Data source
This secondary data analysis used the 2012 South African population-based household survey on HIV [29]. The data was collected using a multi-stage stratified cluster sampling design. A total of 1 000 census enumeration areas (EAs) from the 2001 population census in South Africa were randomly selected using probability proportional to size and stratified by province, locality type and race in urban areas from a database of 86,000 EAs. In each sampled EA a total of 15 visiting points (VPs) or households were used as secondary sampling units. Persons of all ages living in South African households and hostels were eligible to participate and formed the ultimate sampling unit.
Four questionnaires were administered in the survey to solicit information including among other socio-demographic characteristics, HIV knowledge, self-received risk of being infected with HIV and HIV related stigma and discrimination against PLHIV. In addition, blood specimens were collected from consenting individuals for HIV testing. The current study is based on a sub-sample of youth and adult individuals 15 years and older who responded to the questions on psychological distress.
Measures
Endogenous variables
Psychological distress was the observed endogenous variable based on the respondent’s experience of depressive and anxiety disorders measured using the Kessler 10 scale [30], which consists of the following 10 items that describe how they felt during the previous 30 days: How often did you feel: Tired out for no good reason? So nervous that nothing could calm you down? Hopeless; Restless or fidgety: So restless that you could not sit still; Depressed? That everything was an effort? So sad that nothing could cheer you up? Worthless?’ Responses to these items were recorded using a 5-point Likert scale (1 = never, 2 = rarely, 3 = some of the time, 4 = most of the time, 5 = all of the time). The scores from these responses were then summed to calculate a total score indicating whether the respondents were likely to experience psychological distress. The scores were then dichotomized into a binary outcome those who scored <19 absence of psychological distress = 0) and those who scored ≥20 (presence of psychological distress =1).
Exogenous variables
The selected exogenous variables included a set of demographic variables such as age (15-24, 25-34, 35-49, 50 years and older), sex (male and female), race (black African and other races), educational level (primary/ no education, secondary, tertiary), employment status (unemployed and employed), locality type (urban formal, urban informal, rural informal/ tribal areas, rural formal/ farm areas) and asset based socio-economic status (low and high). This also included HIV-related variables such as self-perceived risk of HIV infection (no and yes), HIV knowledge and myth rejection (no and yes), ever tested for HIV (no and yes), awareness of HIV status based on the question “Have you been told/informed of the result of your most recent test? (no and yes), external HIV-related stigma (high and low), and self-rated health (fair/poor and good/excellent).
Mediator variable
HIV status was included as a mediator in the relationship between the endogenous and exogenous variables. Although HIV testing was anonymous and returning of the results optional and dependent on study participant, it is assumed that the majority of study participants were aware of their HIV status, and this is hypothesized that HIV status mediates the effects of demographic, health and HIV-related variables on psychological distress.
Conceptual model and analysis
Generalized structural equation modelling (G-SEM)-path analysis was used to explore the direct and indirect relationships of key variables with psychological distress using HIV status as a mediator variable (see Figure 1). The conceptual model follows the Fundamental Causes Theory which suggests that individuals’ health condition is influenced by contextual factors [31] such as demographics (age, gender, race, locality), socio-economic status (educational level, employment), social contexts (social support), and persistent health disparities (self-rated health, HIV related stigma). This model also includes health and HIV-related factors such alcohol use AUDUT score, self-rated health, HIV testing history (ever had an HIV test), awareness of HIV status, self-perceived risk of HIV, and experiences of externalised HIV-related stigma.
G-SEM was used to measure linear and non-linear causal relationships among selected variables, while simultaneously accounting for measurement error. G-SEM is a combination of three statistical techniques: multiple regression, path analysis, and factor analysis. Its purpose was to determine the extent to which a proposed theoretical model, expressed by a set of relations among different constructs, is supported by the collected data. Mediation analysis for each variable was performed and a final path analysis including the goodness of fit was conducted. Goodness-of-fit chi square test, root mean square error of approximation (RMSEA), Tucker–Lewis’s index (TLI), and comparative fit index (CFI) were used to assess the model fit. All variables with p<0.05 were considered statistically significant and statistical analyses were performed using Stata (V.16, Stata Corp, College Station, Texas, USA) statistical software.