Ethics statement

This study was approved by the Johns Hopkins Bloomberg School of Public Health (JHSPH) institutional review board (IRB) (Ref #: 00007065) as well as the Muhimbili University of Health and Applied Sciences (MUHAS) IRB and National Institute of Medical Research (NIMR) IRB in Tanzania (Ref #: 1593). Oral informed consent was received from all participants. Oral informed consent was used given that the study worked with individuals with historically marginalized identities at the intersection of HIV- and sex work-related stigma. The study opted for oral consent because documenting their signature created additional possible risks related to participants’ confidentiality. This decision was informed by prior experience conducting the informed consent process for HIV prevention studies among people who sell sex and in the cultural context of Iringa. The informed consent procedures for this study were designed to maximize understanding of potential risks to participants. Forms were translated into local languages, and content read aloud to participants. Understanding was ensured by asking participants to summarize the study and explain the reasons why they wanted to participate. Individuals were also provided with a copy of their consent form if they wanted it as well as information on how to contact the study staff to report adverse events associated with their participation in the research. This oral consent procedure was approved by the JHSPH IRB as well as MUHAS and NIMR in Tanzania as part of the approval of the study.

Study setting

This study draws on data collected from FSWs in Iringa, Tanzania, situated in the Southern Highlands of Tanzania. The Tanzania-Zambian (Tan-Zam) highway passes through Iringa, making mobility–be it by truckers traveling or residents–fundamental to life in this region [31–33]. Agricultural production is also an important aspect of the economy in Iringa, with residents and FSWs alike traveling seasonally for the harvest of crops like tomatoes, tea, and timber. Within this context, FSWs often travel along the Tan-Zam highway as well as seasonally for sex work [33]. The agricultural and transport industries both increase demand for sex work across various locations in and outside of the region, with a seasonal ebb and flow. A key type of clientele for FSWs along the highway is truckers; FSWs sometimes accompany truckers on spontaneous long-haul trips, with uncertainly about when and how they will return. As in other areas of Tanzania [34], FSWs in Iringa also travel to other areas–such as mining communities–for the purposes of sex work.

According to the 2016–2017 Tanzania HIV Impact Survey, HIV prevalence is higher in Iringa (11.3%) among adults 15 years and older than nationally (4.7%) [35]. Government HIV prevention programming in Iringa include HIV testing and counseling, both standalone and integrated into existing maternal and child healthcare services, as well as care and treatment services for people living with HIV [36]. Non-governmental organizations have worked in Iringa as well, with some services, including peer education and mobile outreach, tailored for women at increased risk for HIV, including FSWs [37–39].

Study design and sampling

To examine the longitudinal relationship between sex work-related mobility and experience of GBV, this study drew on baseline and 18-month follow-up data collected as part of Project Shikamana. As described elsewhere [32,40], Project Shikamana was a prospective community-randomized trial conducted to examine the impact of a community empowerment-based combination HIV prevention intervention in two communities in the Iringa region of Tanzania. HIV prevention activities as part of Project Shikamana included mobile HIV testing and counseling, peer education, a drop-in center, and wider engagement with FSWs that focused on improving HIV service utilization, risk reduction, and fostering skills development and social cohesion [40,41]. One of the key components of this intervention focused on violence prevention, which was included at the request of women themselves based on their lived experiences [41,42]. Specific activities included workshops and trainings for FSWs as well as police focusing on how to prevent and respond to GBV. A working group was also formed for FSWs to share their experiences of violence. Resources, such as a safety tip card, were also developed to provide FSWs with information for how to prevent and report violence.

After a mapping of sex work venues in each community, time-location sampling was used to recruit both HIV-positive and HIV-negative FSWs for the baseline survey. Eligible FSWs were at least 18 years old and reported that they had exchanged sex for money, based on self-report, within the past month. Oral informed consent was received from all participants before beginning data collection.

Four-hundred ninety-six (496) FSWs were recruited at baseline, with 387 also participating in the 18-month follow-up survey. The final sample analyzed here included all FSWs participating in both baseline and follow-up surveys (n = 387).

Measures

The outcome of interest was any recent experience of physical or sexual GBV from any sexual partner, defined as any self-reported experience of either physical or sexual violence within the past six months from a new or regular client or non-paying partner. Any recent experience of physical or sexual GBV from a sexual partner at baseline was assessed based on responses to nine questions following the WHO and work by Decker and colleagues [9,32,43]. At follow-up, questions were simplified and recent GBV was measured based on recent experience of physical violence–defined as being hit, slapped, kicked, pushed, shoved, or otherwise physically hurt–and sexual violence–forced sex by an individual or by a group–from new or regular clients or partners. This modification allowed the study to capture more accurately the forms of violence perpetrated by specific actors, rather than among all sexual partners (clients or partners). For the purposes of this analysis, recent experience of physical violence was defined, at both baseline and follow-up, if participants reported at least one form of physical or sexual violence from any sexual partner (clients or partners). The explanatory variable of interest was sex work-related mobility, defined as recent mobility outside of the Iringa region within the past six months specifically for sex work. Sex work-related mobility was measured at both baseline and follow-up. Refusal to respond was coded as missing for the purposes of analysis (n = 3 at follow-up only).

Confounding and correlate variables considered for final models were both time variant and time invariant and included relevant socio-demographic characteristics and aspects of FSWs’ living situation and work environments.

Time-variant covariates included in final adjusted models were marital/cohabitation status, average monthly income, alcohol consumption, recent drug use, average number of clients per week, and perceived financial security. Marital/cohabitation status was defined as currently married or living with a sex partner vs. not. Average monthly income was defined as self-reported average monthly income, including sex work, over the last six months and dichotomized at 120,000 Tsh (median average monthly income at baseline; approximately $51USD) to compare higher vs. lower average monthly income. Alcohol consumption was measured using the AUDIT-3 short-form [44] and was based on aggregated scores across three questions assessing frequency of alcohol consumption per week, number of drinks consumed when drinking, and frequency of drinking six or more drinks on a single occasion. Summary scores were dichotomized into hazardous drinking (scores of three or more) and not hazardous drinking (scores of less than three). Drug use was measured based on self-reported use of any drugs within the past six months, including substances such as marijuana, khat, or heroin. Average number of clients per week was a binary variable dichotomized at the baseline median response, comparing an average of more than two clients per week to an average of two or fewer clients per week. Perceived financial security was self-reported and based on responses to the question “In the last six months, how would you describe your overall financial security?” This variable compared participants who reported “excellent, very good, good, or fair” to those who reported “poor.”

Time invariant covariates measured at baseline included participants’ age (categorized into four quartiles based on exploratory analyses: younger than 23, 23–25, 26–30, and over 30 years of age), HIV status (HIV-infected at baseline vs. not), community (intervention vs. control community), participants’ ethnic group (Hehe vs. other), and educational attainment (secondary school or higher vs. less than secondary school).

Statistical analysis

Descriptive analyses examined recent experience of physical or sexual GBV, sex work-related mobility, and other covariates of interest at baseline and/or follow-up. Key baseline socio-demographic characteristics were compared between those individuals participating at baseline and follow-up (n = 387) and those not participating in the follow-up survey (n = 109) to examine potential effects of loss to follow-up using Pearson’s χ² tests. To examine bivariate associations between sex work-related mobility as well as other covariates and the outcome of interest, Pearson’s χ² tests were conducted to examine associations at baseline and at follow-up. As recent experience of physical or sexual GBV was commonly reported among study participants, unadjusted Poisson regression models were fit using generalized estimating equations (GEE) with robust variance estimations and an exchangeable correlation structure to examine the longitudinal relationship with sex work-related mobility as well as with other variables of interest. The use of robust Poisson regression was proposed as an alternative to logistic regression for common outcomes [45]. Longitudinal models accounted for clustering within FSWs across time points. All analyses were conducted in Stata15 [46].

It was hypothesized that age was a potential effect modifier of the relationship between sex work-related mobility and recent experience of physical or sexual GBV. An interaction term between the explanatory variable of interest and this socio-demographic characteristic was included in bivariate robust Poisson regression models to examine this hypothesis. There was no significant evidence of effect modification at the bivariate level (p>0.05), and the interaction term was dropped in final multivariate models presented here.

An adjusted robust Poisson regression model first included relevant factors related to FSWs’ socio-demographic characteristics, living situation, and work environments. A manual backward elimination method from this initial full robust Poisson model was conducted. Wald tests were used to assess overall contribution of variables to the final model. Variables were removed if they did not significantly contribute to the model assessed based on p>0.2. Following this step, retained covariates were included in the final model examining the longitudinal relationship between sex work-related mobility and recent experience of physical or sexual GBV. As part of a sensitivity analysis, the final model suggested following this manual backward selection approach was compared to models suggested using forward and backward stepwise selection with a cut-off of p = 0.2. Model fit was assessed using Cui’s quasilikelihood under the independence model criterion (QIC), an adaptation of Akaike’s information criterion (AIC) for use with GEE analyses, to identify the best group of covariates for the final model [47]. The final model presented here had the lowest QIC among adjusted models fit. Variance inflation factors (VIFs) were calculated to assess collinearity of explanatory variables. All VIFs were less than two, suggesting minimal collinearity of covariates.

The final model adjusted for study visit (follow-up vs. baseline), community, HIV status at baseline, age, educational attainment at baseline, ethnic group, marital/cohabitation status, average number of clients per week, perceived financial security, alcohol consumption, and recent drug use. All coefficients were exponentiated and reflect either unadjusted or adjusted incidence rate ratios (IRR).

Limitations

This study extends previous cross-sectional analyses [21] to examine longitudinal correlates of any recent experience of GBV. First, analyses presented here include data collected from two time points, with FSWs lost to follow-up excluded from analyses. As a result, it is possible that those FSWs who dropped out may have been different from those who remained in the study. While there were no significant differences in the outcome or explanatory variables of interest at baseline between those who participated in both baseline and follow-up and those who did not, those who were lost to follow-up tended to be younger, not living with HIV, and not married/living with a sex partner at baseline. As mobility is often correlated with age and is a common reason for participants being lost to follow-up and a common challenge for surveillance [12], it is likely that the estimates presented here underestimate recent mobility for sex work among FSWs in Iringa at follow-up. Future research should examine the decline in sex work-related mobility from baseline to follow-up (26% to 11%) to understand the factors, be it contextual (e.g. seasonal variations in mobility patterns) or related to participation, engagement, and retention in the parent study over time, that may influence such mobility.

Second, this study is limited by its measures of GBV and FSWs’ mobility. Measures of GBV at baseline and follow-up limited analyses to any recent experience of GBV from a client or partner. Future research should ensure consistent measurement of GBV across time points, and may need to include GBV from other perpetrators, such as police or community members. In addition, there remains a need for future research to understand the nuanced relationships between mobility and GBV. Recent work in Zimbabwe has begun to identify typologies of mobility based on destination, duration, and other aspects of FSWs’ mobility [57]. Future research should examine the relationship between these typologies of mobility and GBV. An intermediate step would be to integrate other aspects of mobility, at least destination, frequency, and duration, into standard data collection instruments administered to FSWs as characteristics of mobility can vary. A qualitative investigation of mobility trajectories could elaborate on or identify emergent mobility typologies for FSWs. Previous research in north-western Tanzania, for example, used qualitative, ethnographic methods to identify typologies of women and men living and working in a mining community that described their unique risks for HIV and other infections [34]. These typologies illustrated how FSWs’ mobility was often circular, following the pay periods of those working in the mines. The expansion of such an approach to exploring mobility among FSWs in greater depth could be used to understand the mechanisms by which mobility and GBV intersect in FSWs’ lives.

Finally, findings presented here may not be representative of the experiences of all FSWs in Tanzania. Analyses were conducted using baseline and follow-up data from Project Shikamana, a community-randomized trial conducted in two communities in Iringa. While Project Shikamana included activities focused on violence prevention, there were no significant reductions in violence found in adjusted models, suggesting the need for more comprehensive approaches that address the socio-structural factors that influence GBV in this context. The socio-structural environment and context in communities where this study was conducted may differ from other communities in Iringa and elsewhere in Tanzania, given their proximity to the Tan-Zam highway [31] that may influence not only FSWs’ mobility, but also their work environments.