Study Population and Setting

The study was conducted in three sites in Zimbabwe; Mutare, a small city bordering Mozambique, Victoria Falls, a tourist town bordering Zambia, and Hwange, the site of a large colliery. Women were eligible for inclusion if they were aged >18 years, worked at the study site and reported that they had exchanged sex for money in the past 30 days. The surveys were conducted around 8 months after the introduction of NAC's HIV and STI prevention programme for FSWs.

Sampling

Rapid ethnographic mapping to determine the spatial and social organisation of sex work was conducted at each site to identify key geographic areas where sex work was being conducted, the main typologies of sex work, and inform preliminary estimates of the size of the FSW population.

A bio-behavioural survey was conducted at each site. Respondents were recruited to the survey through respondent-driven sampling (RDS). This variant of chain-sampling can be used to recruit research participants from “hidden” populations. In contrast with other chain sampling approaches, RDS rations the number of recruits per respondent, tends to deploy a larger number of waves of recruitment, and provides financial incentives to recruiters[8]. The aim is to recruit a representative sample of research participants of the hidden population of which the respondents are members. There is ongoing debate in the literature about the extent to which RDS achieves this aim and the conditions under which this may be more or less likely[9-11]. Nevertheless, RDS offers a systematic and theoretically grounded approach to recruiting hidden populations.

The initial recruiters (seeds) were FSWs with close links with other FSWs. The seeds were purposively selected so as to represent a range of ages, geographic areas and sex-work typologies identified during preliminary mapping. Ten seeds were recruited from Mutare and six each from Hwange and Victoria Falls.

Each seed respondent completed a questionnaire, had a finger prick blood sample collected for HIV antibody testing, and was issued with two uniquely identified coupons. The seeds passed these coupons on to individuals meeting the study inclusion criteria, and on presenting a coupon to study staff these individuals were asked to give consent and be recruited to the study. These recruits were themselves given two coupons to refer further peers, and so on. Each respondent was provided with US$5 to cover costs of participation. A further US$2 was provided to the recruiter for each successfully recruited referral. Six waves of coupons were issued. In Mutare, up to 630 individuals could be recruited if all coupons resulted in referral and 378 each in Hwange and Victoria Falls. These “total possible” numbers were roughly double initial estimates of the size of the sex work population at each site. We chose to limit each respondent to two referral coupons to increase the number of waves since we anticipated that estimates would stabilise after 4-5 of our potential 6 waves of recruitment.

Voluntary counselling and testing was made freely available for those participants who wanted to know their HIV results.

Data collection

The questionnaire was developed in English and translated into Shona and Ndebele, the local languages. It was piloted before being implemented to check that all wording was understood and unambiguous. Responses were compiled directly into a computer-assisted survey instrument (QDS™ Nova research Company) by trained female interviewers. The questionnaire data were collected anonymously and included information on socio-demographic and economic characteristics, sexual behaviour, psychological health, physical health, past history of STIs, sexual and social networks, social capital, utilisation of services including HIV testing, provision of antiretroviral therapy (ART), prevention of mother to child transmission (PMTCT) and contraception. Where appropriate we used questions that had been previously used or validated in Zimbabwe or elsewhere[12-15]. The questionnaire took an average of 40 minutes to complete.

Laboratory procedures

Finger prick blood samples were collected[16] by nurse counselors and transported to the National Microbiology Reference Laboratory in Harare. They were tested for HIV-1 antibody in series using AniLabsytems EIA kit (AniLabsystems Ltd, FIN-01720, Vantaa, Finland) with those specimens testing positive also tested using Vironostika® HIV Microelisa System bioMerieux, Inc., Durham NC 27704. Discrepant results were resolved using western blot.

Key variables

We identified a set of primary outcome variables: laboratory-confirmed HIV status, reported HIV testing within the past 6 months among those HIV negative, and whether currently taking ART among those HIV positive. We explored discrepancies in self-reported and laboratory-confirmed HIV status.

We also investigated consistent (“always”) condom use with commercial clients and with steady/permanent non-commercial partners. The question used to assess condom use was “In the past month, how often did you use condoms with your steady partner?” Participant responses were grouped as “Always” and “Not always”.

The variable used to estimate network degree was "how many of those sex workers whom you know personally would you consider recruiting into this study?”

Statistical analysis

We first used unadjusted statistics to describe the population. All of our analyses were stratified by site. As recommended for RDS, we excluded seeds from the analysis.

Our analysis sought to account for the RDS design of the study, while recognising that there remains debate about which methods best achieve this. It is theorised that successive waves of recruits should become more representative of the source population. Individuals may recruit individuals more like themselves than would be expected on average. To explore these issues we assessed whether the prevalence of key outcome measures stabilised over waves of recruitment and whether socio-demographic characteristics of recruiters were associated with characteristics of their recruits.

To estimate outcome prevalences we first used the Stata RDS package [17] The multiplicity estimate of average network degree was used to correct for over-representation of those with large degree in the sample. Bootstrapping with 1000 repetitions was used to obtain percentages and standard errors except for subgroup analyses, for which Taylor linearization was used. Next, we adapted a method proposed by Szwarcwald to estimate predicted prevalences of the key outcome variables in the source population[18]. We used random effects logistic regression for this purpose. We fitted models including weighting for the inverse of participant degree (using the same question as for RDS analysis). We also specified a random effect term for “seed” such that the variance of responses from individuals who were part of the same recruitment chain was modeled to take some account of potential clustering.

Institutional Review Board approval

All participants gave written consent collected according to the principles of Good Clinical Practice. Institutional Review Board approval (IRB) for the study was given by the Medical Research Council of Zimbabwe (MRCZ) The UCL Ethics Committee. The consent form was approved by all IRBs in English and the translations to Shona and Nedebele were approved by MRCZ.

Recruitment

Over six rounds of recruitment 370 FSWs were recruited in Mutare, 237 in Hwange and 229 in Victoria Falls. All 22 seeds recruited women into the study. The recruitment chains are depicted in Figure 1. One batch of dried blood samples from 32 individuals was misplaced by the laboratory and so were not tested for HIV, these individuals were associated with two recruitment chains in Mutare (also shown).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Chains of recruitment in the RDS sites for Hwange (A), Victoria Falls (B) and Mutare (C).

Each circle represents a participant in the survey and the lines represent the passing of recruitment coupons. Orange circles represent women who tested HIV positive and blue circles HIV negative. The green circles in Mutare represent women for whom the HIV status is missing. Finally, the pink circles represent the seeds.

https://doi.org/10.1371/journal.pone.0077080.g001

About 90% of seeds in Mutare and Hwange and all the Victoria Falls seeds were HIV positive (Figure 2). HIV prevalence was higher in the first wave of recruitment before falling to a roughly stable prevalence in subsequent waves. Seeds were more likely than recruits in all waves to report consistent use of condoms in all three sites, but there was no discernible pattern in consistent condom use by wave after the first wave of recruits. There was no consistent trend in HIV testing in the previous 6 months by wave in the three sites. Figure 2 suggests that the number of waves required to reach equilibrium was less than the number of waves in the study for all of the outcomes. Later waves also include more women. There was little evidence for strong association between socio-demographic characteristics of recruiters and outcomes in their recruits. The associations for which there was some evidence of autocorrelation were different across sites. There were no associations in any site that showed strong evidence for autocorrelation between recruiters and recruitees (p values of less than p=0.01).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. RDS validation: Outcome measure estimates over waves of recruitment.

https://doi.org/10.1371/journal.pone.0077080.g002

Characteristics of sampled population

In the three sites FSWs were on average 27-30 years old, and in Mutare and Hwange had generally been long term residents (Table 1). The majority in all sites were widowed, or divorced from a partner (68.3%-75.0%). FSWs in Hwange and Mutare had longer duration in sex work than those in Victoria Falls. Most solicited clients in bars, with significant numbers also soliciting by phone or at home or on the street. In all sites, 17 years was the median age of first sex, and 22-23 years the median age of first commercial sex. Client load was highest in Victoria Falls (median 14 in the last month) and lowest in Hwange (6). FSWs reported a median of 2 commercial partners in the past week, most of whom were new clients. A minority reported anal sex, though among those, 35-58% reported it in the last month. Most participants had at least one close friend who was a sex worker, with 2-4 sex worker close friends most common in all three sites (62.4% of participants in Mutare, 44.5% in Hwange and 58.0% in Victoria Falls).

Most FSWs (72-77%) reported having a “permanent partner” with whom they had had sex 4.5-6 times in the last month (A steady or permanent partner was defined as a husband, or a boyfriend with whom there is an expectation of commitment to a longer term relationship).

Levels of intimate partner violence (IPV) were high with 13-24% reporting physical and 8-23% reporting sexual violence in the last year. IPV was more common within permanent partnerships than with clients. Many FSWs reported they had experienced harassment or abuse from the police. The percentage of FSWs who had been stopped by the police at least once in the previous month was lower in Hwange at 33.6% than in Mutare (60.7%) or Victoria Falls (66.5%). 27-32 percentage points of FSWs reported recent experience of genital sores, ulcers or unusual genital discharge, with about 66% having sought treatment for this.

Outcome variables

The approaches to calculating prevalence estimates for the outcome variables made a small difference to the estimates obtained. For example, in Mutare and Hwange, the weighted HIV prevalence was slightly lower than the unweighted, while in Victoria Falls weighted results were slightly higher. The RDS weighted prevalence and the regression weighted prevalence were similar for each of the outcomes except for the percentage testing for HIV in the previous 6 months (table 2).

The majority of FSWs recruited were HIV positive. HIV prevalence was highest among FSWs recruited in Victoria Falls with RDS-weighted prevalence of 69.6% and a regression weighted prevalence of 70.0%. In Mutare, 50.6% (RDS) - 52% (regression) of FSWs were HIV positive, while in Hwange, this figure was similar at 50.6% (RDS) - 51.8% (regression).

Because we had information on testing, self-reported HIV status amongst FSWs and laboratory confirmed serostatus in the whole sample, we were able to investigate the proportion of FSWs who were aware of their true HIV status and who were aware and on treatment (Figure 3). The category of most concern is FSWs with laboratory-confirmed infection but who are unaware of their status, either because they had never tested, their status had since changed or because they never collected the results of their test. In all three sites, this category was approximately half of all HIV positive participants (percentages are weighted by degree), see Figure 3. Of those who reported in the questionnaire that they were HIV positive, between 51% and 74% reported being on ART (see Figure 4a). This means that of all those with laboratory-confirmed HIV infection, the majority (61.9% to 73.9%) were not on ART (see Figure 4b). Of those who were HIV negative, the majority (65-82%) were also unaware of their status (see Figure 3). This was either because they had not tested in the last 6 months, because they never collected their results or because, in the case of Mutare only, 8 had previously thought they were HIV positive. Only 21-28% of self-reported HIV negative individuals reported having had an HIV test in the last 6 months.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Previous knowledge of HIV status.

https://doi.org/10.1371/journal.pone.0077080.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Percentage of HIV positive female sex workers who knew their status and were on ART (a).

Percentage of all HIV+ female sex workers who reported that they were on ART (b).

https://doi.org/10.1371/journal.pone.0077080.g004

The most common location of last HIV test amongst those who had ever tested was the hospital (38.7% in Mutare, 58.8% in Victoria Falls and 63.2% in Hwange). In Hwange and Victoria Falls, almost all those who reported that they were on ART were receiving treatment at the hospital (42/43 and 41/44 respectively). In Mutare, 55.5% were receiving treatment through the primary care ART clinic while 41.2% were treated at the hospital.

In all sites, FSWs were more likely to report that they always used condoms with commercial partners than with steady/permanent partners. The regression-weighted prevalences of consistent condom use with transactional partners were 65.2% (Mutare), 72.6% (Victoria Falls) and 66.3% (Hwange). For consistent condom use with steady/permanent partners, prevalence was 33.6%, 33.3% and 34.2% respectively.