and health problems, we fit multivariable GEE models using a conservative confounding model selection approach [26]. We included all variables that were associated with reporting health problems in unadjusted analyses at p < 0.10 in a full multivariable model, and used a stepwise approach to fit a series of reduced models. After comparing the value of the coefficient of the crack pipe acquisition source in each reduced model, we dropped the secondary variable associated with the smallest relative change. We continued this iterative process until the minimum change exceeded 5%. In order to examine if the estimates differed for women and men, we have also repeated the model using an interaction term for the primary explanatory variable and sex. In order to examine whether the attrition towards the end of the study period biased the estimates, we also conducted a sensitivity analysis where we repeated the analyses among those whose last study visit was earlier than December 2013 (i.e., 1 year before the end of the study period). All p-values are two sided. All statistical analyses were performed using SAS software version 9.4 (SAS, Cary, NC).