Study Design
We will undertake a cross-sectional, sexual behaviour survey. This study is registered with and approved by the Stellenbosch University Health Research Ethics Committee (N11/03/093) and it is a subsidiary study to the Zambia South Africa TB and AIDS Reduction study (ZAMSTAR) (ISRCTN36729271), a community randomised trial aimed at reducing the prevalence of tuberculosis in communities with a high burden of TB and HIV, by novel public health interventions.
Study Setting
The study outlined here will take place in three peri-urban, disadvantaged communities in the greater Cape Town area with a high burden of HIV. These communities represent one predominantly black community and two racially diverse communities. They were chosen in order to gather a representative sample of the different races most affected by HIV in South Africa.
Participants
The final primary outcome measurements of the ZAMSTAR study were collected through an HIV and TB prevalence survey, conducted in 2010. In this prevalence survey, participants provided written informed consent to complete a questionnaire on basic demographic information and medical history, to provide sputum for TB culture, and to have repeated visits by the research team. Through an additional consent procedure, participants gave consent for HIV counselling and testing. Our study will randomly select 500 previous ZAMSTAR participants in each of the three communities. In the ZAMSTAR survey, some participants chose not to have HIV testing done. For this reason, 20% of our sample per study site (100/500) will consist of participants who did not consent for HIV testing. By sampling both participants who did, and who did not, consent to an HIV test, we may be able to detect selection bias.
Study participants will be included if they are between the ages of 18 and 64 at the time of the ZAMSTAR survey, able to give informed consent, and agree to complete the study questionnaire. Individuals who do not speak English, Afrikaans or isiXhosa, or who are enrolled in any other ongoing study, will be excluded.
Measurement Instruments
A touch screen questionnaire, utilizing an audio computer-assisted self-interviewing (ACASI) application has been developed to facilitate the collection of sexual history data. The respondents will wear headphones to hear questions in their choice of English, Afrikaans, or isiXhosa, and simultaneously be able to read the question and select their answers on a 22-inch touch screen monitor. This question format was derived from a UNAIDS "Best Practice" questionnaire of sexual partnership data [23], as well as the Relationship History Calendar (RHC) and an Events History Calendar (EHC) [24]. These calendars have been previously validated and tested in adolescent and adult respondents in sub-Saharan Africa [24, 25] and the United States [26]. The modified RHC used in this study will collect detailed retrospective data on sexual histories of the participants for the year preceding the survey, for a maximum of 5 'main' sexual partners and 15 'casual' sexual partners. Before starting the actual survey, participants will view a short demonstration video and will practise using ACASI and the touch screen by answering a series of example questions.
The questionnaire is set up to ask questions along a temporal trajectory. It will begin by asking the participant basic demographic information and then proceed to ask if the participant had a main sexual partner one year ago and if that relationship is still ongoing. The participant will be able to indicate, on a touch screen timeline, the periods they were in this relationship. For each relationship, we ask whether or not the participant or his/her partner used drugs or alcohol at their first intercourse. To gauge the level of spatial connectedness, we ask how long the participant usually travelled to their partner and his/her degree of proximity. For each distinct period that the respondent engaged in sexual activity with a particular partner, measured in weeks, additional questions will be asked about the frequency of intercourse and condom use. This series of questions will be asked for each main sexual partner, as well as the casual sexual partners. The onset, dissolution, and duration of each relationship will be displayed on the touch screen timeline for the participant to see, using different colours for each partner. At the end of the questionnaire, participants will be asked how many partners they have ever had. In addition to the sexual behaviour questionnaire we added three questions to assess the ease of use, perceived confidentiality and self-reported truthfulness when answering questions in this survey. Finally, we ask them to indicate their preferred medium of answering questions about sexual behaviour: touch screen computer with ACASI, researcher-administered verbal questionnaire, self-administered written questionnaire, or telephonic survey.
All questions were developed and refined after conducting cognitive interviews with six people who were representatives of the study communities. These interviews assessed the clarity, comprehensibility, and cultural sensitivity of the proposed questions, allowing us to improve the phrasing of questions, incorporate meaningful slang, and define suitable categorical answer options that maximize the precision and accuracy of responses. Furthermore, the cognitive interviews allowed us to gauge the anticipated community response to a survey utilizing ACASI. Our interlocutors responded positively to the idea of answering questions about their sexual histories in a completely anonymous way and articulated that the communities would embrace a survey that could accomplish this.
Survey administering and data management
The senior data manager will have access to ZAMSTAR study participant information and will therefore be able to obtain names and addresses for each individual. Consent forms will be printed that include a new unique barcode - different from but linked to the barcode allocated in the ZAMSTAR study - along with names and addresses of the individuals. The barcodes from the ZAMSTAR study will not be printed on these consent forms. Therefore, none of the study staff, except the data manger, will be able to link the HIV test results obtained in the ZAMSTAR survey to our survey data.
Residents of the community, who were confirmed with names, addresses and dates of birth to be previous ZAMSTAR paricipants, will be asked if they consent to participation after reading an information sheet and listening to verbal information given by the research assistants in their home language. The research assistants will make it known that the anonymised HIV test results from the ZAMSTAR study will be accessed and used in the analysis. If the participant consents to the survey and signs the consent form, he/she will be escorted to a camper van - located in a safe part of the neighbourhood - where the survey will commence. The camper van has been refurbished to provide a private and safe office space, containing a partition, in which two participants can take the survey simultaneously at their own desk, chair and computer. Inside the van, the research assistant will scan the barcode on the consent form into the touch screen computer.
All subsequent answers to the touch screen questionnaire will be answered in private by the participant only. No one will be able to see their responses, including the research assistants who will be waiting outside of the van or in the driver's cab. They will remain close by at all times in the event that a participant requests their assistance. While two participants take the survey, two additional research assistants will be out recruiting the next two participants from the same community. The participant's name will not be asked in the questionnaire, hence the senior data manager will be the only person who can link the name from the consent form via the unique barcode. The touch screen questionnaire data will be uploaded from the laptop onto the central database designed for this study. All data will be backed up onto a second server. The consent forms will be stored in locked cabinets at the study centre.
To ensure the quality of the data collection process, the fieldwork coordinator will randomly select 5% of the completed consent forms every week and do a home visit to verify that the participant was in fact enrolled in the study and to confirm the signature on the consent form.
HIV Status
The HIV status of participants was determined in the 2010 ZAMSTAR Tuberculosis Prevalence Survey. In that study, for those who consented to HIV testing, Abbott Determine HIV-1/2 screening tests were used, and a second, confirmatory test was conducted for those who tested positive. The data manager of this study is the only person to have access to the ZAMSTAR HIV test results and after the completion of our survey, he will link these test results to the new survey data.
Statistical Analyses
In an initial descriptive analysis, the prevalence of HIV infection and the point prevalence of concurrent relationships and age-disparate relationships will be calculated, along with the average number of partners per year, the average frequency of sexual intercourse, the frequency of condom use for each of the reported relationships, condom use at last sexual intercourse with each of the reported partners, the average age difference between individuals and their partners and the variance of these age differences in the study population. Since many individuals are likely to report on more than one relationship, linear mixed-effects models will be used to analyze the inter- and intra-subject variability in the age difference between sexual partners, and covariates associated with large age differences. To take into account the fact that multiple sexual partnerships may co-exist, and the timing of partnership initiation and dissolution are correlated, or dependent on each other, survival analysis for correlated event times will be conducted to model concurrency patterns, partnership durations and rates of partnership initiation and termination. This type of survival analysis also accommodates right-censored observations where current partnership durations are only known up to the time of the survey. Marginal survival models for correlated event times will be used to enable comparisons between genders or other covariate groupings.
Next, logistic regression models for clustered data (due to respondents reporting on the characteristics of multiple partners, constituting repeated measures) will be fit to the data to determine whether HIV status is associated with age-disparity and sexual connectedness. Age-disparity and sexual connectedness will be operationalised using the following individual and community characteristics: (a) the mean difference between the age of an individual and the age of his/her partners; (b) the variability in age difference between an individual and his/her partners; (c) being engaged in a concurrent relationship; (d) having ever engaged in concurrent relationships; (e) the number of past and present concurrent relationships; (f) the cumulative overlapping time engaged in concurrent relationships; (g) the number of lifetime sexual partners; (h) spatial proximity to sexual partners; (i) the population mean age differences between individuals and their partners; (j) the population variability in age difference between individuals and their partners; (k) point prevalence of concurrency; (l) prevalence of having ever engaged in concurrent relationships; (m) population mean number of past and present concurrent relationships; (n) the population per capita cumulative overlapping time engaged in concurrent relationships; (o) population mean number of lifetime partners; (p) population spatial assortativeness (i.e. choosing partners from one's own community).
In a supplementary analysis, socio-demographic and behavioural predictors of egocentric and community sexual network structures will be ascertained. Potential risk factors under consideration are: race, gender, proximity to city centre, education level, socio-economic status, age, religion, and alcohol and drug use at first intercourse. Confirmed risk factors will be added to the models as confounders in the primary analyses and adjusted associations for age-disparity, concurrency and HIV status at the individual and cluster level will be calculated.
Sample size calculations
For an alpha level of 5%, a design effect of 2.0 (effect of clustered data rather than independent random sample) and a prevalence of HIV infection of 15%, with the width of the 95% confidence interval at 6% (i.e. +/-3%), 1089 study participants with HIV test results are needed. Building a 10% margin for inconsistent and incomplete data, we will aim to administer the survey to 1200 study participants for whom HIV test results are available. Additionally, we will include 300 study participants (20% of the total sample) who did not opt for HIV testing, to investigate whether non-universal consent for HIV testing may have introduced selection bias.