Study design
We used baseline data from a randomized controlled study funded by the National Institute of Mental Health (NIMH). The study aimed to assess the efficacy of a combination intervention when added to the traditional HIV risk reduction (HIVRR) in reducing the incidence of STI and HIV among WESW in Southern Uganda. Between June 2019 and March 2020, the study team employed a randomized cluster design to recruit 542 WESW from 19 HIV hotspots (areas that are reported to have high HIV prevalence rates) in the region. According to Grabowski et al., HIV hotspots serve as drivers of transmission to the neighboring areas of lower prevalence [27]. Women were eligible to participate in the study if they fulfilled the following inclusion criteria: 1) at least 18 years of age, 2) engaged in transactional sex, defined as having vaginal or anal sexual intercourse in exchange for money, alcohol, or other goods in the last 30 days, and 3) engaged in at least one episode of unprotected sexual intercourse in the preceding 30 days.
Study setting
As of December 2020, the estimated total number of people living with HIV in Uganda was 1.4 million, with a prevalence rate of 31.3% among WESW, a percentage higher than the national prevalence of 5.4% [3]. The HIV prevalence in Masaka region where the study was conducted, is 11.7%, which is twice higher than the national average [3]. In addition, there is an estimated total of 1,332 WESW within the study target hotspots [28]. For this study, the research team engaged community stakeholders to identify and recruit WESW into the study. Details of the study’s methodology are described in our published protocol [29].
Ethics and informed consent
The study received approval from the Washington University in St. Louis Institutional Review Board (IRB #201811106), Columbia University IRB (IRB #AAAR9804), the Uganda Virus Research Institute (UVRI #GC/127/18/10/690), and the Uganda National Council of Science and Technology (UNCST #SS4828). All methods were performed in accordance with the relevant guidelines and regulations, and all women in the study provided voluntary written informed consent prior to participating in the study. All the research assistants received ethics training in the form of Collaborative Institutional Training Initiative (CITI) Human Subjects training and Good Clinical Practice training. The interviews were conducted in a secluded area to ensure privacy and to avoid interference. All consent forms and related study materials, including data collection measures, were translated into Luganda by a certified translator from the School of Languages, Literature, and Communication at Makerere University, then reviewed by a team of stakeholders and research staff to ensure that they are culturally and linguistically appropriate [30].
Study measures
Outcome
The primary outcome of this study was sexual risk-taking behaviors, measured using four items adapted from Schilling et al. [31]. Items included 1) the total number of customers that the participant had sex within the past 30 days, 2) the number of times in total a participant had vaginal sex with any of the customers in the past 30 days, 3) the number of times a condom was used during these encounters, and 4) whether participants could accept more money, goods or extra services from their paying customers for unprotected sex. A latent variable for sexual risk-taking behaviors was generated from the four questions.
Independent variable
Asset ownership was defined in terms of ownership of financial and physical resources. It was measured using the asset index, a scale comprising 21 items that assessed whether the women owned various items such as land, gardens, house, rental property, a small business, car, cell phone, and cattle, to mention a few. We coded each item with a “1” if the participant owned the item and a “0” if the participant did not. We used principal components analysis (PCA) to generate a latent factor variable for asset ownership.
Mediator variables
Mediator variables included depression and access to healthcare. Depression was measured using a 6-item scale from the Brief Symptoms Survey [32]. The 6-items assessed whether respondents had experienced any of the depressive symptoms, including 1) thoughts of ending life, 2) feeling lonely, 3) feeling sad, 4) feeling no interest in things, 5) feeling hopelessness about the future, and 6) feelings of worthlessness. Responses were coded from 1 = “Not at all,” 2 = “A little bit,” 3 = “Moderately,” 4 = “Quite a bit,” and 5 = “Extremely.” The theoretical range was 0–24 (alpha = 0.82). Using PCA, we extracted one factor that we used to measure depression.
Access to health services was measured using a 6-item scale related to seeking medical care in the past 12 months [33]. Responses were rated on a 5-point scale with 1 = “Strongly Agree,” 2 = “Somewhat Agree,” 3 = “Uncertain,” 4 = “Somewhat Disagree,” and 5 = “Strongly Disagree.” The six items included: a) “If I need medical care, I can get admitted without any trouble,” b) “It is hard for me to get medical care in an emergency,” c) “Sometimes I go without the medical care I need because it is too expensive,” d) “I have easy access to the medical specialists that I need,” e) “Places, where I can get medical care, are very conveniently located,” and f) “I am able to get medical care whenever I need it.” Two items were captured in the inverse direction and had to be reverse coded to match the other items. We used PCA to generate a latent variable for access to health care in the SEM.
Participants’ age and education level were included in the model as control variables.
Data analysis
We conducted the descriptive analysis using STATA SE, Version 17 (StataCorp, college station, Texas 77845). We declared data to be survey data to adjust for clustering at the level of the HIV hotspots. Continuous variables were summarized using means and standard deviations. Categorical variables were summarized using percentages. We used SPSS and MPlus version 8.1 (Muthen and Muthen) to fit the principal component analysis (PCA) and the structural equation models (SEM), respectively, that assessed the effects of asset ownership on sexual risk-taking behaviors among the WESW. Prior to performing PCA, we assessed the data for suitability of PCA using the KMO test for sampling adequacy, which yielded a value of 0.898 (critical value 0.5). We also computed the Bartlett’s test for sphericity, where we aimed for a significant p value. The corresponding p value was < 0.001.
The results tables reported both the non-standardized coefficients (B) and the standardized coefficients (β) and their 95% confidence intervals. For all the analyses, the level of significance was set at 0.05.
To test for the goodness of fit of the final model, we used several parameters, including 1.) the chi-square for the goodness of fit, where we aimed for a non-significant p-value. 2.) Root Mean Square Error of Approximation (RMSEA) less than 0.063) Standardized Root Mean Square Residual (SRMR) less than 0.08, and 4.) Comparative Fit Index (CFI) of at least 0.95. We estimated the direct effects, indirect effects through the mediators, and total effects of asset ownership on sexual risk-taking behaviors. The direct effect is the effect of asset ownership on sexual risk-taking behavior in the absence of mediators. A specific indirect effect represents how much that mediator explains the asset ownership effect. The sum of the individual-specific indirect effects constituted the total indirect effects. The total effect is the sum of the direct and indirect effects. Using the indirect and total effects, we determined the proportion of the mediated effect.
