High variability of HIV and HCV seroprevalence and risk behaviours among people who inject drugs: results from a cross-sectional study using respondent-driven sampling in eight German cities (2011–14)

Standard probability methods are generally difficult to apply in hard-to-reach populations, where a sampling frame for the targeted population is not available. RDS was introduced by Heckathorn in 1997 as a modified snowball method to recruit hard-to-reach populations [32]. Globally, more than 460 studies from 69 countries applying RDS have been published, and several studies have used RDS to recruit PWID in recent years [33]. Due to their strong social networks and because PWID often buy from and inject drugs with other PWID, RDS worked well as a recruitment method in the majority of studies [34, 35]. RDS works effectively as a sampling method, when four requirements are met [32]: first, participants need to know one another through the network of the group under study. Second, the network needs to be dense enough to attain a sample with sufficient socio-metric depth in order to reach equilibrium. The statistical rationale of RDS depends on the stabilization of the sample composition after a sufficient number of recruitment waves - the point at which the characteristic proportions remain stable, even if the recruitment continues is known, as the equilibrium [32]. The number of waves required to reach equilibrium is again linked to the third requirement: random recruitment must set in at some point to avoid that sampling is limited to a specific sub-group and only reflecting the characteristics of the seed with which the chain began. The tendency to recruit persons who are similar and thereby causing bias in the samples is termed homophily. Fourth, an enabling system to motivate participants to recruit other participants must be in place [36].

In this paper we present descriptive results of the first sero-behavioural study of PWID using RDS performed in Germany. The objectives are to describe basic characteristics of participants in the respective study cities focusing on i) socio-demographic factors, ii) seroprevalence of HIV and HCV, including co-infections, iii) use of health services, and iv) injecting and sexual risk behaviours. Furthermore, we assess whether RDS was effective for sampling PWID in the study cities.

From 2011 to 2014, we recruited PWID using RDS across eight cities in Germany targeting a sample of 200–400 PWID in each city. All cities have a relatively large PWID community and were selected for their geographic and demographic diversity as well as the availability of low-threshold drop-in facility services. Four of the cities - Berlin, Cologne, Munich and Hamburg - have more than one million inhabitants; the four others - Essen, Leipzig, Frankfurt on the Main (Frankfurt) and Hanover- between 500,000 and 700,000.

Eligibility for participation was defined as i) aged 16 or older, ii) self-reported injecting drug use within the past 12 months in the respective city, iii) willingness to take part in an questionnaire assisted-interview and to provide a capillary blood specimen for serological and molecular testing iv) willingness to give informed consent, and v) not having participated in the study previously.

Sampling started with a small number of initial recruits (‘seeds’) in each city, selected by local partners of low threshold drug services to represent a range of characteristics (gender, country of birth, residential area and preferred low-threshold drug service, self-reported HIV serostatus, mainly preferred substance, former experience of sex work and imprisonment). All seeds were selected based on an anonymous list of PWID and their characteristics provided by the local partners. The recruitment expanded through so called ‘recruitment waves’ of peers; after the seeds recruited the first recruitment wave of participants the first recruitment wave continued to recruit the second recruitment wave of participants and so on until the targeted sample size was reached.

In each city we established between one and four RDS study sites in local low-thresholds drop-in facilities, where participants enrolled in the survey and redeemed their coupons. The recruitment coupons were valid for two weeks. Each individual received 10 EUR for participating in the study, and was paid an additional 5 EUR for each eligible drug user they recruited. To ensure anonymity and to track the recruitment process we assigned a unique numeric identifier to each participant. If a seed turned out not to be productive additional seeds were selected if needed to keep the recruitment process ongoing. Recruitment and data collection was conducted by staff of low-threshold drug services who are trained to work with PWID. This recruitment process continued until the end of the scheduled recruitment period which was reached after 7 to 9 weeks.

Staff of the Robert Koch Institute (RKI) conducted a two-day pre-survey training on study design, RDS methodology, standardised interviews, blood sample collection procedures and logistics for the staff of low-threshold drug services in the respective study cities. Eligible PWID had to undergo a questionnaire-assisted interview in German or Russian, wherever Russian-speaking staff was available. We asked questions regarding respondent’s demographic characteristics, their knowledge, attitudes, behaviour and practices as well as their network. Minor modifications were made in the questionnaire throughout the four years while conducting the survey. Therefore, some variables are not available for all cities. The network size was determined by asking respondents how many PWID (fulfilling the inclusion criteria for the study) they know by name who would also know the respondent by name. We also asked how many of these persons they believed they could recruit for the study. The questionnaire was based on a model questionnaire developed by the European Monitoring Centre for Drugs and Drug Addiction (EMCDDA), and additional indicators proposed by the European Centre of Disease Control (ECDC) and the Global AIDS response progress reporting (GARPR) [38‐40]. Dried blood spots (DBS) on filter cards (Whatman #903) were obtained from participants’ capillary blood.. During the pilot phase of the study (cities of Essen and Berlin), DBS testing was validated in the Institute of Virology, National Reference Centre (NRZ) for Hepatitis C, at the University of Duisburg-Essen, which subsequently also performed the regular analyses on serological and molecular markers of HIV and HCV. The Division for HIV and other Retroviruses and the Division for Viral Gastroenteritis and Hepatitis Pathogens and Enteroviruses in the Department for Infectious Diseases at the RKI were in charge of the laboratory testing for the remaining six cities. The study flow and laboratory procedures including possible shortcomings arising from DBS testing are described in detail elsewhere [37, 41]. Prevalence of infection was determined by detection of anti HIV or anti HCV antibodies and detection of molecular markers for HIV and HCV by nucleic acid amplification tests. Pre- and post-test counselling were offered to participants according to international and national recommendations [42].

For evaluation of the sampling method we applied criteria following the guidelines for “Strengthening the Reporting of Observational Studies in Epidemiology for RDS Studies” (STROBE-RDS), a checklist of essential items to present in RDS publications [33]. We provide information about the relationship of respondents with their recruiters and calculated the equilibrium and the number of recruitment waves for five key variables: I. participants’ mean age; II. proportion of male participants; III. proportion of PWID born in Germany; IV. HCV prevalence; and V. HIV prevalence. Furthermore, we describe the level of homophily among the study population. Homophily (Hx) was analysed for the following three outcomes: age, gender and HIV serology. As recommended a graphical representation of the entire recruitment network for all study cities is included. Finally, we assess whether the incentives could motivate PWID to participate in the study. Detailed material of this evaluation is attached in the Additional file.

For data entry we used EpiData 3.0. We applied descriptive statistics by using Stata version 14.0. The crude sample proportions are presented for all cities in Tables 1, 2 and 3. The results are shown as range of proportions (min. and max. values (%)) for the respective study cities. Based on the reported network size of each participant, we used the respondent driven sampling analysis tool RDSAT version 7.1 (http://​www.​respondentdriven​sampling.​org) to define population proportions and variance estimates of each dataset [43]. We included seeds in the analysis. The number of re-samplings to determine bootstrap 95 % confidence intervals (CI) was set to 15,000 to improve the accuracy of the variance estimates and the network size outliers pulled in by 5 %. The enhanced smoothing algorithm type was employed as recommended by Johnston [44]. The RDS estimated population proportions based on the reported network are provided in the (Additional file 1: Table S1). We used RDSAT 7.1 to calculate homophily for all eight data sets. The homophily (Hx) metric is between −1 und 1. In line with Heckathorn’s suggestion we defined any value of Hx ≥ 0.3 as intermediate homophily and any value ≤ −0.3 as strong heterophily [45]. We applied Stata 14.0 to calculate equilibrium and the number of recruitment waves. Equilibrium was attained when the sample distribution from one recruitment wave to the next fell within a discrepancy of less than 2 % [46].

Overall, we recruited a total of 2,079 participants in the multicentre survey, of which two did not meet our eligibility criteria. Most of the interviews took around 45 min to 1 h to complete. In each city except Leipzig (n = 130) a sample size between 200–400 PWID was achieved (see Table 4). In all cities the proportion of female participants ranged between 18.5 and 35.3 %. The median age of participants varied between 35–41 years – except in Leipzig where the median age was 29 years. Accordingly, the proportion of PWID younger than 25 years was higher in Leipzig (26.9 %) compared to the remaining seven cities (2.1–9.0 %). Leipzig was also an exception with regards to country of origin of the participants. Foreign-born participants accounted for 9.2 % in Leipzig and ranged between 16.6 and 30.6 % in the other seven cities. The proportion of participants born in Eastern Europe and Former Soviet Union ranged between 6.3 in Leipzig and 24.7 % in Berlin.

The majority of participants in all cities had completed lower secondary school (40.4–58.3 %) and between 8.5 and 20.8 % had not completed any school. Between 72.3–90.4 % reported currently receiving social benefits/pensions. Furthermore, more than half of the participants in all cities had been homeless at least once in life (52.8–76.9 %). Between 6.8 (in Hanover) and 28.7 % (in Frankfurt) of the participants reported being homeless or staying in homeless shelters as their main residence in the past 12 months (Table 1).

HIV prevalence amongst participants varied between 0 % in Leipzig and 9.1 % in Frankfurt. HCV prevalence (Anti-HCV or HCV-RNA positive or both) ranged from 42.3 in Leipzig to 75.0 % in Hanover (Fig. 1), while HCV viremic infections (HCV-RNA positive) were found to range from 23.1 in Leipzig to 54.0 % in Hanover. HCV-RNA in the absence of anti-HCV antibodies was detected in 0.9 % of the cases in Munich and in 5.4 % of the cases in Leipzig, indicating recent HCV infections before seroconversion. HCV co-infections amongst the HIV positive participants were detected between 60.0 % of cases in Cologne and 100 % in Essen.

Data on utilisation of low-threshold drug services (in the last 30 days) was collected in five of the eight study cities. The proportion of PWID who visited a low-threshold drug service in the last 30 days ranged from 78.7 to 90.5 %. About three out of five participants in each of the eight cities (54.6–88.9 %) reported ever receiving opioid substitution therapy (OST). Currently receiving OST varied between 30.8 in Leipzig and 66.2 % in Cologne. In all cities the vast majority had been tested for HIV during their lifetime (76.6–97.5 %). Undergoing an HIV test in the last 12 month was reported by 43.9 % (Leipzig) to 69.9 % (Munich). The majority of the study population in each city reported ever being tested for HCV antibodies (70.3–96.0 %), while having been tested recently (12-month prevalence) was reported between 28.8 (Leipzig) and 75.4 % (Munich) (Table 2).

The median number of years since first injection was 10 years in Leipzig, 13 years in Berlin and between 16 and 18 years in the remaining cities. In seven cities, the proportion of participants who initiated injecting in the last two years ranged from 3.2 in Hanover to 7.8 % in Berlin. In Leipzig one out of ten (11.1 %) had started injecting in the last two years.

Injecting drugs in the last 30 days was reported by more than three out of four participants in all eight cities (76.0–88.4 %) and daily injection in the last 30 days varied between 17.2 in Munich and 39.1 % in Berlin. In the last 30 days Heroin was the most frequently used substance (all routes of administration) in five cities (Berlin: 83.1 %; Essen: 78.2 %; Cologne: 85.4 %; Hanover: 75.0 % and Munich 56.8 %); while in Leipzig methamphetamine was the most frequently reported substance (67.4 %). In the other seven cities methamphetamine was less common (0.0–6.4 %). Crack was used by a high proportion of participants in three cities (Frankfurt: 71.6 %; Hanover: 58.6 % and Hamburg: 45.9 %) while reported by much lower proportions in the remaining five cities (0.4–3.1 %). Cocaine use was lower in Berlin (37.1 %), Leipzig (17.7 %) and Munich (20.9 %) compared to Frankfurt (44.0 %), Cologne (46.7 %), Essen (60.9 %), Hanover (65.9 %) and Hamburg (79.9 %). Sharing of unsterile needles and syringes (n/s) in the last 30 days was reported by 10.6 % in Hamburg and up to 22.3 % in Hannover among participants who reported having injected during the last 30 days.; only in Frankfurt the proportion was lower (4.7 %). Recent sharing of unsterile equipment like spoons, filters or water for injection with other injectors was reported by 33.0 % in Hanover and Hamburg and by up to 43.8 % in Frankfurt among persons who injected during the last 30 days.

Between 72.9 and 84.6 % of participants reported engaging in sexual intercourse in the last 12 months, of which 44.9 % in Essen and up to 69.1 % in Munich reported not having used a condom at last sexual intercourse. More than half of the participants reported that their last sexual partner was also injecting drugs (51.4 % in Essen and up to 69.4 % in Munich).

Imprisonment (ever) was reported by the majority of participants in all cities (72.8–85.8 %) and median of total duration of incarceration ranged between 2.0 years in Leipzig, Berlin and Munich to 5.0 years in Hanover. Of those who had ever been in prison, between 17.8 % in Leipzig and 39.3 % in Berlin reported injecting drugs while incarcerated. Of these, between 33.0 and 49.4 % had shared n/s or other equipment when injecting during their last incarceration. Tattooing and piercing during imprisonment were reported by 19.4 % of the participants in Munich and up to 34.5 % of the participants in Essen.

The detailed data on substance use and risk behaviours in the last 30 days are shown in Table 3.

Between 7 and 19 seeds started the recruitment process in the respective cities. The sample of Leipzig (n = 130) reached a maximum of eight recruitment waves, while the samples of Frankfurt and Hamburg reached a maximum of 20 recruitment waves (Table 4). Equilibrium and homophily were assessed after the recruitment process had been completed. We reached equilibrium for four of the following five key variables: I. participants’ median age; II. proportion of male participants; III. proportion of PWID born in Germany; IV. HCV prevalence; and V. HIV prevalence in all study cities except in Leipzig. Equilibrium for HIV prevalence was not attained in our sample of Frankfurt, Hanover and Cologne. The results of these analyses are presented in the (See Additional file 2: Figure S1a, Additional file 3: Figure S1b, Additional file 4: Figure S1c, Additional file 5: Figure S1d, Additional file 6: Figure S1e and Additional file 7: Figure S2). Respondents and recruiters had the following relationships: Most participants (54–86 %) received their coupons from their partner or from an acquaintance. Between 14 % of the participants in Cologne and up to 46 % of the participants in Hanover received their coupons from a stranger (Table 4). The reported network size defined as “how many people who injected drugs in the last 12 months do you know (and they know you)” ranged from 0–1400 individuals. We did observe random recruitment among the participants. In Frankfurt, Cologne and Hamburg young participants (<25 years) demonstrated a strong negative homophily, indicating that younger participants only recruited older participants (Hx = −1). Among the female participants, only women in Leipzig demonstrated a negative homophily (Hx = −0.37). In Cologne and Munich HIV positive participants only recruited HIV negative participants (Hx = −1) while HIV negative participants in Cologne demonstrated intermediate homophily recruiting mostly other HIV negative participants (Hx = 0.67). The recruitment chains in Cologne and Hamburg show a very late recruitment of HIV positive PWID in the study sample (See Additional file 8: Figure S4e and Additional file 9: Figure S4h). In those two city samples the recruitment chains have often ended once HIV positive participants were recruited. A graphical representation of the recruitment networks (including HIV and HCV serostatus) in each study city is displayed in the (See Additional file 10: Figure S4a-b, Additional file 11: Figure S4c-d, Additional file 8: Figure S4e-f, Additional file 9: Figure S4g-h).

In all cities we observed a decreased interest in participation in the days following the monthly “social benefit”-payment. We did not experience recruitment challenges such as commercial exchange of coupons, imposters or duplicate recruits.

Risk behaviours (30-day prevalence) like using unsterile paraphernalia (spoons, filters or water), sharing of unsterile n/s and practicing unsafe sex, also with non-IDU (12-month prevalence), was reported in all city samples. In the last two decades, harm reduction interventions like NSP in high income countries, including Germany have led to a remarkable decline in the re-use of unsterile n/s and in HIV incidence among PWID [60‐63]. From our findings we must conclude that either the access to clean n/s is still insufficient and/or that there are still knowledge gaps around transmission and preventive behaviour of HIV and HCV. In other studies, sharing of other paraphernalia still persists at higher levels than sharing n/s among PWID [64‐66]. This was also reflected in our study: sharing other paraphernalia appeared much more common than sharing n/s which was most prevalent in the sample of Frankfurt. In this particular case, the high discrepancy might be related to the high number of crack users. Crack is generally linked to a high consumption frequency, and thus with an increased use of paraphernalia.

Several studies have demonstrated that HCV is more infectious than HIV and a prolonged survival of HCV in syringes and non-syringe injecting paraphernalia has been shown [67, 68]. Not surprisingly and in line with data from other Western European countries, HCV infections therefore appeared to be more prevalent within the study population than HIV [26].

High proportions of the participants reported that their last sexual intercourse was unprotected. The last sexual partner was frequently reported to be an injecting drug user as well, but it was also not uncommon that the sexual partner was a non-PWID. The reported sexual behaviours thus demonstrate the potential risk of spreading HIV through the sexual route. While other studies have demonstrated the higher risk of non-PWID to acquire HIV [69], in Germany, little is known about HIV prevalence among the non-injecting sexual partners of PWID and their potential risk of being a bridge population between PWID and the general population. Therefore further research is needed to better understand the HIV/HCV prevention needs of sex partners of PWID who do not inject drugs themselves.

The study found high rates of incarceration (at least once in lifetime) among the study participants. Unsafe drug use and tattooing/piercing in prison were reported as common practices while in prison, thus constituting important risk factors for the transmission of HIV and HCV. Several studies have shown that not only drug use but also HIV and HCV infections among people in prison are of major concerns in Germany [70, 71]. The provision of harm reduction services in the criminal justice system seems to be insufficient, only one of 186 prisons in Germany offers NSP [72], and there are large variations regarding the availability of OST in prisons across the federal states [70, 73].

Our study shows that HCV and HIV testing rates (12-month prevalence) remained moderate to high in the study populations in comparison to other risk groups, like men who have sex with men [74]. Especially the large variation in the HCV testing rates across the study cities may be linked to the variation of participants who reported undergoing OST at the time of participating, but this needs further investigation.

An important limitation of our study is that it only provides a snap shot of the HIV and HCV epidemiology among PWID in Germany but it does not allow determination of cause-effect relationships. Furthermore, the selection of our study cities was based on the availability of low-threshold drug services in the cities willing and able to participate in the study. Since national representative data on PWID in Germany are not available, we cannot claim that the PWID recruited in the chosen cities are representative for all PWID in Germany.

The application of RDS as a method to recruit PWID was successful. We reached the targeted sample size within the set time frames in all cities except in one, and our primary and secondary incentives seemed appropriate to motivate PWID to enroll in the study. This is in line with other studies, showing that RDS is an effective recruitment method among PWID [35, 75].

The choice of low-threshold drug services as study venues probably increased the willingness of participation. We observed long recruitment chains in seven of the eight city samples, indicating that PWID are well connected via social networks or through making use of the low-threshold drug services. However, we cannot exclude selection bias of the city samples due to oversampling of persons as initial seeds as well as participants who showed communicative competence, well understood the study flow and the background of the study, and who were willing to recruit others for participation. Persons with a lower bonding to the drug scene or less communicative skills might be underrepresented in the samples.

Due to the extensive questionnaire, we refrained from asking the participants how many PWID rejected their coupons during the recruitment process. Reasons for rejecting and the number of PWID who refused are therefore unknown. It is possible that unknown barriers restricted participation and potentially created a bias. Yet we assume that our samples mostly attained adequate socio-metric depth, given that equilibrium was reached for four out of five key variables while reaching up to 20 waves in our samples. However, not all city samples allow robust weighting of results, as equilibrium was not reached in all and the length of the recruitment chains was too short in the city of Leipzig. In this city we recruited PWID in two low-threshold drug services with alternating study operating hours. This seems to have confused some study participants and it is possible that further potential participants were lost due to this fact. Equilibrium could be reached in the other seven samples, showing that bias introduced by the initial non-randomly selected seeds could be eliminated in these samples.

Despite the popularity and the widely applied methodology of RDS as a sampling method it is not known whether RDS can generate unbiased estimates. The assessment of RDS as a method of data analysis (RDS inference) is challenging as it often fails to produce precise results due to the unknown underlying truth [76]. Also, the key variable used to generate the RDS-generated estimates is the reported network size of the participant which may not have been consistently addressed by all interviewers or not consistently understood by all participants, leading to a large range and thereby further uncertainty about the validity of the RDS estimates. In 2012 McCreesh et al. have performed a RDS study with known characteristics in order to assess the precision and relevance of RDS inference and found that RDS failed to reduce bias when it occurred, and even tended to overestimate biased adjusted results [77]. In case biases occur in practice the method is not designed to correct for the sources of biases. RDS-generated estimates should therefore be interpreted with caution and are only shown in the Additional file 1: Table S1.