J Epidemiol Community Health 67:491-497 doi:10.1136/jech-2012-202035
  • Research report

Social participation and drug use in a cohort of Brazilian sex workers

  1. Sheri A Lippman2
  1. 1Division of Epidemiology, School of Public Health, University of California, Berkeley, California, USA
  2. 2Division of Prevention Science, Department of Medicine, Center for AIDS Prevention Studies, University of California, San Francisco, California, USA
  3. 3Reprolatina, Campinas, Brazil
  4. 4Department of Health, Behavior and Society, Johns Hopkins School of Public Health, Baltimore, Maryland, USA
  1. Correspondence to Hannah Hogan Leslie, Division of Epidemiology, School of Public Health, University of California, 101 Haviland Hall, Berkeley, CA 94720-7358, USA, Hannah.leslie{at}
  • Received 4 October 2012
  • Revised 25 January 2013
  • Accepted 26 February 2013
  • Published Online First 16 March 2013


Background Structural interventions focused on community mobilisation to engender an enabling social context have reduced sexual risk behaviours among sex workers. Interventions to date have increased social participation and shown an association between participation and safer sex. Social participation could modify risk for other health behaviours, particularly drug use. We assessed social participation and drug use before and after implementation of a clinical, social and structural intervention with sex workers intended to prevent sexually transmitted infections/HIV infection.

Methods We followed 420 sex workers participating in the Encontros intervention in Corumbá, Brazil, between 2003 and 2005. We estimated the association of participation in external social groups with drug use at baseline and follow-up using logistic regression and marginal modelling. Follow-up analyses of preintervention/postintervention change in drug use employed inverse probability weighting to account for censoring and were stratified by exposure to the intervention.

Results Social participation showed a protective association with drug use at baseline (1 SD higher level of social participation associated with 3.8% lower prevalence of drug use, 95% CI −0.1 to 8.3). Among individuals exposed to Encontros, higher social participation was associated with an 8.6% lower level of drug use (95% CI 0.1 to 23.3). No significant association was found among the unexposed.

Conclusions A structural intervention that modified sex workers’ social environment, specifically participation in external social groups, was associated with reduced drug use. These findings suggest that sexual risk prevention initiatives that enhance social integration among marginalised populations can produce broad health impacts, including reductions in drug use.


Global efforts to prevent the spread of HIV and other sexually transmitted infections (STI) increasingly include structural interventions: approaches that target the social, economic and political environments that shape individual vulnerability to effect broad and sustained risk reduction.1–7 Structural interventions for groups traditionally excluded from power have sought to strengthen community ties, extend social support networks, build partnerships and otherwise integrate and empower marginalised groups highly vulnerable to HIV/STI.2 ,8–11 Sex workers are one such group and have been a focus of HIV/STI prevention efforts.12 Beginning with the pioneering community-based Sonagachi project in India, a number of structural interventions that focus on community mobilisation and social inclusion have been implemented among sex workers.1 ,13 These interventions have reduced STI or HIV infections,2 ,4 ,14 increased condom use with clients2–4 ,15 ,16 and increased community development indicators.2 ,17 One component of community development that may be particularly important to reduce marginalisation is increasing capacity to interact beyond the immediate social group, defined in the social capital literature as extracommunity ties/linkages or bridging social capital and termed here social participation.18 ,19 Existing studies that explicitly measure participation in external (non-sex worker) social groups have documented a cross-sectional association between this participation and condom use17 ,20 as well as a longitudinal increase in social participation among those exposed to structural interventions.2 ,17 ,21 Structural interventions that meaningfully alter social participation would plausibly affect other risk behaviours for sex workers, but this has not been examined to date.

Substance use operates as a mutually reinforcing risk behaviour with sex work: drug users may participate in sex work to support their drug use, while sex workers may use drugs as a coping strategy.22 ,23 Beyond individual factors, the social environments of drug use and sex work are intertwined; multiple studies document their overlap within venues such as clubs and brothels.24–28 In addition to direct negative effects on health, drug and alcohol use are associated with physical and sexual violence among sex workers, as well as with greater risk-taking behaviour.23 ,24 Studies in disparate settings have identified associations between use of drugs such as crack and unprotected sex.26 ,29 ,30 Despite the increasing emphasis on both structural interventions and drug use harm reduction among sex workers, there are no studies to date on how structural interventions with sex workers might impact drug use or its associated harms. Furthermore, while existing theory and research suggest that social integration is inversely related to risk behaviour,31 research on drug use has identified an increased risk of substance initiation and use associated with high drug use in family and peer networks.32 To date, limited research addresses whether social participation in external groups, as distinct from the immediate social group, affects drug use.32–34

This paper presents results from Encontros, a multilevel structural and clinical STI/HIV prevention intervention with sex workers in Brazil. The primary objective of Encontros was to decrease STI/HIV incidence and improve condom use by creating a more enabling social environment for protective behaviours (including increased cohesion and social participation).20 Previous analyses found that Encontros increased social participation and consistent condom use with regular clients.2 We hypothesise that social participation as a component of a supportive environment can reduce harmful behaviour. In this analysis we examine how the intervention and social participation shape change in drug use (figure 1).

Figure 1

Conceptual model.

Materials and methods

Study procedures

The Encontros intervention and study procedures are fully described elsewhere.2 ,20 The study took place from 2003 to 2005 in Corumbá in the Pantanal region of Brazil. Corumbá is a city of 100 000 permanent residents that borders Bolivia and supports an active commercial sex economy for seasonal fishermen, eco-tourists and local residents; cocaine is widely available owing to Corumbá's location along a drug trafficking route.28 Individuals who identified as sex workers, spoke Portuguese or Spanish, and did not plan to relocate permanently from the study area in the month following recruitment participated in the programme. Participants were recruited using convenience and snowball sampling, facilitated by peer educators drawn from local sex work establishments. Participants received free transportation to the study site, condoms and small gifts at each visit. Full participation included enrolment and four scheduled follow-up visits throughout the year. Follow-up visits included administration of a structured questionnaire, STI/HIV prevention counselling and clinical exam and specimen collection for STI testing.


The Encontros intervention employed a combined clinical and social-structural approach to engage sex workers on individual, interpersonal and community levels. Project activities were developed within a human rights framework with the goal of reinforcing the Brazilian national strategy of HIV prevention through destigmatising sex work, reinforcing social inclusion of all citizens and emphasising the provision of universal high-quality care. Intervention activities were based at clinics and within the community. Community activities included skills-based workshops for participants and cultural showcases to publicly display the workshop results. Participation in political forums, advocacy campaigns and the establishment of a sex worker association were also part of the Encontros intervention process.

The research was approved by the Ethics Committee in Mato Grosso do Sul, Brazil; the Brazilian National Committee of Ethics in Research; the Institutional Review Board at the Population Council and the Committee for the Protection of Human Subjects at the University of California, Berkeley.


Social participation was measured through a series of yes/no questions regarding involvement in groups in the broader community (eg, participation in a church, club, neighbourhood association or volunteer organisation).20 Intervention activities were not included in this measure. Baseline measures were used in the cross-sectional analysis, and the average of participation measured at interim visits (approximately 3, 6 and 9 months) was used in the preanalysis/postanalysis.

Drug use was assessed at baseline and at the final follow-up visit (12 months); respondents were asked whether and how recently they used cocaine, crack or cocaine paste, heroin and any other injection drugs. The primary outcome in all analyses was drug use in the past 12 months; the preanalysis/postanalysis was replicated using drug use in the past 3 months as the outcome to assess sensitivity to temporal ordering.

Intervention exposure was assessed using 20 questionnaire items that queried contact with peer educators, contact with project materials, participation in social events or collective workshops and adherence to scheduled clinical visits. Item responses were pooled into scores using item response modelling; the resulting reliability estimate of 0.86 (akin to Cronbach's α) indicated good reliability.2 The scores were dichotomised into high exposure and no/low exposure at each visit using an a priori cut point based on whether or not respondents participated in activities that aimed to modify the social environment. We classified as exposed respondents who attended scheduled appointments, sought additional contact with intervention personnel, and participated in intervention social and community events for at least one 3-month follow-up period. We classified as unexposed respondents who attended clinic visits but participated little or not at all in community-based intervention activities.

Additional covariates measured at baseline included gender (male, female, transvestite (individuals whose gender expression does not conform to biological sex35)); age, which was categorised as 18–24, 25–30, 31–35 and over 35; educational attainment (primary, secondary, high school, some technical or university); income by terciles; marital status (married or cohabiting vs not); primary site of sex work (brothel vs not) and days of the past year engaged in sex work.


Associations of social participation, intervention exposure and demographic characteristics with drug use at baseline and follow-up were assessed in bivariate analysis. We examined the cross-sectional association between social participation and drug use at baseline and the relation between social participation and drug use over the follow-up period, adjusting for baseline use. Analyses were adjusted for individual characteristics conceptually considered confounders: age, gender, income, educational attainment, marital status, location of sex work and days of sex work in the past year.i Each analysis employed logistic regression and a marginal modelling approach.36 With this approach, we estimated the difference in the prevalence of drug use if all respondents experienced high social participation compared with low participation. We set exposure levels of participation to one-half SD above and below the mean to capture a 1 SD difference centered on the mean. This provided the following parameters: θ(high)=EW{E(Y|A=high, W)}, θ(low)=EW{E(Y|A=low, W)}, where A is social participation, W is the vector of confounders and Y is drug use. We compared low to high to estimate the marginal association between social participation and drug use: θ(low−high)=EW{E(Y|A=low, W)−E(Y|A=high, W)} (for a detailed discussion of the method, see refs. 36 and 37). Traditional analytic estimates of variance are not available for marginal models; instead, CIs were generated through bootstrapping: we resampled the data with replacement 1000 times, analysed each resample and report the resulting 2.5th and 97.5th percentile coefficients as CI bounds.38

The preanalysis/postanalysis was performed separately within strata of intervention exposure to assess whether the relation between social participation and drug use differed if an individual was exposed to Encontros; previous analyses found that the intervention modified the effect of social participation.2 In the marginal models, we estimated differences associated with a 1 SD difference around the stratum-specific means to ensure that participation values represented the range observed in each stratum.

Because 51.4% of the cohort was lost to follow-up (censored), we used inverse probability weighting techniques to account for potential non-random censoring in the preanalysis/postanalysis.39 This approach weights the non-censored observations by the inverse probability of being observed at the final follow-up. Weights were estimated using logistic regression with censoring as the outcome and confounders for drug use as the covariates. Analyses were conducted using Stata V.11 (StataCorp, College Station, Texas, USA).


Of the initial cohort of 420 participants, 204 attended a final visit. Table 1 displays characteristics of the sample at baseline and follow-up; there were no significant differences in demographic characteristics despite attrition. Brothel-based sex workers, who tended to be younger, more educated and from outside of the study city, were somewhat more likely to leave the study (p=0.09).

Table 1

Characteristics of Encontros cohort and of drug users at baseline and follow-up

Prevalence of drug use at baseline was moderate, with 82 individuals (19.6%) reporting drug use in the past year (table 1). Reported drug use consisted of cocaine (80.5%) or crack (46.3%); no participants reported injecting drugs. Drug use was more common among women and those under 30. Individuals using drugs were significantly more likely to be high income and working in brothels; they also reported more days practicing sex work in the past year. Of the 75 drug-using participants who had a sexual encounter in the week before the baseline interview, 43 (57.3%) reported using drugs or alcohol before or during sex all or most of the time, confirming the strong link between drug use and sex in this cohort. Average social participation was significantly lower among respondents using drugs at baseline.

At follow-up, 13.2% of respondents reported drug use in the past year; all drug use consisted of crack or cocaine (table 1). Drug use remained significantly more common in participants with higher income and those practicing brothel-based sex work. In addition, drug use at the final study visit was significantly higher among those using drugs at baseline; only 10 individuals moved from use to non-use. There was no association between intervention exposure and drug use at the final study visit. However, those exposed to Encontros and those unexposed did differ in levels of social participation during the study (3.0 vs 2.1, p<0.001). This difference had also been present at baseline (2.8 vs 2.2, p=0.0213), providing further justification for stratifying by intervention exposure to isolate the association of social participation with drug use.

Logistic regression analysis of the cross-sectional relation between social participation and drug use at baseline (arrow A in figure 1) is shown in table 2, with the corresponding marginal model shown in the first panel of table 3. Social participation had a significant albeit weak protective association with reported drug use (β −0.17, 95% CI −0.34 to 0.00). In the marginal models, low social participation was associated with 20.7% prevalence of drug use while high participation was associated with 16.9% prevalence of drug use, resulting in a difference of 3.8% (95% CI −0.1 to 8.3) associated with low versus high social participation.

Table 2

Logistic model of the relation between social participation and prior year drug use at baseline (N=412)

Table 3

Predicted marginal drug use by level of social participation

The relation between social participation across the intervention follow-up period and reported drug use at the final study visit—arrows B1 and B2 in figure 1—is presented in the second panel of table 3 (marginal model) and table 4 (logistic regression).

Table 4

Logistic models of the relation between social participation during the intervention and prior year drug use at follow-up, stratified by intervention exposure

The regression models are stratified by intervention exposure, weighted for censoring and adjusted for confounders as well as for baseline drug use; the resulting smaller sample size necessitated eliminating three of the covariates considered to be confounders: gender, marital status and education. Among those unexposed to the intervention, social participation was not significantly associated with drug use (β 0.23, 95% CI −0.25 to 0.70; marginal difference −2.0%, 95% CI −12.7 to 5.1). However, among those exposed to the intervention, social participation was protective against drug use (β −2.94, 95% CI −5.28 to −0.59). In the marginal models for the exposed, low social participation was associated with 17.1% drug use while high participation was associated with 8.5% drug use. This results in a difference of 8.6% (95% CI 0.1 to 23.3) associated with a 1 SD difference in social participation. Sensitivity analyses restricted to drug use in the past 3 months provided similar results: a 1 SD difference in social participation was associated with a marginal difference in drug use of −0.8% (95% CI −6.3 to 5.6) among the unexposed and of 6.0% (95% CI 1.1 to 16.7) among the exposed. (Low power in this analysis required removal of one further confounder, income level.) Among those exposed to Encontros, increased social participation was associated with a substantial and statistically significant drop in drug use.


This prospective cohort study found a significant association between social participation and reduced drug use among sex workers exposed to a structural intervention to prevent STI/HIV. Individuals with higher baseline social participation were more likely to be active in Encontros; their social participation beyond the intervention appeared protective while social participation among the unexposed showed no association with drug use. Although the cross-sectional association is weak, the magnitude of the marginal association in the preanalysis/postanalysis among the exposed suggests a notable link between social participation and drug use despite the fact that the intervention did not directly target drug use. This finding contributes to the growing literature on structural interventions among sex workers by relating such interventions to broader health behaviour impacts. In addition, these findings contribute to a wider literature on how social participation may affect drug use. The association between social participation and health may be mediated by social influence; by social engagement reinforcing an individual's role and value, whether as a drug user or non-user; and by access to resources such as drugs or drug treatment.31 By measuring participation in social and community groups, the present study specifically addresses the impact of links between a marginalised group and the broader community. The few existing observational studies of drug users with related measures have provided limited or no evidence of a protective effect of social group participation.33 ,34 This study strengthens the evidence for a protective role of external social participation in the context of a community-building intervention such as Encontros.

Moving from associations in observational research to a causal interpretation requires a number of assumptions fully elaborated elsewhere40; we address each assumption in turn. Temporality: the prospective study design enabled estimation of change in drug use from baseline to follow-up. However, social participation during interim study visits and drug use in the past year were measured in overlapping periods, making it impossible to definitively establish temporal ordering. Sensitivity analyses restricted to drug use after the last measurement of social participation estimated similar associations, buttressing the assumption that social participation preceded drug use. Ignorability: although we controlled for baseline sociodemographic confounders, all observational studies have some degree of unmeasured confounding. Sample size at follow-up necessitated eliminating confounders from the analysis. The initial sample size was established to power the primary outcome of STI incidence. Stable unit treatment value: the assumption that social participation by one individual will not affect potential outcomes for other individuals is plausible but not guaranteed in this context. The concern is somewhat mitigated because the participation measure focuses on engagement with the external community (eg, churches, clubs and voluntary organisations) rather than internal relations among sex workers.

Beyond requiring the assumptions above for causal interpretation, the study is limited by the substantial attrition over the course of the intervention, although we found no significant differences between those lost to follow-up and those retained. In addition, the context of sex work and drug use may vary considerably by population and setting. Although the prevalence of drug use in this cohort was similar to other estimates among sex workers in Brazil,30 the international border location shapes drug use in Corumbá: drug availability is high, policing of lower-level users and traffickers is aggressive and inhalation is the predominant method of cocaine use. However, these unique elements would not necessarily affect the relationship under study. Further research is required to assess the generalisability of these findings, including whether other interventions that increase social participation have the same impact.

The principal methodological strengths of this study include the follow-up analysis, the careful measurement of social constructs, and the use of censoring weights to address attrition. This study adds to the existing literature on structural interventions by providing evidence that social participation catalysed by the intervention may affect health beyond sexual risk behaviours. The association of social participation with reduced drug use among sex workers engaging in a community-building intervention is a novel finding. Future research should include assessment of the broader health implications of changes in social structural factors in other populations and settings. In theory, successful structural interventions modify individual and community characteristics in myriad ways with potential for multiple and synergistic health benefits; measuring a range of health-related outcomes will improve our understanding of the full impact of these interventions.

What is already known on this subject?

  • Structural interventions to prevent HIV and reduce sexually transmitted infections (STI) have simultaneously increased sex workers’ integration and social participation and improved condom use. Drug use among sex workers poses a direct risk to health and can increase sexual risk behaviour. Although social participation could theoretically exacerbate or reduce drug use, no studies to date have assessed the association of social participation with drug use among sex workers in the context of a structural intervention.

What this study adds

  • Participation in external social and community groups, such as clubs or voluntary organisations, was associated with lower crack and cocaine use among sex workers exposed to an intervention intended to engender a more supportive social environment and reduce sexual risk behaviour. Social participation was not protective for drug use among those unexposed. These findings suggest that STI prevention initiatives that enhance social integration among marginalised populations can produce broad health impacts, including reductions in drug use.


We acknowledge the Encontros investigators, study advisory committee, and our institutional partners (the Population Council, the Brazilian National STI/AIDS Program, Ministry of Health, Pathfinder do Brasil, Rede Brasileira de Prostitutas, and the state and municipal Secretary of Health) for their dedication to the project and support of this work. We also thank Angela Donini and Adriana Pinho for study coordination and assistance with data cleaning and Alan Hubbard for providing statistical guidance.


  • Contributors SAL and MC designed the research protocol and supervised its implementation. SAL and DK designed social environmental measures. HHL analysed the data and drafted the article; she is the guarantor. JA and SAL conceptualised the article; JA defined the analytic approach. All of the authors reviewed drafts of the article and contributed to the conceptual framework.

  • Funding Data collection for this research was supported by the Population Council, Pathfinder do Brasil, and the Ministry of Health in Brazil. The first author received support for this manuscript from the Hellman Family Faculty Fund at the School of Public Health, University of California, Berkeley. The funding sources played no role in the study design, data analysis, and manuscript writing, or in the decision to submit this manuscript for publication.

  • Competing interests None.

  • Ethics approval Ethics Committee in Mato Grosso do Sul, Brazil; the Brazilian National Committee of Ethics in Research; the Institutional Review Board at the Population Council; and the Committee for the Protection of Human Subjects at the University of California, Berkeley.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • i All models were assessed for collinearity through calculation of the variance inflation factor (VIF), with VIF >4.0 considered evidence of correlation. Overall VIF statistics were 1.42, 1.21 and 1.36 for the cross-sectional, pre/post low-exposure, and pre/post high-exposure analyses, respectively, indicating no significant collinearity. The highest individual VIF statistic was 2.69 on middle school education in the cross-sectional model.


Free sample

This recent issue is free to all users to allow everyone the opportunity to see the full scope and typical content of JECH.
View free sample issue >>

Don't forget to sign up for content alerts so you keep up to date with all the articles as they are published.