Accuracy and determinants of perceived HIV risk among young women in South Africa

We used data from the Cape Area Panel Study (CAPS) [21]. The first wave of CAPS, conducted in 2002, surveyed a representative sample of 4752 young adults (ages 14 to 22) using a two-stage sample stratified by race. In the first stage, primary sampling units (PSUs) were selected based on the 1996 Census enumeration areas, and in the second stage households were randomly selected from PSUs. Respondents were re-interviewed (face-to-face in participants’ first language) up to four more times. The last wave was fielded in 2009, at which time respondents were 20–30 years old. Written consent was obtained from all respondents, and written parental consent was obtained for interviews with respondents under the age of 18.

Our study focused on the sample of black women, given the elevated risk of HIV infection in this sub-population [19]. The initial 2002 CAPS sample comprised 1221 black women of whom 833 (68%) and 746 (61%) were re-interviewed in 2005 and 2009 respectively. Extensive face-to-face questionnaires were administered in each wave. HIV testing was offered to black participants in (only) the 2009 wave and 94% of the sample consented (see Additional file 1: Supplemental Digital Content 1 for details). We restricted the data to those reporting having ever had full penetrative sexual intercourse, as heterosexual transmission is the predominant mode of HIV transmission in the study setting. We also excluded individuals who self-reported being HIV positive at baseline (2005).

Using data on HIV status from the HIV test conducted as part of the 2009 CAPS survey wave, we created a binary indicator for HIV-positive status (1 = yes) in 2009.

HIV-risk perceptions were assessed in the 2005 and 2009 waves using the question ‘Do you think you have no risk, a small risk, a moderate risk or a great risk of getting the AIDS virus?’ This survey instrument for risk perception has been used elsewhere [16, 22, 23]. Respondents had the option of reporting an HIV-positive status. “Don’t know” was included as a potential response. In our analysis of the accuracy of HIV risk perception, we use the 5-category (don’t know, no, small, moderate, or great risk) variable. For analysis of the correlates of risk perceptions in 2009, we created a binary variable to separate individuals who responded that they had no or small risk from those who perceived themselves at moderate or great HIV risk. Women answering “don’t know” were excluded from this measure. We justified this choice, which has precedent in the literature [11], on two grounds. First, only 73 respondents (12%) answered “don’t know.” Second, a “don’t know” response could reflect failure to disclose an HIV-positive status to study fieldworkers because of social desirability bias.

In analysing the correlates of risk perceptions, we used data on three sets of covariates, which were chosen a priori as factors that could theoretically influence risk perceptions or have been shown to do so in the literature. These covariates are all drawn from the 2009 data, as this survey wave contained the richest measures of demographic, sexual behaviour, and socio-psychological variables.

The first set of covariates comprised socioeconomic and demographic factors. Specifically, we considered measures of age, years of schooling, marital status, employment, and monthly household income per capita.

The second focused on current and historical sexual behaviours. We created a measure of condom use that identified individuals who reported consistently using condoms (always or usually) with their most recent partner. As multiple sexual partnerships are often associated positively with perceived HIV infection risk [22, 24], we included a marker of having had three or more lifetime partners. (We chose this cut-off given that participants who did not provide an overall estimate of their number of life-time partners often reported at least three most recent partners in the survey’s partner history table.) We created a measure of age-disparate partnerships as those in which the male partner was 5 or more years older, which is consistent with the UNAIDS definition [25] and the literature [26, 27]. We also included binary indicators of previous sexually transmitted disease (STD) history, and whether or not women perceived their most recent partner to have other partners (following a growing literature around the effects of partner concurrency on self-perceived HIV risk) [11, 22, 24]. An indicator of any alcohol consumption over the previous month was used, as alcohol consumption has been associated with risky sexual behaviours [28]. It is worth noting that examining sexual behaviours is also a way to assess the validity of our measure of risk perceptions [29]. For example, a lack of correlation could signal a poor positive predictive value of the survey question, in addition to individual risk perceptions being inaccurate.

The third set of covariates focused on socio-psychological aspects relating to HIV, including beliefs and experiences. We created an indicator of HIV knowledge, based on a summation of correct responses to four questions (see Additional file 1: Supplemental Digital Content 2). We also included a binary indicator of beliefs about the protective benefits for men of male circumcision, following recent findings that women who believed that male circumcision protects men from HIV infection also believed the protection extended to women, and accordingly downgraded their perceived HIV risk [30, 31]. Such beliefs may be relevant in the context of our study, as the vast majority of black African men in the study region are circumcised [32]. HIV-related symbolic stigma (negative moral judgements of people living with HIV) may influence perceived risk if a blaming or ‘othering’ response leads to a heightened sense of invulnerability to HIV [33‐35]. We created a measure of symbolic stigma based on the number of the following questions for which a stigmatising response was given: ‘Do you think HIV/AIDS is a punishment for sleeping around?’; ‘Do you think that some people with HIV/AIDS want to infect other people with the virus?’; ‘Do you think many people who get HIV infected through sex have only themselves to blame?’ To capture the effects of social networks, which have been shown in other work to be highly predictive of risk perceptions [28], we included a binary measure of knowing someone living with HIV or believed to have died of AIDS. Finally, we included binary measures of prior HIV testing and knowledge of partners’ HIV status.

First, we assessed the accuracy of perceived HIV risk. To do so, we identified all black women who were surveyed both in 2005 – the first year the perceived HIV-risk measure was fielded – and in 2009, the year when HIV tests were conducted. We used χ ² tests to examine the association between respondents’ perceived risk in 2005 of contracting HIV and HIV test results in 2009. We excluded participants who reported an HIV-positive status in 2005 (n = 10), as it is not clear how HIV-positive women aware of their serostatus would have responded to the perceived risk question. We also analysed the association between perceived risk in 2009 of getting HIV and actual HIV status in 2009, in order to assess whether risk perceptions improved over time. To do so, we used all observations from the 2009 survey wave with the requisite information on both variables.

Second, we examined transitions in HIV risk perception between 2005 and 2009. Specifically, we calculated the HIV prevalence for each of four groups: those reporting low risk perceptions in both 2005 and 2009; those reporting high risk perceptions in both waves; and both transitions between high and low perceived risk. This analysis helps assess how risk perceptions evolve with age and whether the individuals adopt information relevant to their HIV risks over time. If, for example, individuals were correctly updating perceived HIV risk between survey waves on the basis of actual risk, we would expect to find higher HIV prevalence for individuals moving from low to high risk over that period than for those continuing to perceive low risks. We examined potential bias from attrition in the analysis of risk perception accuracy and transitions by assessing whether the probability of follow-up in 2009 was associated with perceived HIV-risk in 2005.

To account for non-response and complex survey design, we used CAPS sample weights and adjusted standard errors for clustering at primary sampling unit. All analyses were conducted with Stata 14.0 (Stata Corp. College Station, TX, USA).

Descriptive statistics are presented in Table 1. Column 1 presents baseline (2005) data for the sample used for our prospective analysis of accuracy of perceived risk (N = 530). The average age was 21.3 years. Twenty-one percent had completed grade 12 and 40% were still enrolled in education. Few in the sample were married (6%) or employed (24%). Reflective of the high HIV prevalence in the region, more than two thirds knew someone living with HIV or who had died of AIDS (69%). Column 2 presents the characteristics of the cross-sectional sample (n = 539) used in the analysis of factors associated with perceived HIV risk in 2009. The average age was 24.8 years. Forty-two percent had completed grade 12 and 8% were still enrolled in education. Sixty-nine percent reported ever testing for HIV and 66% knew persons living with HIV or who had died of AIDS.

Figure 1a displays the proportion of individuals by perceived risk in 2005 who tested HIV positive in 2009. Women who had perceived themselves at no or small risk of getting HIV in 2005 were as likely to test positive for HIV in 2009 as individuals who had perceived themselves to be at moderate or great risk (31% vs 29%, p = 0.80), with the proportions ranging from 27 to 31%. Furthermore, the distribution of perceived risk responses in 2005 was extremely similar among individuals who tested HIV-negative compared to those who tested HIV-positive in 2009 (χ² = 1.43; p = 0.85), which indicates that respondents were highly inaccurate at evaluating their own HIV risk.

Figure 1b displays results of the same analysis using the 2009 data. Among this sample (n = 649), the majority (55%) perceived themselves at no (24%) or small (31%) HIV risk, 33% at moderate/great risk and 12% were uncertain. Compared to the panel analysis (Fig. 1a), the proportion who tested HIV positive among those with a moderate/high perceived risk was very similar, while the HIV-positive percentage among women who perceived no/small HIV risk (24.4%) dropped. While this suggests that the accuracy of perceived risk improved somewhat over time, the difference in the proportion who were HIV positive between the no/small perceived risk group and the moderate/great perceived risk group remained small and statistically insignificant (24% vs 30%, p = 0.12). Furthermore the distribution of perceived risk responses in 2009 was not statistically different across HIV-positive and HIV-negative women (χ² = 5.8; p = 0.24).

Table 2 presents HIV status in 2009 by changes in perceived risk between 2005 and 2009 among the panel sample (n = 446) of individuals with the perceived risk data in both years and who did not self-report being HIV positive in any survey. Approximately half (51%) reported a low perceived risk of contracting HIV in both 2005 and 2009. Almost a quarter (24%) of these women tested HIV-positive in 2009. Only 11% of the sample perceived themselves at high risk of contracting HIV in both 2005 and 2009. The difference in proportions of those who were HIV-positive among these women and those with low perceived risk in both years was small and not statistically significant (32% vs 24%, p = 0.31). Overall, the proportion testing HIV positive across all four transition categories was statistically similar (χ² = 1.6; p = 0.68).

Notably, we found no association between perceived risk in 2005 and attrition from CAPS between 2005 and 2009 (Additional file 1: Supplemental Digital Content 3), suggesting that differential attrition did not bias our findings.

Table 3 presents the results from the multivariable logistic regression analysis examining correlates of self-perceived HIV risk reported in 2009 (bivariate estimates, which were, in general, substantively similar, are provided in Additional file 1: Supplemental Digital Content 4). In the full estimation sample (all available observations for African women - Column 1), three factors relating to sexual behaviour were substantively associated with perceived risk. Women who reported using condoms consistently with their most recent partner were less likely to perceive themselves at moderate/great risk (aOR: 0.57; 95% CI: 0.36, 0.89; p = 0.02). Women were more likely to perceive themselves at risk if they had had three or more lifetime partners (aOR: 2.37; 95% CI: 1.53, 3.73; p < 0.01) or their most recent partner was five or more years older, i.e. an age-disparate partnership (aOR: 1.73; 95% CI: 1.09, 2.76; p = 0.02). We found positive associations between perceived risk and both perceptions about partner concurrency and alcohol consumption, but these associations were small and not statistically significant. We found no association between risk perceptions and age. Several of the socio-psychological factors were associated with perceived risk. Women who had heard that male circumcision reduces a man’s HIV infection risk perceived themselves less likely to get HIV (aOR: 0.38; 95% CI: 0.22, 0.64; p < 0.01). A negative association was found between symbolic stigma and perceived risk: compared to women who did not report stigma on any item, those who reported stigma on one item had slightly lower odds of perceiving themselves at risk (aOR: 0.69; 95% CI: 0.37, 1.28; p = 0.24) and those that reported two or more stigma items had about half the odds of perceiving themselves at risk (aOR: 0.53; 95% CI: 0.26, 1.09; p = 0.09). Sensitivity analysis using a stigma score (a 0–12 scale based on summing 5-point Likert scale responses) confirmed a negative relationship (aOR: 0.91; 95% CI: 0.84, 0.99; p = 0.03).

Women who reported knowing someone living with HIV or who had died of AIDS were more likely to perceive themselves at risk (aOR: 2.27; 95% CI: 1.36, 3.78; p < 0.01). Women who reported knowing their partner’s HIV status were less likely to perceive themselves at moderate/great risk (aOR: 0.67; 95% CI: 0.43, 1.07; p = 0.09). Finally, previously having had a HIV test was associated with much lower odds of perceiving oneself to be at risk (aOR: 0.21; 95% CI: 0.13, 0.35; p < 0.01).

The analysis restricted to HIV-negative women (Table 3, column 2) found substantively similar associations between perceived HIV risk and having had three or more lifetime partners (aOR: 2.68; 95% CI: 1.54, 4.67; p < 0.01), hearing that male circumcision reduces male HIV risk (aOR:0.31; 95% CI: 0.17, 0.59; p < 0.01), reporting stigma on 2/3 items (aOR: 0.47, 95% CI: 0.20, 1.12; p = 0.09), knowing someone living with HIV or who had died of AIDS (aOR: 2.42; 95% CI: 1.28, 4.60; p < 0.01) and having had an HIV test (aOR: 0.23; 95% CI: 0.13, 0.39; p < 0.01).