Data for this study were obtained from the AIDS Care Cohort to evaluate Exposure to Survival Services (ACCESS), an ongoing observational prospective cohort of HIV-positive illicit drug users [8, 9]. Following recruitment through street outreach and the provision of informed consent, participants provide a blood sample and complete an extensive interviewer-administered questionnaire as well as a nurse-administered examination. Follow-ups occur semi-annually. Most of these individuals reside in the Downtown Eastside (DTES) neighbourhood, an area in Vancouver, Canada, with high prevalence of illicit drug use, poverty and homelessness, as well as HIV and hepatitis C infection [8, 9]. Individuals from the ACCESS cohort were aged 18 years or older at recruitment, tested seropositive to HIV-1 and had used non-cannabinoid illicit drugs in the month prior to enrolment. ACCESS has been approved by the University of British Columbia/Providence Healthcare Research Ethics Board.
Data on HIV clinical monitoring and drug-using behaviour were augmented with information on HIV care and treatment outcomes from the province-wide centralized ART dispensary and HIV clinical monitoring laboratory at the British Columbia Centre for Excellence in HIV/AIDS (BC-CfE) [8, 9]. As a result, we had access to a complete profile of CD4 cell count and plasma HIV-1 RNA level for each participant. ART adherence was defined as the number of days ART was dispensed over the number of days an individual was eligible for ART in the previous 6 months; the resulting proportion was dichotomized as > 95% vs. ≤ 95% adherence. Previous studies have validated this method of measuring ART adherence using pharmacy refill data . Measurements of plasma HIV-1 RNA were obtained using the Roche Amplicor Monitor assay (Roche Molecular Systems, Mississauga, Canada.)
This study included all ACCESS participants who were ART-naïve at recruitment, initiated treatment during the study period, and had at least one clinical test measuring CD4 cell count and plasma HIV-1 RNA level during the first 12 months on ART. Our main outcome, time to plasma HIV-1 RNA suppression, was operationalized as the date of the first observation with a plasma viral HIV-1 RNA < 500 copies/mL.
The primary explanatory variable of interest was physician experience, defined as the number of patients that the participant's prescribing physician had previously enrolled in the province-wide HIV treatment registry. This was considered as a continuous variable and, since physicians could become more experienced over time, was fixed for each participant at the time they initiated ART. In British Columbia, antiretroviral prescribing physicians could be located anywhere in the province and not at only one institution. Patients selected antiretroviral-prescribing primary care or specialist physicians in a non-random manner through self-selection or referral from other physicians. Physician experience was fixed as a baseline characteristic at the time the patient initiated ART as we believe that this was the best way to estimate physician experience. We recognized certain behaviours (e.g. drug use activity) could confound this association so these measures were treated as time-updated variables based on each participant's semi-annual follow up visit.
Several variables that could influence the association between physician experience and plasma HIV-1 RNA suppression were also assessed, including age, gender (female vs. male), Aboriginal ancestry (yes vs. no) and Downtown Eastside residence (yes vs. no). Clinical variables included ART adherence (≥ 95% vs. < 95% in first year of treatment), current enrolment in methadone maintenance therapy (no vs. yes), protease inhibitor as part of the first ART regimen (yes vs. no), CD4 cell count at baseline (per 100 cells), plasma HIV-1 RNA level at baseline (per log10 increase) and the year of ART initiation (per more recent year). Illicit drug use measures included daily cocaine use (yes vs. no) and daily heroin use (yes vs. no) and referred to the six-month period prior to the interview. Patients were prescribed antiretroviral therapy consistent with therapeutic guidelines which, beginning in 1996 for all patients, recommended triple combination therapy. While the drugs used over the study period changed markedly over time, consistent with previous analyses, we adjusted for whether protease inhibitors were used in the initial regimen or not as a strategy to adjust for confounding that could occur as a result of regimen type. Given the large number of nucleosides used in the backbone of the ART regimen during the study period, we elected to not adjust for this in the analysis.
We initially examined baseline characteristics of our cohort, including physician experience, and tested for significant differences using Pearson's χ
2 statistic. We subsequently used univariate and multivariate Cox proportional hazards regression to evaluate the impact of the variables considered on the time to plasma HIV-1 RNA < 500 copies/mL. The multivariate model was built using an a priori model building strategy developed by Greenland and colleagues . This strategy aims to produce a parsimonious set of covariates to better estimate the adjusted relationship between a primary explanatory variable and an outcome of interest. It has previously been used, for example, to assess the independent relationship between incarceration patterns and non-adherence to ART among injection drug users . To start, we fitted a multivariate model that included physician experience and the full set of secondary explanatory variables. After noting the value of the regression coefficient associated with physician experience in the full model, we used a manual stepwise approach to fit a series of reduced models, each with one secondary explanatory variable dropped from the full set. Comparing the value of the coefficient in the full model to the value of the coefficient for physician experience in each of the reduced models, we dropped the variable associated with the smallest relative change from further consideration. This iterative process was continued until the maximum change exceeded 5%. This technique has been employed in several studies to estimate the independent effect of an explanatory variable on an outcome of interest .