Study design and settings
This cross-sectional analytical study (follow-up exit interview) was conducted between January to May 2018 on a sample of parturient women with HIV enrolled in the electronic database of the East London Prospective Cohort Study [21]. All pregnant women attending maternity service would have received HIV testing, commenced on ART if diagnosed with HIV and documented in the antenatal medical records per the PMTCT guideline [13]. An electronic database was created for research purposes by the investigators between September 2015 and May 2016 to track the PMTCT outcomes of parturient women with HIV and their infants in three hospitals in Buffalo City and the Amathole district of the East Cape Province of South Africa. These hospitals serve a combined population of 1.7 million people residing in the rural and urban communities of the central region of the Eastern Cape [22].
Participants and Sample size
The sample size for this sub-study (exit interview) was estimated as 485, using the formula for categorical data, at a confidence level of 95%, a precision level of +/−4% and 10% possible attrition. All the women enrolled in the East London Prospective Cohort Study database (N=1709) signed informed consent at the baseline to be contacted for the follow up (exit) study. However, only those who were accessible telephonically were considered eligible for this follow-up survey. Each participant was offered a choice to either complete an interviewer-guided interview face-to-face or telephonically. A few participants (n=43) who chose to attend interviews at one of the three hospitals were reimbursed for the cost of transportation. Those who chose to complete telephonic interviews agreed to a scheduled time with our research team. A convenience sampling of the participants who were available and willing to participate was conducted.
We employed and trained two research assistants, who were fluent in both IsiXhosa (local language) and English for this study. The research team successfully contacted 509 participants out of the total of 1709 participants (29.9%). Some of the eligible participants were no longer accessible through any of the three contactable mobile numbers obtained from the electronic database. We designed a questionnaire specifically for the exit interview, which was piloted with 12 parturient women with HIV in one of the hospitals to ascertain the validity and reliability of the instrument. We subsequently adjusted the questionnaire using feedback from the participants and the investigators.
Measures
With the aid of a validated questionnaire, the outcome (dependent) and independent variables were obtained from the interview. These variables were informed by the existing knowledge in the literature and are described briefly.
Outcome variable: The key outcome variable for this study was disclosure of HIV to a sexual partner, and this was measured by asking the participants if they had disclosed their serostatus to their sexual partners. Responses were categorized as "yes" or "no" which was coded as 1 for 'yes' and 2 for 'no'. Also, we used an open-ended question to elicit information on the reasons for non-disclosure.
Independent variables: Socio-demographic, lifestyle behaviours and clinical charateristics were the covariates included in this study. The selection of these covariates was based on the existing knowledge of HIV disclosure in the literature [8-11].
Socio-demographic characteristics: We obtained information on the participants' ages which were coded as continuous variables, level of education and marital status. We obtained information on the employment status of the participants, occupation in the preceding 12 months at the time of the study, and whether they were engaged in a salary paying job. We obtained additional information on whether participants were receiving child support grants (social grants) from the South African government.
Lifestyle behaviours: We obtained information on the smoking status and alcohol consumption as categorical data with "yes" or "no" response, which was coded as 1 for 'yes' and 2 for 'no'. Smoking status was obtained by self -report of cigarette or any tobacco product within the preceding month of the study. Similarly, the consumption of any alcoholic beverage in the prior month of the study was obtained by self-report and recorded.
Clinical characteristics: The following clinical information was obtained through self-reporting by the participants: awareness of partner's HIV serostatus and complete adherence to ART (no missed dose of ART in the preceding week of the study). A binary response of "yes" or "no" was provided for the participants, which was coded as 1 for 'yes' and 2 for 'no'. We also documented the duration of HIV infection (period since diagnosis) among the participants, which was categorized as a continuous variable. We asked participants to provide open responses to reasons for not disclosing their status to their respective partners.
Data analysis
Data analysis was performed on responses from 485 out of 509 participants contacted (95.3%). Of the 24 participants excluded in the analysis, nine were not interested in discussing their HIV serostatus and thus, withdrew during the interview. Six participants refused to participate in the exit interview, and nine were confirmed to have died by the family member who responded to the phone call. Disclosure of HIV serostatus to sexual partners was the main outcome measure of this study. Complete responses were available for 485 respondents on the main outcome measure and were included in this analysis. All analyses were conducted with the IBM Statistical Products and Service Solutions, version 27.0 (SPSS, Chicago, IL, USA). Given that this data is a convenient sample of the 1709 women with HIV who gave birth between September 2015 and May 2016 in the three largest health facilities in Eastern Cape, South Africa, we calculated sampling weight using respondents' age distributions. We applied the sampling weight for all analyses.
Descriptive statistics (means, frequency, and percentages) were used to summarise the characteristics of the participants disaggregated by their disclosure status. Given that the main outcome variable is dichotomous ("yes or no") and there are many explanatory variables, we performed adjusted and unadjusted logistic regression models to determine the independent and significant influencing factors of non-disclosure in the study. Variables were included in the final model if they have been shown by previous studies to influence disclosure of status significantly. Knowing partner status was positively correlated with the outcome variable and was therefore removed from the multivariate analysis. Also, smoking behaviour was highly related to alcohol use and was removed from the model. The 95% confidence intervals were reported for all analyses, and p-values less than 0.05 were considered statistically significant.