Effectiveness of interventions for the prevention of HIV and other sexually transmitted infections in female sex workers in resource poor setting: a systematic review

Authors


Corresponding Author Maryam Shahmanesh,Centre for Sexual Health and HIV Research, Royal Free and University College School of Medicine, 3rd floor, Mortimer Market Centre, off Capper Street, London, UK. Tel.: +44 7776 185 572; Fax: +44 207 530 5044; E-mail: bamaryjoon@yahoo.co.uk

Summary

Objective  To systematically review the evidence for effectiveness of HIV and sexually transmitted infection (STI) prevention interventions in female sex workers in resource poor settings.

Method  Published and unpublished studies were identified through electronic databases (Cochrane database, Medline, Embase, and Web of Science), hand searching and contacting experts. Randomized-controlled-trials and quasi-experimental studies were included if they were conducted in female sex workers from low and middle income settings; if the exposure was described; if the outcome was externally measurable, it was after the discovery of HIV, and if follow-up was longer than 6 months. A priori criteria were used to extract data. Meta-analysis was not performed due to the heterogeneity of studies.

Results  Twenty-eight interventions were included. Despite methodological limitations, the evidence suggested that combining sexual risk reduction, condom promotion and improved access to STI treatment reduces HIV and STI acquisition in sex workers receiving the intervention. Strong evidence that regular STI screening or periodic treatment of STIs confers additional protection against HIV was lacking. It appears that structural interventions, policy change or empowerment of sex workers, reduce the prevalence of STIs and HIV.

Conclusion  Rigorous evaluation of HIV/STI prevention interventions in sex workers is challenging. There is some evidence for the efficacy of multi-component interventions, and/or structural interventions. The effect of these interventions on the wider population has rarely been evaluated.

Abstract

Objectif  Revue systématique de l évidence de l’efficacité des interventions de prévention des infections sexuellement transmissibles et du VIH chez les professionnelles du sexe dans les régions à ressources pauvres.

Méthodes  Des études publiées et non publiées ont été identifiés grâce à des bases de données électroniques (base de données Cochrane, Medline, Embase et Web of Science), à la recherche manuelle et aux contacts avec des experts. Des études randomisées contrôlées et quasi expérimentales ont été inclues lorsque: elles ont été menées sur de professionnelles du sexe de régions à faibles et moyens revenus, l’exposition a été décrite, le résultat était mesurable extérieurement, c’était après la découverte du VIH et le suivi était sur plus de six mois. A priori, des critères ont été utilisés pour extraire les données. Une méta-analyse n’a pas été réalisée en raison de l’hétérogénéité des études.

Résultats  28 interventions ont été inclues. Malgré des limites méthodologiques, les évidences suggéraient que la combinaison de: la réduction des risques sexuels, la promotion du préservatif et l’amélioration de l’accès au traitement des IST, réduisait l’acquisition des infections sexuellement transmissibles et du VIH chez les professionnelles du sexe recevant l’intervention. Nous n’avons trouvé aucune évidence solide prouvant que le dépistage régulier ou le traitement périodique des IST conférait une protection additionnelle contre le VIH. Il semble que les interventions structurelles, le changement de politique ou la responsabilisation des professionnelles du sexe réduit la prévalence des infections sexuellement transmissibles et du VIH.

Conclusion  L’évaluation rigoureuse des interventions dans la prévention des IST et du VIH chez les professionnelles du sexe est un défi important. Il existe certaines évidences pour l’efficacité des interventions à multiples composantes et/ou pour les interventions structurelles. L’effet de ces interventions sur la population plus étendue a rarement étéévalué.

Abstract

Objetivo  Realizar una revisión sistemática de la evidencia sobre la efectividad de las intervenciones para la prevención del VIH y otras infecciones de transmisión sexual en mujeres trabajadoras del sexo en lugares con pocos recursos.

Métodos  Se identificaron estudios publicados y no publicados a través de bases de datos electrónicas (Cochrane, Medline, Embase, y Web of Science), búsqueda manual y contacto con expertos. Se incluyeron ensayos controlados y aleatorizados y estudios casi-experimentales, cuando: habían sido conducidos en mujeres trabajadoras del sexo en lugares con un ingreso bajo o medio; se describía la exposición; el resultado podía medirse externamente; era posterior al descubrimiento del VIH; y el seguimiento era mayor de seis meses. Los criterios se utilizaron a priori para la extracción de datos. No se realizó un meta-análisis debido a la heterogeneidad de los estudios.

Resultados  Se incluyeron 28 intervenciones. A pesar de las limitaciones metodológicas, la evidencia sugiere que el combinar una reducción en el riesgo sexual, la promoción del uso del preservativo y un mejor acceso al tratamiento de las enfermedades de transmisión sexual, reduce la adquisición del VIH /ETS en trabajadoras sexuales que reciben la intervención. No encontramos una fuerte evidencia de que la búsqueda regular de ETS o el tratamiento periódico de infecciones de transmisión sexual confirieran una protección adicional frente al VIH. Parecería que las intervenciones estructurales, los cambios en política y el otorgamiento de poderes a las trabajadoras sexuales reduce la prevalencia de ETS y VIH.

Conclusión  La evaluación rigorosa de las intervenciones para la prevención de VIH/ETS supone un reto. Existe evidencia de la eficacia de intervenciones con componentes múltiples, y/o intervenciones estructurales. El efecto de estas intervenciones en un sentido más amplio de la población prácticamente no ha sido evaluado.

Introduction

The HIV epidemic continues to spread; 95% of the estimated 33 million people living with HIV reside in resource poor countries (UNAIDS 2007). Several systematic reviews have studied the effectiveness of HIV prevention strategies at both an individual and population level. One concluded that well designed condom promotion interventions targeting core-groups (groups with high rates of partner exchange) are effective (Merson et al. 2000). A Cochrane review of sexually transmitted infection (STI) control concluded that, with the exception of the trial of syndromic management of STIs in Mwanza(Grosskurth et al. 1995), there is limited evidence from randomized controlled trials (RCT) for STI control as an effective HIV prevention strategy (Sangani et al. 2004). A systematic review of STI prevention interventions found that just over half of 41 interventions identified were effective at reducing STIs (Manhart & Holmes 2005). Authors of a systematic review of structural facilitators and barriers to HIV prevention suggest the need to address macro-social determinants of risk, such as economic policy, migration, gender inequality and sex work legislation (Parker et al. 2000).

Mathematical models suggest that targeting core-groups, such as sex workers (SWs), is an effective way to reduce HIV transmission, particularly in the early and accelerated phase of the epidemic (Aral & Blanchard 2002; Boily et al. 2002). Given the scale of sex work, with incomes equivalent to 2–14% of Southeast Asia’s gross domestic product (Lin 1998), there is an urgent need to identify which interventions are effective in reducing HIV in SWs.

Two important position papers have sought to summarise key strategies for HIV prevention in SWs. One approached HIV as an occupational hazard, advocating harm reduction strategies such as empowering SWs to use condoms and removing structural barriers to safety (Rekart 2005). The other examined strategies to provide STI treatment for SWs and concluded that, using presumptive periodic treatment (PPT) with single dose antibiotics, followed by regular algorithm-driven screening, was likely to be the most effective strategy (Steen & Dallabetta 2003). The effectiveness of either – harm reduction or STI treatment as an HIV prevention intervention – for SWs has not been systematically assessed. This paper presents the findings of a systematic review of the evidence for the effectiveness of HIV and STI prevention interventions, in female SWs, in resource poor settings.

Materials and methods

Inclusion criteria

Any intervention which intended to prevent HIV and STIs through targeting female SWs in resource poor settings, and which was evaluated in an experimental (RCT) or quasi-experimental (controlled but without randomly assigned control groups, or, time-series) study was eligible for inclusion. Study participants were limited to female SWs, defined as women who exchange sex for money or other gifts and commodities. Studies were only included if they reported at least one outcome measure that could be externally validated (Peterman et al. 2000; Zenilman 2005). This included biological outcomes (HIV incidence and/or STI incidence/prevalence), or measurable health outcome (e.g. condom disposal, health service utilisation). Studies were excluded if they targeted male and transsexual SWs, were conducted prior to the advent of HIV, were based in rich industrial countries, if the focus was harm reduction for injection drug use (IDU), the intervention was not adequately described, or if the duration was less than 6 months.

Search strategy

Databases searched are listed in Table 1. Medline and Embase were searched using the Key Mesh terms and text words (in italics): (Prostitution OR prostitut* OR ‘sex work*’) AND (HIV OR HIV infection OR HIV seroprevalence OR HIV OR sexually transmitted disease OR ‘sexually transmitted infection’). The text words were used to search the other databases. A key non-indexed journal ‘Research for Sex Work’ (http://www.researchforsexwork.org) and references of review articles and selected studies were hand searched. Web sites of agencies involved in HIV-prevention (UNAIDS, Family Health International and Population Council) and conference abstracts (through Gateway, National Library of Medicine) were searched. First authors and experts in the field were contacted to obtain information on unpublished work, forthcoming manuscripts and research in progress. Unpublished studies and studies published in non-English languages journals were considered for inclusion.

Table 1.   Databases and years searched
DatabaseYears searchedDate last search performedNumber of articles identified
Cochrane controlled trial register and Cochrane database of systematic reviews 1998–2006July 2006 41
Embase1980–2006June 20061912
Medline1966–2006June 20062175
Web of Science1984–2006July 20062660

Review methods

Titles and abstracts were entered into Reference Manager Professional Version 10 (ISI ResearchSoft, Philadelphia, USA) and screened in three stages using a ten-item checklist (Figure 1). Data from studies that met the inclusion criteria were extracted using a data collection form. Heterogeneity of interventions precluded a summary statistic of effectiveness. Instead, the qualitative results were summarized in tables categorized by main intervention focus and outcome. The interventions are classified according to the conceptual framework presented in Figure 2. The order in which the studies are reported in the tables reflects methodological vigour.

Figure 1.

 Flow chart: Selection of interventions for the systematic review.

Figure 2.

 Conceptual framework for examining the interventions to prevent HIV and sexually transmitted infections in settings.

Results

Intervention characteristics

Of 6788 articles and 1318 abstracts (including duplicates) identified across databases, 1272 were related to HIV and STIs in female SWs in resource poor settings. Hand-searching references and journals, searching websites and conferences and contact with experts identified a further 22 studies. The flow chart (Figure 1) shows that from the relevant articles located, 26 published and two unpublished studies met the inclusion criteria.

Study populations

Twenty-five studies were conducted with SWs; one with couples (transactional and non-transactional sex partners) visiting a motel; two studied interventions with high-risk women associated with mines and truck stops. Four studies evaluated the effect on clients.

Study settings

Sixteen (57%) of the studies were in Africa and the remainder were in Asia (= 8) and Latin America (= 5). Eleven (39%) were in dedicated SWs clinics; the remainder were conducted in brothels (= 7), communities (= 7), motels (= 1) or truck stops (= 1).

Study design

Eleven studies (39%) were RCTs, three of which were cluster-RCTs. Seventeen (61%) were quasi-experimental including uncontrolled before-and-after studies (= 11), studies with a non-randomised control arm (= 3), or a combination of both (= 3).

Interventions evaluated

Seven studies (25%) evaluated interventions to increase condom use. Four (14%) evaluated the efficacy of the vaginal microbicide nonoxinol-9 (N-9). Fourteen (50%) evaluated a combination of a behavioural intervention and STI treatment, six of which were able to separate out the effectiveness of adding the STI treatment component. Three (11%) structural interventions were multifaceted, with improved STI care and an enabling atmosphere for risk reduction either through community mobilisation or political/legal sanction.

Outcomes

Twenty-six (93%) studies assessed changes in incident or prevalent HIV or STIs, of which 12 measured HIV incidence. Other outcomes were verifiable measures of condom use such as provision, disposal or use with simulated clients (= 4), and service utilisation (= 2).

Summary of findings

There were only two RCTs which examined the effect of behavioural interventions combined with condom promotion (Table 2). In Madagascar, the addition of clinic based risk-reduction counselling to community based peer-counselling reduced incident STIs corresponding to increased self-reported condom use (Feldblum et al. 2005). An RCT in Nicaragua found that condoms placed in the rooms or handed to clients were more likely to be used than if made available at reception. Paradoxically, condoms were less likely to be used in the presence of educational material in the rooms (Egger et al. 2000).

Table 2.   Behavioural interventions and condom promotion
Place
Year
Study design
Outcome
Population
Sample
Duration
Response rate/follow-up (%)
Intervention Results*,†
  1. SW, sex worker; STI, sexually transmitted infection; RCT, randomised controlled trial; IEC, information and education campaign; I, intervention arm; C, control arm; OR, odds ratio; adjOR, adjusted odds ratio; RR, risk ratio; adjRR, adjusted risk ratio.

  2. *Where possible RR are calculated from data presented in the papers. Unless otherwise stated RR and OR are quoted for Intervention arm compared with control.

  3. †Numbers in brackets following OR and RR are the 95% confidence intervals.

  4. ‡Percentage of all the motels approached who agreed to participate/percentage of condoms distributed that was retrieved.

Incident HIV
 India 1991–1993
 (Bhave et al. 1995)
Cluster non-random – CT
Incident HIV
Syphilis
Hep BsAg
= 2 areas in red-light district
= 541 SW & 37 brothel owners
6 months
?/97%
(I) IEC, peer risk reduction counselling, condom promotion
(C) No intervention
HIV incidence: (I) 0.05/100 py; (C) 0.16/100 py
HIV incidence RR 0.32 (= 0.002)
Sexually transmitted infection
 Madagascar 2001
 (Feldblum et al. 2005)
Single blind RCT
Incident STIs
= 1000 SW
Stratified by city
6 months
?/90
(I) Clinic-based +  community-based peer risk reduction counselling
(C) Community-based counselling only
Aggregate STIs OR 0.7 (0.5–0.9)
Gonorrhoea OR 0.7 (0.3–1.0)
Chlamydia OR 0.7 (0.5–1.0)
Trichomonas OR 0.8 (0.6–1.2)
 India 1991–1993
 (Bhave et al. 1995)
Cluster non random – CT
Incident HIV
Syphilis
Hep BsAg
= 2 areas in red-light district
= 541 SW & 37 brothel owners
6 months
?/97%
(I) IEC, peer risk reduction counselling, condom promotion
(C) No intervention
Syphilis RR 0.36 (p 0.002)
Hep B sAg RR 0.27 (0.001)
 Singapore 1994–2002
 (Archibald et al. 1994;  Wong et al. 1998;  Wong et al. 2004)
Before and after study
Gonorrhoea rates
Brothel based
= 2737 old SW and 1986 new SW
8 years
100/60
(I) Peer risk reduction counselling, condom promotion, IEC material, and deregistration of brothels with high STI rates
(C) None
(Both arms mandatory STI screen)
Gonorrhoea reduced from >30–45/1000 person months to <5/1000 person months
Gonorrhoea RR 0.11–0.17
 Bali 1997–1998
 (Ford et al. 2000a;  Ford et al. 2002)
Cluster non-random controlled trial
STI incidence
= 7
= 1566 SWs
24 months
?/50% turn over per 6 months
(I) Three risk reduction sessions in 6 months
(C) One risk reduction session in 6 month
(Peer counselling and condom promotion in both arms)
Gonorrhoea OR 0.53 (0.33–0.83)
Chlamydia OR 0.63 (0.40–0.99)
Trichomonas OR 0.91 (0.46–1.81)
STIs reduced in both arms and the differences between high and low effort areas declined over time.
 Thailand, 1994–1995
 (Fontanet et al. 1998)
Cluster RCT
Incident STI
Brothel = 71
Brothels
= 548 SWs
24 weeks
?/26
(I) Addition of female condom
(C) Male condom
Aggregate STI RR 0.76 (0.50–1.16)
Female condoms were used 12% of the time
 Madagascar 2001–2003
 (Hatzell et al. 2007)
Before and after study
Incident STI
Research clinic
= 1000
18 months
?/82%
Addition of female condom to ongoing risk reduction counselling and male condom promotion Aggregate STIs adj OR 0.7 (0.58–0.86)
Female condoms accounting for 20% of the final condom use.
Condom disposal
 Nicaragua 1990
 (Egger et al. 2000)
Cluster RCT with Factorial design
Used condoms retrieved from rooms
= 19 motels
= 6463 couples
24 days per motel 53/48‡A. (I.i) Condoms were placed in the room
(I.ii) Condoms were handed to couple as they registered
(C) Condoms available on demand at reception
B. (I) IEC material in room
(C) No IEC in room
In rooms: Condom retrieval OR 1.3 (1.09–1.75)
To couples: Condom retrieval OR 1.3 (1.03–1.6)
Presence of IEC material: Condom retrieval OR 0.89 (0.84–0.94)

Two non-random-cluster-CT looked at the impact of peer education and condom provision in brothel-based SWs in India (Bhave et al. 1995) and Singapore (Archibald et al. 1994; Wong et al. 1998, 2004). They found reductions in incident HIV and STIs (India) and gonorrhoea (Singapore), which corresponded to increased condom use. Another non-random-cluster-CT compared three risk reduction sessions per 6 months to one and found a lower STI risk in the intervention arm that was not sustained over time (Ford et al. 2000a,b, 2002).

As for female controlled methods (Tables 2 and 3), one cluster-RCT with only 25% follow-up in Thailand (Fontanet et al. 1998) and one longitudinal study in Madagascar (Hatzell Hoke et al. 2007) examined the effect of adding female condoms to ongoing programmes. Both found a shift to female condom use, which only corresponded to a decrease in STI prevalence in Madagascar. Four placebo-controlled-RCTs examined different doses of the vaginal microbicide N-9 delivered in a variety of ways (Kreiss et al. 1992; Roddy et al. 1998; Richardson et al. 2001; Van Damme et al. 2002). They showed either no effect or an increased risk of HIV.

Table 3.   Vaginal microbicides
Place
Year
Study design
Outcome
Population
Sample
Duration
Response rate/ follow up (%)
InterventionResults
  1. SW, sex worker; STI, sexually transmitted infection; RCT, randomised controlled trial; IEC, information and education campaign; I, intervention arm; C, control arm; OR, odds ratio; adjOR, adjusted odds ratio; RR, risk ratio; adjRR, adjusted risk ratio.

Incident HIV
 Benin, Cote d’Ivoire,  South Africa &  Thailand 1996–2000
 (Van Damme et al. 2002)
Triple blind RCT
Incident HIV
Incident STI
Genital lesions
STI clinics and truck stops
= 765 SW
48 weeks
76/68
(I) 52.5 mg: nonoxinol-9 vaginal gel
(C) Identical placebo
HIV incidence adjRR 1.5 (1.0–2.2)
HIV incidence (>3.5 applications per day) adjRR 3.5 (2.1–5.8)
 Cameroon 1994–1996
 (Roddy et al. 1998)
Double blind RCT
Incident HIV
Incident STI
Genital lesions
= 1170
SWs
21 months
65/73
(I) 70 mg nonoxinol-9 film
(C) Identical placebo
HIV incidence RR 1.0 (0.7–1.5)
 Kenya 1987–1990
 (Kreiss et al. 1992)
Un-blinded RCT
Incident HIV
Lesions
Research clinic
= 138 SWs
14–17 months
100/84
(I) 1000 mg nonoxinol-9 vaginal sponge
(C) non-identical placebo
HIV incidence adjRR 1.6 (0.8–2.8)
Genital lesions RR 3.3 (< 0.001).
 Kenya 1996–1998
 (Richardson et al. 2001)
Double blind RCT
Incident HIV
Incident STIs
Research cohort
= 278
19 months
?/69
(I) 52.5 mg nonoxinol-9 gel
(C) Placebo
HIV incidence RR 0.75 (0.37–1.53)
Sexually transmitted infection
 Benin, Cote d’Ivoire,  South Africa & Thailand
 1996–2000
 (Van Damme et al. 2002)
Triple Blind RCT
Incident HIV
Incident STI
Genital lesions
STI clinics and truck stops
= 765 SW
48 weeks
76/68
(I) 52.5 mg: nonoxinol-9 vaginal gel (C) Identical placebo Gonorrhoea adjRR 1.2 (0.9–1.6)
Chlamydia adjRR 1.2 (0.8–1.6)
 Cameroon 1994–1996
 (Roddy et al. 1998)
Double blind RCT
Incident HIV
Incident STI
Genital lesions
= 1170
SWs
21 months
65/73
(I) 70 mg nonoxinol-9 film
(C) Identical placebo
Gonorrhoea RR 1.1 (0.8–1.4)
Chlamydia RR 0.9 (0.7–1.3)
 Kenya 1987–1990
 (Kreiss et al. 1992)
Un-blinded RCT
Incident HIV
Lesions
Research clinic
= 138 SWs
14–17 months
100/84
(I) 1000 mg nonoxinol-9 vaginal sponge
(C) non identical placebo
Gonorrhoea adjRR 0.4 (P < 0.001)
 Kenya 1996–1998
 (Richardson et al. 2001)
Double blind RCT
Incident HIV
Incident STIs
Research cohort
= 278
19 months
?/69
(I) 52.5 mg nonoxinol-9 gel
(C) Placebo
Gonorrhoea RR 1.8 (1.0–3.1)
Chlamydia RR 1.4 (0.6–3.1)
Trichomonas RR 0.8 (0.5–1.3)

Treatment of bacterial sexually transmitted infections combined with behavioural interventions

Three RCTs (Table 4) tested the effectiveness of different STI treatment strategies for SWs, in two of which the primary outcome was incident HIV. The groups in Nairobi (Fonck et al. 2000; Kaul et al. 2002, 2004) and Benin/Ghana (Labbe et al. 2003) looked at the effect of PPT while the group in Cote d’Ivoire (Ghys et al. 2001) tested regular screening for STIs. Neither the Cote d’Ivoire nor the Nairobi studies found a difference in HIV incidence between the arms. In Cote d’Ivoire the follow-up was less than 50%. The study of PPT in Nairobi was the only one that reported significant reductions in bacterial STIs in the intervention arm.

Table 4.   STI screening and treatment combined with condom promotion
Place
Year
Study design
Outcome
Population
Setting
Duration
Response rate/follow-up (%)
InterventionResults
  1. SW, sex worker; STI, sexually transmitted infection; RCT, randomised controlled trial; IEC, information and education campaign; I, intervention arm; C, control arm; OR, odds ratio; adjOR, adjusted odds ratio; RR, risk ratio; adjRR, adjusted risk ratio.

Incident HIV
 Cote d’Ivoire 1994–1997
 (Ghys et al. 2001)
RCT
Incident HIV
Incident STI
SW clinic
= 542
42 months
45/42
(I) Monthly genital examination, microscopy & treatment
(C) Examination and treatment only when symptomatic
(Peer education & condom promotion both arms)
HIV incidence
(I) 5.3/100 py (C) 8.5/100 py (P = 0.5) HIV incidence RR 0.62 (0.5)
HIV reductions in BOTH arms adjRR 0.42 (0.18–0.96).
Women attending 80% of scheduled clinic visits less likely to seroconvert P = 0.04
 Kenya 1998–2002  (Fonck et al. 2000;  Kaul et al. 2002, 2004)Double blind placebo controlled RCT
Incident HIV
Incident STI
= 466 SWs969 person years
89/73
(I) Monthly presumptive treatment with 1 g azithromycin
(C) Placebo
(Peer education & condom promotion both arms)
HIV incidence (I) 4/100 py & (C) 3.2/100 py
HIV incidence RR 1.2 (0.6–2.5)
 Zaire 1988–1991
 (Laga et al. 1994)
Longitudinal cohort
Incident HIV
Incident STI
Dedicated SW clinic
= 531
36 months
?/?
Monthly STD screen & treat
3 monthly voluntary counselling and HIV testing & risk reduction counselling
Peer education & condom promotion
HIV incidence rates dropped from 11.7/100 py to 4.4/100 py
HIV incidence RR 0.4 (P 0.003)
HIV incidence in irregular compared with regular clinic attendees RR 6.2
 Kenya 1985–1986
 (Ngugi et al. 1988, 1996)
Longitudinal cohort &
Non-random CT
Incident HIV
Incident STI
Cohort (1985) = 595:
(I.1) = 91 SW
(I.2) = 67 SW
(C)New recruits (1986) = 205
1–23 months
?/?
(I.1) Peer education, condom promotion, 6 monthly group risk reduction counselling, and individual counselling.
(I.2) As above without individual counselling
(C) Recent recruits before any intervention
(Intervention groups also received periodic STI screening or treatment when symptomatic)
Reported condom use was associated with reduced incident HIV OR 0.34 (0.13–0.92)
Condom use (I.1) 78%, (I.2) 64% and (C) 52%
Sexually transmitted  infection
 Cote d’Ivoire 1994–1997
 (Ghys et al. 2001)
RCT
Incident HIV
Incident STI
SW clinic
= 542
42 months
45/42
(I) Monthly genital examination, microscopy & treatment
(C) Examination and treatment only when symptomatic
[Peer education & condom promotion both arms]
No significant difference between STI incidence
 Kenya 1998–2002
 (Fonck et al. 2000;  Kaul et al. 2002, 2004)
Double blind placebo controlled RCT
Incident HIV
Incident STI
= 466
SWs
969 person years
89/73
(I) Monthly presumptive treatment with 1 g azithromycin
(C) Placebo (Peer education & condom promotion both arms)
Gonorrhoea RR 0.46 (0.31–0.68)
Chlamydia RR 0.38 (0.26 to 0.57)
Trichomonas RR 0.56 (0.40–0.78)
No significant reductions in the incidence of Syphilis
 Benin and Ghana  2001–2002
 (Labbe et al. 2003)
Double blind placebo controlled cluster RCT
Incident STI
18 clusters = 384 SW, = 706 clients
= 252 SW, = 1073 clients
Individual randomisation
= 181 SW
9 months
?/80
(I) Presumptive periodic treatment azithromycin 1 g first month and Ciprofloxacin 500 mg given second and third month. Cycle repeated (C) placebo
(Peer education & condom promotion both arms)
Gonorrhoea RR 0.78 (P =  0.37)
Chlamydia RR 1.9; (P = 0.77)
No significant difference in STI incidence in clients
There was a drop in gonorrhoea in both groups after enrolment.
 South Africa 1996–1997
 (Steen et al. 2000)
Before and after study of SWs and miners
Non-random control group distant from intervention
Prevalence of STIs
Mobile SW clinic
= 407 SWs
= 608 & = 928
miners
9 months
?/32
 (I) Monthly presumptive periodic treatment
 1 g azithromycin to SWs
 (Condom promotion & IEC)
FSWs
Gonorrhoea & chlamydia RR 0.24 (P < 0.001)
Genital ulcer disease RR 0.17 (P < 0.001)
Miners Gonorrhoea & chlamydia RR 0.6 (P < 0.001)
Genital ulcer disease RR 0.22 (P < 0.001)
Inverse relation between attending mine clinic with a genital ulcer and distance to intervention (p for trend 0.002).
 Nicaragua 1995–2004
 (Borghi et al. 2005;  Gorter et al. 2005;  McKay et al. 2006)
Observational study of time trends repeat cross sectional studies
Service utilisation STI prevalence
Community based
= 1500 SW
9 years
50% vouchers utilised/na
(I) 50000 vouchers distributed for free STI treatment at designated clinics. The package consists of presumptive treatment with azithromycin 1 g, screening for syphilis, trichomonas, candida, bacterial vaginosis and cervical cytologyAggregate STIs RR 0.5
 Annual drop: Gonorrhoea (8%), Trichomonas (9%), & syphilis (16%)
Optimal gap for voucher distribution <6 months
 Zaire 1988–1991
 (Laga et al. 1994)
Longitudinal cohort
Incident HIV
Incident STI
Dedicated SW clinic = 531 36 months
?/?
Monthly STD screen & treat 3 monthly voluntary counselling and HIV testing & risk reduction counselling
Peer education & condom promotion
Incidence of all STDs except chlamydia decreased over 3 years (P < 0.01)
 Kenya 1985–1986
 (Ngugi et al. 1988, 1996)
Longitudinal cohort &
Non-random CT
Incident HIV Incident STI
Cohort (1985) = 595: (I.1) = 91 SW
(I.2) = 67 SW
(C) New recruits (1986) = 205
1–23 months
?/?
 (I.1) Peer education,  condom promotion, 6  monthly group risk  reduction counselling,  and individual  counselling.
(I.2) As above without  individual counselling
(C) Recent recruits before  any intervention
  (Intervention groups also  received periodic STI  screening or treatment  when symptomatic)
Annual gonorrhoea rate woman RR 0.23 (P < 0.001)
Decline in men attending STI clinic in intervention site compared to non- intervention site (P < 0.001)
 Peru 1994–1995
 (Sanchez et al. 2003)
Longitudinal cohort
Incident STI
= 917 SW22 months
95/?
  Risk reduction  counselling, condom  promotion and monthly  STI screen and treat Chlamydia adjOR 0.47 (0.28– 0.79.)
Gonorrhoea adjOR 1.16 (0.61–2.3)
Trichomonas adjOR 0.19 (0.09–0.37)
 China 1998–1999
 (Ma et al. 2002)
Longitudinal cohort
Incident STI
= 9666 months
?/53
Risk reduction counselling 2 monthly STI screen and treatGonorrhoea RR 0.3 (0.11– 0.75)
Trichomonas RR 0.14 (0.04– 0.45)
Chlamydia RR 0.24 (0.14– 0.4)
 Cote d’ Ivoire 1991–1997
 (Ghys et al. 2002)
Before & after repeat cross-sections
HIV Prevalence
STI Prevalence
Community based
= 5218
6 years
90/na
Peer education & IEC & condom promotion
Voluntary counselling and HIV testing & STI care
Gonorrhoea RR 0.3 (P < 0.001)
Syphilis RR 0.1 (P < 0.001)
 Benin 1993–1999  (Alary et al. 2002)Before & after repeat cross-sections
HIV Prevalence
STI Prevalence
= 374
= 365
= 591
6 years
?/na
Peer education & IEC & condom promotion
Monthly STI screen & treat
Syphilis adjOR 0.24 (0.09– 0.56)
Gonorrhoea adjOR 0.47 (0.39–0.65)
 Bolivia, 1992–1995  (Levine et al. 1998)Before & after Repeat cross-sections
STI prevalence
Brothel based
= 508
3 years
80/na
Periodic STI screen & treat
 Condom promotion
Clinic based individual counselling
Out reach workers
Gonorrhoea RR 0.6 (P < 0.001)
Syphilis RR 0.4 (0.02)
Genital ulcer disease RR 0.8 (P < 0.006)
 South Africa 1998–2000
  (Williams et al. 2003)
Before & after cross-sections
STI prevalence
SW = 121 & = 93
Stratified random sample of men, women & miners = 899 & = 769
2 years
?/na
Community level intervention: Peers educators from SW, Mine workers & youth.
Condom promotion Train health care workers in syndromic STI management
Monthly presumptive treatment with azithromycin
Miners
Chlamydia adjOR 4.23 (P < 0.001),
Gonorrhoea 2.61 (P < 0.001),
Syphilis (RPR) 1.57 (P = 0.02)
Men Chlamydia adj OR 3.54 (P < 0.001)
Women Chlamydia adj OR 1.88 (P < 0.001)
Syphilis adjOR 2.06 (P < 0.001)
Condom distribution increased three fold
Service uptake
 Tanzania 1993–1994
 (Nyamuryekung’e et al. 1997)
Cluster RCT
Service utilisation
= 7 truck stops
= 330 high risk women
12 months
?/na
Different STI treatment delivery:
(I.1) Primary Health Care worker led outreach clinic twice per week
(I.2) Primary Health Clinic with STI drugs
(I.3) Doctor led outreach clinic every 3 months
(C) Primary Health clinic without STI drugs (standard of care)
Intervention (1) 1.43 visits/woman
Intervention (2) 1 visit/woman
Intervention (3): 1.23 visits/woman
Control: 0.4 visits/woman

A quasi-experimental study of PPT in South Africa found reductions in STIs in SWs after the introduction of the intervention and an inverse relationship between distance from intervention and genital ulcer disease in miners (Steen et al. 2000). Only one of three women were followed-up over the 9 months. An intervention in Nicaragua found that STI treatment vouchers, redeemable at quality approved clinics led to significant drops in STIs in the SWs. This intervention, which in effect provided presumptive treatment to half of the known SWs, showed more substantial reductions in prevalence of STIs if the rounds of voucher distribution were less than 6 months apart (Borghi et al. 2005; Gorter et al. 2005; McKay et al. 2006). A cluster-RCT of STI delivery systems found that high-risk women at truck-stops preferred dedicated outreach clinics to primary health care centres (Nyamuryekung’e et al. 1997).

Four cohort studies examined the effect of regular STI screening, peer education and condom promotion. In Zaire (Laga et al. 1994) and Nairobi (Ngugi et al. 1988, 1996) they examined the effect on incident HIV while in Peru (Sanchez et al. 2003) and China (Ma et al. 2002) the primary outcome was incident STIs. Only the Chinese cohort reported the loss to follow-up, which was 50%. All interventions showed an increase in self reported condom use that corresponded with a reduction in incident HIV and/or STIs.

Four studies from Cote d’Ivoire (Ghys et al. 2002), Benin (Alary et al. 2002), Bolivia (Levine et al. 1998) and South Africa (Williams et al. 2003) compared the situation before and after introducing similar combinations of peer education, condom promotion and regular STI care. In South Africa this was part of a larger intervention that also targeted miners and youth. Only Cote d’Ivoire and Bolivia reported their response rates, which were 90% and 80% respectively. Bacterial STI prevalence dropped in all sites except for South Africa where it paradoxically rose despite increased condom use.

Structural interventions

The best described structural intervention has been the Thai 100% condom programme (Hanenberg et al. 1994; Rojanapithayakorn & Hanenberg 1996; Visrutaratna et al. 1995; Table 5). The countrywide, government led project improved access to affordable STI treatment, and increased condom use through changing social norms and imposing sanctions on dissident sex work establishments. Although there is no control group, various indicators suggest an impact, namely increased condom supply, an 80% reduction in the five major STIs in men (Hanenberg et al. 1994), and a tenfold decrease in STI incidence in new military recruits (Nelson et al. 1996; Celentano et al. 1998). The same magnitude of effect could not be demonstrated in SWs (Kilmarx et al. 1998, 1999).

Table 5.   Structural interventions (Thailand)
Country, YearPrimary interventionStudy populationStudy design
Outcome variables
Results
  1. SW, sex worker; adjRR, adjusted risk ratio; STI, sexually transmitted infection.

Thailand (Hanenberg et al. 1994; Rojanapithayakorn & Hanenberg 1996) 1989–1994100% condom program: Government led supply of condoms to SW establishments
Sanctions for brothels fail to adhere to 100% condom Large scale media campaign targeting male clients to use condoms with SWs
Increased number of STI clinics Free weekly STI tests for SW
Country wideHIV surveillance data from blood donors, pregnant women, SWs, male STI clinic attendees, 21-year-old army conscripts
Statistics on the SW establishments from male STI clinic attendees and annual field surveys
STI data from STI clinics and hospital out patient departments.
Condoms procured by the government and distributed Condoms sold to retailers
Condom use in commercial sex establishment increased from 14 to 94% Government supplied condoms for 70% of SW and private sector for 50% (1992)
Five major STDs decreased by 79% in men
Thailand (Visrutaratna et al. 1995) 1989–1992Pilot for 100% condom program Superstar peer-educators
Condom promotion
Model brothel
Encourage peer pressure amongst brothel owners
Supply free condoms
Cost benefit for brothel owners
500 brothel based sex workers in Chiang MaiBefore and after X-section
Participation in intervention
Before and after behaviour data
Refusal of simulated client w/o condom
Participation up to 100% of identified female SWs
No decline in clients or net income
Before intervention 40% refused sex without condom
After 90% refused simulated client after 2 months and around 80% after one year
Thailand (Nelson et al. 1996; Celentano et al. 1998) 1991–1995100% condom programme
STD treatment at baseline
Incident STDs treated
2417 and 1669 military conscripts in the north of Thailand (random 19–23 year olds as selected by lottery)
90% contribute person time to the analysis.70% followed up 24 months
Comparing two Cohorts
Six monthly surveys
HIV incidence
STD incidence
Sexual behaviour
10 fold decrease in STD incidence between 1991 cohort and 1993 cohort from 17/100 py to 1.8/100 py (P < 0.0001)
HIV incidence from 2.48/100 py to 0.55/100 py RR 0.22 (P < 0.0001)
Brothel visits down from half to 1/3
Inconsistent condom use with SWs down from 14% to 2.5% (P < 0.0001)
Thailand (Kilmarx et al. 1998, 1999) 1991–1994100% condom use program:Brothel based female sex workers over 16 and Thai nationals = 500
16% loss to FU
Cohort study – before and after intervention
HIV incidence STD incidence
adjRR for incident HIV brothel based c.f. non-brothel based 7.3 CI 2.5 to 21.9 (p 0.05)
Brothel based higher HIV incidence throughout

Another well-described structural intervention was the empowerment of SWs in the Sonagachi red-light area (Table 6). Politicised and empowered SWs created an environment conducive to condom use and improved STI care through collective bargaining with structures of power (police, brokers and brothel-owners). Again without a control arm, the impact of the intervention cannot be quantified; however, HIV prevalence among the SWs of Sonagachi remains in single figures compared with prevalences of over 50% reported from similar settings elsewhere in India. A three- to fivefold reduction in prevalent STIs was documented (Chakraborty et al. 1994; Das et al. 1994; Jana et al. 1994, 1998, 2004; Pal et al. 1994; Jana & Singh 1995). There has been one quasi-experimental study comparing Sonagachi with neighbouring brothels; but marked baseline differences, particularly higher client numbers in Sonagachi, limit interpretation of the finding of no difference in STI prevalence (Gangopadhyay et al. 2005).

Table 6.   Structural interventions (Sonagachi)
Country, Year:Primary interventionStudy populationStudy design
Outcome variables
Results
  1. SW, sex worker; STI, sexually transmitted infection; RCT, randomised controlled trial; IEC, information and education campaign; I, intervention arm; C, control arm; OR, odds ratio; adjOR, adjusted odds ratio; RR, risk ratio; adjRR, adjusted risk ratio.

India, Calcutta (Chakraborty et al. 1994; Das et al. 1994; Jana et al. 1994, 1998, 2004; Pal et al. 1994; Jana & Singh 1995) 1991–nowSonagachi (red-light area wide) project:
(i) Empowerment: through self organisation of SWs
(ii) Defining and tackling needs; legal advice, child immunisation, literacy and HIV prevention
(iii) Collective bargaining with police, brokers, and brothel owners in HIV prevention
(iv) Condom promotion
(v) Improved STI treatment
Women living in SonagachiCross sectional surveys
Surveillance data for STIs and HIV
Collective represents 60 000 SWs
HIV prevalence in SWs has remained at <10% which is three to tenfold less than SWs elsewhere in India, e.g. Mumbai. Since 1992 drop in Trepanoma Pallidum hemagglutination assay (TPHA) from 63.5% to 17% (P = 0.001) and trichomonas 15 to 5% (P < 0.001)
Sonagachi vs. NACO (Gangopadhyay et al. 2005) 2003(I) As above
(C) Condom promotion & IEC & peer education
(I) Stratified random sample of 200 brothel based SW in Sonagachi
= 173 (87% response rate) (C) All SW from neighbouring area = 169 (65% response rate)
Non-random CT
Outcome measures
Behaviour and a combined clinical and laboratory diagnosis of STIs
Significant baseline differences between intervention and control arms
Sonagachi women had significantly better health seeking behaviour and optimism scores
No difference in bacterial STIs.

There has been one controlled study of a structural intervention combining elements from both the group empowerment model of Sonagachi and the political sanctions of Thailand (Table 7). The study, conducted in 68 brothels in two cities in the Dominican Republic, compared the addition of regional policy change, which penalised the brothel management for failing to enforce 100% condom use, against an intervention that combined SW solidarity, environmental cues for condom use, improved STI care, and self-regulation of the brothels. There were greater reductions in STI prevalence and a corresponding increase in likelihood of rejecting unsafe sex in the city where, the 100% condom use policy was in force. Condom use increased in individual SWs and was associated with reduced incident STIs in both arms of the study. However, the likelihood of a brothel adhering to the 100% condom use programme was 10% greater in the policy change area (Kerrigan et al. 2003, 2006).

Table 7.   Structural interventions (Dominican Republic)
Country YearPrimary interventionStudy populationStudy design
Outcome variables
Results
  1. SW, sex worker; STI, sexually transmitted infection; RCT, randomised controlled trial; IEC, information and education campaign; I, intervention arm; C, control arm; OR, odds ratio; adjOR, adjusted odds ratio; RR, risk ratio; adjRR, adjusted risk ratio.

Dominican Republic (Kerrigan et al. 2006) 1999–2000 (C) Brothel-based intervention:
(1) Solidarity through regular meetings between SWs & management
(2) Environmental cues for condoms
(3) Improved clinical care through liaison & training for the government’s mandatory monthly STI screens
(4) Monitoring and reporting the performance of the brothels
(I) In addition to 1–4 above: Policy and regulation: regional policy made condom use between clients and SWs mandatory & implementation the brothel owners and management’s responsibility. This policy was enforced through a mixture of support and sanctions.
(C) Santa Dominga (34 brothels) (I) Puerto Plata (34 brothels)
Participatory observations at all brothels
Cross-sectional survey before and after intervention = 200 SWs per city (recruited from the mandatory government STI clinics-every third SW on a designated day)
Response rate 95%.
Before and after X-sectional studies
Non-random comparison: STI
Condom use
Rejection of unsafe sex
Number of establishments without STIs per month
Exposure to intervention
Decrease in STIs was only significant in intervention arm
(I) adjOR 0.50; CI 0.32 to 0.78
(C) adjOR 0.60; CI 0.35 to 1.03
(I) Increased proportion of brothels with no new STIs OR 1.20; CI 1.0 to 1.31
(I) Increased rejection of unsafe sex adjOR 3.86; CI 1.96 to 7.58
Observed adherence to the intervention was significantly associated with reduced STIs adjOR 0.52; CI 0.35 to 0.78
Adherence increased at an individual level in both arms (P < 0.001)
Adherence at an establishment level only increased in intervention arm adjORg 1.2; CI 1.11 to 1.3

Discussion

To the best of our knowledge, this is the first systematic review of HIV and STI prevention interventions in female SWs in resource poor countries. Although there were a considerable number of descriptive studies of sex work in resource poor settings, we only identified 28 that evaluated interventions with externally measurable outcomes. Less than half of these were RCTs, the robustness of which was compromised by very high attrition rates. We identified four broad categories of intervention: behavioural interventions with condom promotion, addition of vaginal microbicide, addition of STI treatment, and structural interventions. The small number of methodologically rigorous studies reflects the considerable challenges of studying this group. The diversity in type of intervention, study design, and outcome measures made calculation of a summary measure of effectiveness inappropriate.

What interventions worked?

Risk reduction counselling coupled with condom promotion reduced HIV or STI risk or increased condom use in all the five studies that tested this hypothesis (Bhave et al. 1995; Egger et al. 2000; Ford et al. 2003; Wong et al. 2004; Feldblum et al. 2005). Additional support for the effectiveness of condom promotion comes from observed reductions in HIV incidence in both arms of STI treatment RCTs (Ghys et al. 2001; Kaul et al. 2004) and the relationship between increases in self reported condom use and reductions in infections in two of the cohorts (Ngugi et al. 1988; Laga et al. 1994). Despite the methodological limitations of these studies, the consistency of the direction of change, the dose response, the association between participation in the intervention, self-reported condom use and reduced infection rates, and biological plausibility suggest that this is an effective strategy.

Two studies assessed female condom promotion and showed an increase in female condom uptake (Fontanet et al. 1998; Hatzell Hoke et al. 2007). There is only weak evidence from the before and after study of related reductions in STI incidence (Hatzell Hoke et al. 2007). N-9 did not reduce HIV incidence and a meta-analysis of all N-9 studies found a relative risk for HIV of 1.12; CI 0.88–1.42 (Wilkinson et al. 2002). Trials of other microbicides are under way.

The two RCTs of PPT and regular screening of STIs were unable to prove the hypothesis that STI treatment in SWs will reduce HIV acquisition (Ghys et al. 2001; Kaul et al. 2004).The failure of two of the RCTs to show an effect of presumptive treatment or regular screening on STI rates may be explained by a type 2 error (loss of power from sizable reductions in STI rates in the control as well as intervention arms) (Ghys et al. 2001; Labbe et al. 2003). The RCT that did show an effect of presumptive treatment on STI rates detected this sample size problem and lengthened the enrolment period accordingly (Kaul et al. 2002, 2004).

One quasi-experimental study suggests that increasing the interval between rounds of PPT may lessen its impact on STI prevalence (Gorter et al. 2005). Other studies also suggest that the effect of presumptive treatment is short lived (Behets et al. 2005; Cowan et al. 2005a). The effectiveness of the Nicaraguan voucher system in enabling nearly half of the SWs countrywide to access STI health services (Gorter et al. 2005) and the preference for outreach services in truck-stops (Nyamuryekung’e et al. 1997) suggests that innovative outreach services may improve the coverage of dispersed and clandestine SWs.

100% condom use programme was a countrywide multi-component intervention that sought to increase condom use, reduce the number of commercial sexual encounters and improve provision for STI treatment. It is impossible to disentangle the relative importance of the different components of the intervention from each other, or secular trends. The observational data from the Sonagachi Project suggest that empowering SWs may reduce their HIV and STI risk. However, the reproducibility of this approach remains unproven (Basu et al. 2004). The Dominican Republic attempt to disentangle the relative effects of policy and empowerment suggests that while pressure to create ‘model brothels’ through self-regulation resulted in a decrease in STIs, there was a greater effect in the city where the ‘model brothel’ was enforced through policy (Kerrigan et al. 2003).

Potential biases of the review process

Given the heterogeneity of the study designs, a funnel plot for publication bias was not done; however there is likely to be publication bias. While some RCTs were unable to show an effect, almost all quasi-experimental studies reported statistically significant findings in favour of the intervention being tested.

Interventions not published in peer reviewed journals are under-represented (Hopewell et al. 2007). Even within the grey literature there is potential for selection bias, as interventions funded or sanctioned by the larger donors are more likely to be accessed through UNAIDS, FHI or Population Council reports and best practice publications. As in all systematic reviews, despite extensive hand searching, there is still the possibility of indexing bias (Hopewell et al. 2002).

The review was restricted to evaluated interventions that had externally validatable outcome measures of effectiveness. This may have excluded less rigorously evaluated but nevertheless important and potentially effective interventions (Wilson et al. 1990; Chipfakacha 1993; Asamoah-Adu et al. 1994; Nairne 1999; Ganasinghe 2000; Campbell & Mzaidume 2001). However, self reported measures of, for example, condom use are unreliable and were therefore excluded from this review (Peterman et al. 2000; Zenilman 2005).

A limitation of this systematic review is that only interventions that involved women who exchange sex for gifts or money could be included. This means that potentially effective interventions with high risk women such as bar workers, who were not explicit about the transactional nature of their sexual behaviour, were excluded (Riedner et al. 2006).

Potential biases of the studies and other methodological issues

Properly conducted RCTs are the best way of assessing the effectiveness of health care interventions. In this review, fewer than half of the studies were RCTs and only just over half had any controls. The effect size of the quasi-experimental studies is greater than the RCTs, and several RCTs showed no effect.

Studies primarily targeted professional SWs working in brothels or red-light districts. In reality much sex work takes place in less organised settings, which would affect the broader applicability of the findings. Forty percent of the studies recruited participants from an STI clinic that had been specifically established for SWs. Participants in a disease prevention intervention may not be representative of all SWs; they may be more adherent, more visible and more likely to have received HIV prevention information. This may lead to participation bias. Analysis of the Kenyan cohort as an open cohort found a drop in incidence of HIV over time, which the investigators attribute to secular trends and the cohort attracting lower risk SWs with the passage of time (Baeten et al. 2000). In addition, half of new HIV infections occurred within the first 6 months of joining the cohort, and 75% occurred within the first year (Baeten et al. 2000), which may reflect the selection of higher risk individuals early in the cohort’s life (Beyrer et al. 1996). These are alternative explanations for the drop in HIV incidence, detected in two of the cohorts, after introducing the interventions (Ngugi et al. 1988; Laga et al. 1994).

Sex workers are highly mobile. Over half of the studies that followed SWs reported attrition rates as high as 75%. This compromises the validity of the resultant outcome (Beyrer et al. 1996; Beyrer & Nelson 1997). In one cohort, if all the women lost to follow-up were non-compliant, the 50% increase in condom use reported would be a more modest 10% (Ma et al. 2002).

To minimise recall and social desirability bias, only studies with reproducible outcomes were included (Peterman et al. 2000; Zenilman 2005). However, for a study to be powered to detect change subsequent to an intervention there needs to be a low baseline prevalence and high incidence of the outcome of choice. In at least three RCTs, the lower than expected infection rates after enrolment may have resulted in a type 2 error contributing to the lack of effect found (Ghys et al. 2001; Labbe et al. 2003; Kaul et al. 2004).

The HIV prevention in SWs is a core group intervention, STIs are a communicable disease and any intervention to reduce STIs may have a herd effect. Thus, any evaluation of STI and HIV prevention should also consider impact at a population level. None of these studies looked at HIV incidence in the bridge or general population, and only one out of the four studies that measured the effect of the intervention on STIs in clients found an effect.

Sex workers are a heterogeneous group. Factors such as relative number of the SWs in relation to the bridge and general population, as well as the structure of the sexual networks and stage of the epidemic influence the extent to which they behave as a ‘core’ group (Lowndes et al. 2002; Cowan et al. 2005b; Nagot et al. 2005) Given the small number of effective studies, we were unable to explore the relationship between phase of the epidemic and effectiveness of the intervention.

Given the complexity and multifaceted nature of these interventions, indicators of exposure to the intervention would have assisted interpretation. Unfortunately the indicators to measure exposure commonly reported, e.g. number of clinic visits or educational events attended, are also measures of adherence. In the absence of controls, finding an association between these measures of exposure and outcomes may be confounded by other factors associated with being an ‘adherent’ participant in disease prevention. Data collection methods can behave as interventions, e.g. behavioural questionnaires could reinforce the behaviour message or social desirability bias. Equally, legally imposed ‘model brothels’ may encourage management to implement additional, undocumented, interventions.

Few of the cluster-controlled trials accounted for inter-cluster correlation in either the power calculation or in the analysis stage. This could result in a greater measure of effect than if clustering had been considered (Hayes et al. 2000). In three studies, only two areas were compared, so we cannot exclude residual confounding or chance (Bhave et al. 1995; Wong et al. 1998; Kerrigan et al. 2006).

Conclusions

The methodological challenges to conducting studies in such a clandestine and mobile group suggest that caution should be exercised when interpreting the results. None of the RCTs showed an impact on HIV incidence. However, the observational data suggests that there is some evidence for the effectiveness of risk reduction counselling and condom promotion.

There is evidence that condom promotion and regular access to improved STI management reduces STI burden in SWs. There’s no unequivocal evidence that intensive STI management in SWs has any additional benefit in HIV prevention. Innovative STI delivery methods, such as vouchers, may improve coverage.

There is some evidence that policy support for SW interventions as well as strategies that empower the women improve coverage, acceptability and adherence to the intervention. There is still uncertainty around the efficacy of STI treatment in HIV prevention for SWs, what is the best STI treatment strategy, what components of structural interventions work, and what the potential negative ramifications of targeting SWs are (e.g. stigma, violence, and driving sex work underground or into areas less identified with ‘professional’ sex work, such as bars and dance halls). In addition, there is limited data available on the wider public health benefits of targeting SWs. There is a need to explore the effectiveness of comprehensive HIV care packages for SWs, new microbicides, HSV-2 prophylaxis and pre-exposure prophylaxis. Evaluations of interventions that reach community-based SWs who work outside brothel-based settings and red-lights districts are required.

Acknowledgements

We are grateful to Sarah Walker and Helen Weiss for the critical feedbacks on the methodology of the systematic review, and to Michel Alary, Gina Dallabetta, Richard Steen and Theresa Hoke for guidance and sharing unpublished data. This work was supported by a fellowship grant from the Wellcome Trust to the first author.

Ancillary