Our formative research taught us that particularly for Brussels very few studies were done in the past among the PWID population. Based on expert opinions from key stakeholders including PWID and service providers, the PWID population in Brussels was estimated at between 500 and 1000, but apart from treatment centers and other community-based supportive groups there were no users groups of PWID and the connections between PWID were supposed to be weak. As standard probability methods are generally difficult to apply to hard-to-reach populations such as PWID [13] and because the hidden nature of injecting drug use decreased the potential for success of other survey strategies such as time-location sampling or simple random sampling, respondent driven sampling (RDS) was chosen as sampling method. It was decided to include also high risk opiate users (HROU) in the study. They allowed us to reach PWID with a more diverse profile. The HROU, defined as people who used opiates at least once a week for six months in the last year without medical prescription, were seen as a bridge between different PWID sub-populations or individual PWID.
Between 1 February 2019 and 26 April 2019, we recruited PWID and HROU in Brussels using the RDS methodology. Eligibility criteria were: (i) self-reported injecting drug use in the last year and/or opiate use at least once a week for six months in the last year without medical prescription, (ii) aged 18 or older, (iii) having lived or used drugs in Brussels, (iv) having received a coupon from someone who participated already or being selected by one of the participating organizations, (v) willingness to answer to a questionnaire, (vi) willingness to undergo a rapid HCV-test and (vii) not having participated in the study before. Interviews were conducted in Dutch, French or English or in case of native Arabic and Russian speaking interviewers also in these two languages.
Sampling started with seven seeds, selected by low-threshold treatment centers or needle exchange programs in Brussels. The recruitment followed the usual RDS sampling method of chain referral, where seeds recruited ‘first wave’ participants and ‘first wave’ participants recruited ‘second wave’ participants, and so on, until the end of the study period. Interviewers checked for track marks to verify recent injecting behavior. If marks were not found or the user confirmed no injecting practices, the subject had to demonstrate detailed acquaintance with opioid preparation for other modes of administration. At the end of the interview, every respondent received three recruitment coupons. The recruitment coupons were valid for one week, although expired coupons were not rejected in practice. The limitation of one week was mainly applied to encourage participants not to wait too long before introducing the coupons to potential new participants. 90.4% of all participants arrived within this time limit. Because an adequate strategy on non-response or refusals was lacking, no further information is available on the actual response rate, i.e. the number of people who refused to participate. To ensure tracking of subsequent waves, the coupons had a unique identifying number to link the recruiter to the recruited person [14]. Each individual received €5 for participating in the interview and the HCV-test and €10 for each eligible participant they recruited, with a maximum of 3 new recruits per participant. To avoid doubles, we encoded individuals with their initials, sex and date of birth, and used custom-developed coupon manager software. If seeds turned out not to be productive or the recruitment stopped, we recruited additional seeds. With this strategy, we responded to poor recruitment during data collection without negatively impacting the theoretical and methodological requirements [15]. A team of multi-ethnic and multilingual nurses, conducted structured interviews and HCV-testing in a mobile unit. The interviews were immediately imputed in a questionnaire with all mandatory questions, developed in LimeSurvey [16] and saved on a central server. As a result the sample did not contain missing values.
During the study period, participants could come every day (7/7) between 17.00 and 21.00 to this mobile unit. Initially it was decided to park the mobile unit on fixed places, every day of the week another place where PWID could come and participate in the study and to repeat this patterns week after week. However, it turned out that the PWID were less mobile than expected and after ten days we changed the strategy and parked the mobile unit for several days in a row on a location known to PWID where we stayed until no new recruits arrived anymore. In total, people were interviewed and tested on nine different locations in the city.
At the onset of the interview, participants were informed about the study aims and were asked to sign a consent form. We asked questions on socio-demographic characteristics, drug use and injecting history, HCV risk behavior, previous HCV testing, previous and current HCV treatment, previous and current drug use treatment, number of overdoses, network size and the relationship with their recruiter. To distinguish between respondents who were rather in contact with PWID and others who were more in contact with HROU, network size was defined as (i) “How many people in Brussels who have injected opiates such as heroin, morphine, opium or fentanyl in the last 6 months do you know by name who know your name?”, (ii) “How many people in Brussels who have used opiates such as heroin, morphine, opium or fentanyl without prescription and without injecting and who have injected drugs in the last 6 months but not opiates do you know by their name who know your name?”, (iii) “How many people in Brussels who have injected drugs in the last 6 months but not opiates do you know by their name who know your name?” or (iv) “How many people in Brussels who have used opiates such as heroin, morphine, opium or fentanyl without medical prescription without injecting in the last 6 months do you know by their name who know your name?”. The questionnaire was pre-tested and based on the Drug Related Infectious Disease toolkit, developed by the European Monitoring Centre for Drugs and Drug Addiction (EMCDDA) [17], the WHO guidelines for RDS [18], the Belgian HIV register (for the unique person’s identifier) [19], with some additional questions from the Australian national drug strategy household survey [20, 21] and a HIV/HCV risk behavior survey from Yale University [22]. Prevalence of HCV was tested with the whole blood InTec Rapid HCV antibody Test. We did not have the possibility to confirm HCV antibody positive results with a PCR HCV RNA test as recommended by the national guidelines [6]. Respondents were offered pre- and post-counselling according to national and international guidelines [6].
Based on the aforementioned estimated number of PWID in Brussels and with an estimated HCV-prevalence of 40% [23], a CI of 95%, z = 1.96, a design effect of 1.5 [1] and a standard error of 0.05, we reached a sample size on which we applied a finite population correction [24], resulting in a required sample size of 269 PWID. Seeds were included in every part of the analysis. Selection bias was examined for age and sex by comparing the participants’ profile with the profile of patients in low-threshold treatment centers and needle exchange programs in Brussels and within the sample through t-tests and χ2-tests.
RDSAT version 7.1.46 was used to verify the stability of the study sample. The homophily metric (Hx) was between − 1 and 1, with Hx < − 0.3 and Hx > 0.3 used for the identification of significant biases, as is conventional in RDS analysis [14, 25]. We used RStudio version 1.1.442, package ‘RDS’ [26] to calculate equilibrium for sex, mean age and HCV prevalence, which was attained when the sample distribution from one recruitment wave to the next fell within a discrepancy of less than 2% [24], For respondents who answered that their personal network size was 0, degree values were imputed using the weighted mean of the non-missing degrees, calculated with Gile’s SS [26]. Since there is no consensus on which estimator is optimal [27], we give the unweighted sample average for HCV antibody prevalence among PWID as well as the RDS-II and RDS-SS weighted estimate. We did the graphical representation of the complete network in NetDraw version 2.160. For reasons of completeness, descriptive statistics for HROU will be presented as well. The reporting of this study conforms to the STROBE guidelines for RDS studies [28].