Individual and community-level risk factors for HIV stigma in 21 Zambian and South African communities: analysis of data from the HPTN071 (PopART) study

Supplemental Digital Content is available in the text


Introduction
HIV stigma is present whenever HIV infection is linked to negative stereotypes that mark a person living with HIV as different from rest of the population; a separation of 'them' from 'us.' This separation then leads to status loss, which can result in negative outcomes for people living with HIV (PLHIV) [1]. Stigma experienced by PLHIV can include being gossiped about, insulted or physically assaulted in communities and healthcare settings [2]. Internalized stigma occurs whenever PLHIV apply the same negative feelings to themselves and can have mental health consequences [3][4][5][6]. HIV stigma infringes human rights and can inhibit access to HIV testing and care [7,8].
Few studies have compared data both from those whose beliefs and behaviours are thought to drive the stigmatization process and also from those who experience it. The community-level factors that give rise to stigma are under-studied [9,10]. Stigma theories suggest that the beliefs and behaviours of community members and health workers (HW) are drivers of stigma [11,12], but there are few quantitative data to support this.
We analysed baseline data from a large cohort study of HIV stigma nested within the HPTN 071 (PopART) trial [13,14]. The outcomes of interest were experienced and internalized stigma reported by PLHIV. We first explored individual-level risk factors. We then investigated the hypothesis that stigma reported by PLHIV was more common in communities with higher levels of fear and judgement towards PLHIV. Finally, we investigated whether stigma reported by PLHIV was more common in communities with more perceived stigma reported by community members and HW.

Methods
Twenty-one urban communities (9 in South Africa, 12 in Zambia) were purposively selected to take part in the HPTN 071 (PopART) cluster-randomized trial. The trial tests the impact of a combination HIV prevention package, including universal door-to-door HIV testing and offer of antiretroviral therapy (ART) regardless of CD4 þ cell count, on HIV incidence. Using a 'parallel' approach [15,16], we combined outcome and exposure data from three separate populations who were interviewed in two data collection activities (see Appendix S1, http://links.lww.com/QAD/B222 and the following for further details).

Stigma outcome measurement
Outcome data came from individuals recruited to the HPTN 071 (PopART) Population Cohort who both selfreported living with HIV and were laboratory-confirmed as HIV-positive. We refer to this group as PLHIV. HIV status was determined by testing blood samples drawn from consenting survey participants. Blood samples were analysed in-country using a single fourth generation assay (Architect HIV Ag/Ab Combo Assay, Abbott Diagnostics, Delkenheim Germany). Further testing was performed at the HPTN Laboratory Center (Baltimore, Maryland, USA). Samples that had reactive results incountry were tested with a second fourth generation assay (GS HIV Combo Assay, Bio-Rad Laboratories, Redmond, Washington, USA). Samples with discrepant/ discordant test results were tested with additional assays to determine HIV status. The cohort was enrolled between November 2013 and March 2015. In each community, household listing generated a sample frame [13]. The target sample size was 2500 individuals per community, of whom 15% were expected to be living with HIV. In randomly sampled households, one adult resident aged 18-44 years was selected at random. Participants completed an interviewer-administered questionnaire with data captured on an electronic device. Participants were asked if they had an HIV-test previously, and if comfortable to do so, to share the result of their last test [13]. Participants were also offered voluntary counselling and testing using rapid HIV-test kits. Individuals testing positive were referred to a government health facility.
PLHIV were asked about their experiences of stigma. Item and response wording are held in Table S1 (Appendix S1, http://links.lww.com/QAD/B222). Item wording was informed by previous harmonization on measures of HIV stigma [12]. PLHIV responded to three items on internalized stigma (see Table 2 for item and response wording). Responses were summarized into a binary variable describing whether participants agreed to feeling any of three manifestations of internalized stigma. Five items captured experienced stigma in a community setting, and three captured stigma experienced in healthcare settings. Precoded response categories identified the frequency of experiences during the last year. These items were collapsed to create two binary variables capturing experience of stigma, in the community or in a healthcare setting, during the last year. Three thousand eight hundred and fifty-nine PLHIV had complete data on all 11 stigma items and on all sociodemographic variables (Figure S1a, http://links.lww.com/QAD/ B222).

Exposure measurement
Individual-level exposure data came from the same interviews with PLHIV as the outcome data. Exposures considered included sex, age, education, marital status, HIV treatment (i.e. ever started ART), HIV-status disclosure, sexual behaviour and household wealth. We also measured community-level characteristics reflecting the level of HIV fear and judgement and the perceptions of stigma reported by community members and HW. The HIV fear and judgement items captured participants' attitudes towards PLHIV. The perceptions of stigma questions reflected whether participants perceived that stigma was occurring rather than reflecting their attitudes. Data were collected from a random 20% sample of the Population Cohort described above. We included data from participants who did not have a confirmed HIV-positive blood test or self-report being HIV-positive (5088 individuals, range 161-441 per community). We refer to this group as community members (CM). We asked CM about their fears and judgement toward PLHIV (three items), levels of perceived stigma in communities (five items), and levels of perceived stigma in healthcare settings (two items) (Table S1 and Figure  S1b, http://links.lww.com/QAD/B222). Each question was asked on a 4-point Likert scale scored as follows (strongly disagree 0, disagree 1, agree 2, strongly agree 3). Three scores were calculated for each individual as the mean of the item responses. Each score could theoretically range from 0 to 3, with 0 representing all items being responded to as 'Strongly Disagree' and 3 representing all 'Strongly Agree'. Cluster summary variables were calculated as the mean of the individual responses, with higher scores representing communities with a greater presence of stigmatizing attitudes or a higher level of perceived stigma. Thus, for any community, a score of '1' would mean that the average response to all items across all individuals was to 'Disagree' with the statements.
Data on the beliefs and perceptions of HW came from the baseline survey for a separate cohort study conducted as part of the trial [15]. We recruited consenting healthfacility staff and community HW delivering HIV-related services. We also collected data from new trial intervention staff (known as 'CHIPS' [13]) but excluded these from this analysis as these individuals had only just begun to work in the communities at the time of data collection. We included data only from HW who did not self-report being HIV-positive. We refer to this group as HW. Again, three scores were developed reflecting HIV fear and judgement (five items), perceptions of the stigmatizing behaviours of their co-workers (four items) and perceptions of stigma in the community (five items). Scoring at individual and community levels was as above. Some 851 HW contributed data to this analysis (range 13-77 per community; (Table S1 and Figure S1c, http://links.lww. com/QAD/B222).

Statistical analysis
We summarized PLHIV characteristics in each country and describe variation in stigma prevalence by cluster (range). There was one cluster with a low sample size 786 AIDS 2018, Vol 32 No 6 (n ¼ 5 PLHIV) leading to outlier values. Wherever relevant we present the outlier value separately, and the cluster range excluding this value. We calculated Cronbach's alpha to assess inter-item agreement.
In risk-factor analysis, we assessed whether both individual-level and cluster-level characteristics were associated with each of the three PLHIV stigma outcomes (internalized, experienced in the community, experienced in a healthcare setting). We used logistic regression and report the odds ratios and 95% confidence intervals and Wald-test values for each risk factor for each of the three outcomes in turn. Regression analyses were carried out excluding categories wherever a response was 'Don't know' (ever started ART) or missing (time since first positive HIV test, first time had sex, number of sexual partners and condom use). We examined the impact of missing data on these four risk factors on the three stigma outcomes and found that PLHIV with missing data were less likely to report HIV stigma. We adjusted the standard errors using the vce (cluster) command in Stata v14 (StataCorp. College Station, Texas, USA) to reflect the study design, and adjusted all analyses for sex and age. In Appendix S2, http://links.lww.com/QAD/B222, we report on a sensitivity analysis restricted to 2342 PLHIV who, at the time of recruitment to the study, had not yet had a visit from the trial intervention team. We were concerned that this visit may influence stigma reporting. In summary, the prevalence of stigma was largely unchanged and although there were changes in point estimates and significance values for individual variables, there were no systematic differences of interpretation.
For individual-level risk factor analysis, we included sociodemographic and sexual behaviour characteristics that have been associated with stigma in previous analyses [17,18]. For cluster-level risk factor analysis we hypothesized that PLHIV-reported levels of internalized stigma and stigma experienced in the community would be correlated with the level of HIV fear and judgement reported by CM, and with perceived levels of community stigma reported by CM and HW. We hypothesized that stigma reported by PLHIV in healthcare settings would be correlated with the level of HIV fear and judgement reported by HW, with perceived levels of stigma in healthcare settings reported by CM, and with perceptions of stigma among co-workers reported by HW. To aid interpretation, we produced cluster-level scatter plots of the associations between the prevalence of each type of stigma and the cluster-level exposures. Each cluster was represented by a circle proportional in size to the number of PLHIV included in the analysis. We added fit lines from unadjusted, clusterlevel linear regressions of the associations weighted by the size of the PLHIV population in each cluster.

Ethics
The HPTN

Community-level characteristics
The cluster-score reflecting fear and judgement towards PLHIV reported by CM was 0.9 in South Africa and 0.8 in Zambia (Table 3), with substantial variation between clusters (range 0.4-1.2). Note that a score of 0.9 represents that across all communities the average participant response was closer to 'Disagree' (1) than to 'Strongly Disagree' (0). On average, CM also 'disagreed' with statements regarding the perception that stigma was present in communities (1.  Table 3).

Risk factor analysis
Internalized stigma was not significantly associated with sociodemographic or behavioural characteristics, except that it was reported less often by those who had been diagnosed for longer (aOR 0.75, 95% CI 0.59-0.96 and aOR 0.73, 95% CI 0.56-0.96, comparing 1-5 years and more than 5 years since diagnosis with 0-12 months, respectively). There was some evidence of more internalized stigma reported by those of higher wealth (P ¼ 0.065). Internalized stigma was more commonly reported by those reporting stigma experienced in both community and healthcare settings (aOR 4.32, 95% CI 3.47-5.37 and aOR 4.37, 95% CI 2.71-7.06, respectively; Table 4). Internalized stigma was not significantly associated with living in a community with a higher score for HIV fear and judgement held by CM (adjusted odds ratio for a unit increase in the score, aOR score 1.11, 95% CI 0.36-3.44). However, internalized stigma was significantly associated with the average level of perceived stigma reported by CM (aOR score 3.36, 95% CI 1.86-6.10). There was little evidence of an association between internalized stigma and the HWs' perceptions of the level of stigma in the community (aOR score 0.16, 95% CI 0.01-2.34; Table 5). These findings were mirrored in the cluster-level scatter plots (Fig. 1).
Stigma experienced in the community was more frequently reported by women than men (aOR 1.  The aOR for sex is adjusted for age group; the aOR for age group is adjusted for sex; the aOR for all other predictor variables are adjusted for sex and age group. b A P value of less than 0.05 indicates that the predictor creates a statistically significant improvement in the fit of the model. between community-experienced stigma and HW's perceptions of stigma in communities (aOR score 0.34, 95% CI 0.07-1.71) ( Table 5).
Stigma experienced in a healthcare setting was more commonly reported by women than men (aOR 1.64, 95% CI 1.08-2.48), and among those reporting more lifetime sexual partners (e.g. aOR 2.24, 95% CI 0.91-5.49 comparing >20 partners with those with one partner; Table 4). Odds ratios for the associations between community level characteristics and stigma experienced in a health setting had wide confidence intervals (Table 5). Despite this, there was evidence of an association such that CMs' perceptions that stigma was present in healthcare settings was associated with PLHIV reports of this (aOR score 14.93, 95% CI 3.95-56.43).

Discussion
In this large study in 21 urban communities across two countries, 35.5% of PLHIV reported some type of stigma. Most PLHIV participants were women, reflecting both higher response rates and a higher prevalence of HIV among women. Individuals reporting one form of stigma were more likely to report the other types. Experienced stigma in the community and internalized stigma were more common in Zambian communities, whereas experienced stigma in healthcare settings was more common in South African communities. There were few individual predictors of internalized stigma, but experienced stigma was associated with sociodemographic and behavioural characteristics. At cluster level, community members' (but not HW) perceptions of stigma varied substantially across communities and were associated with PLHIV experiences. However, surprisingly, CMs' reported attitudes of fear and judgement toward PLHIV were not associated with PLHIV's reported experiences of stigma.
We have undertaken the largest ever study on experiences of stigma from a random sample of PLHIV, adopting bestpractice measures of core manifestations of HIV stigma [7,12,15]. Although some PLHIV did not participate in the study or did not disclose their status, response rates were high. We have brought data from PLHIV together with independently collected data on the beliefs and perceptions of HIV stigma held by community members and HW. These fears, judgements and perceptions are thought to act as drivers of stigma in communities [19]. This 'parallel' approach to data collection has been discussed in the literature [11,15,16] but not operationalized. Aside from the strengths of our work, there are also limitations. Stigma is a sensitive subject and may have been under reported. Social desirability bias might have affected the validity of responses to beliefs and perception questions [20,21]. As the communities involved in the study were purposively selected it is unclear how generalizable our findings are to other settings in sub-Saharan Africa. Wide confidence intervals for some associations reflect few events for some outcomes, limited intercluster variation for some exposure variables and the small number of clusters [22]. Finally, results for risk factors with missing data should be interpreted with caution noting that PLHIV with missing data were less likely to report HIV stigma.
Reported experience of stigma among PLHIV in our study was lower than in studies employing the PLHIV Stigma index in South Africa [23] and Zambia [24]. In 2009, 51.8% of PLHIV reported having experienced 790 AIDS 2018, Vol 32 No 6 verbal abuse because of their HIV status in Zambia compared to 8.3% in our study [24]. In another study, 16.1% of PLHIV reported physical abuse because of their HIV status compared to 4.6% in our population [19]. Previous studies were conducted on smaller, convenience or snowball samples [25]. Individuals recruited this way may not be representative of all PLHIV [26]. Participants may be more likely to discuss stigma in the PLHIV Stigma Index studies as these are partly used to encourage reflection on life experiences living with HIV [19,23,24]. Our findings are consistent with the hypothesis that some forms of stigma may be decreasing over time and as ART access expands [27]. Stigma manifestations may also be shifting with more nuanced forms of stigma replacing overt acts of stigma and discrimination [28].
Some findings were as hypothesized. Experienced stigma was more common among those reporting more risk behaviour. Those who had been diagnosed for longer and who had disclosed to others reported more experienced stigma, perhaps reflecting their greater visibility [29,30]. They also reported less internalized stigma, perhaps reflecting having had a longer period to 'accept' their status [31][32][33]. Other findings were unexpected. Although community members' perceptions of levels of stigma were correlated with the reported experiences of PLHIV, neither their beliefs, nor the beliefs or perceptions of HW were. This may reflect misreporting of either stigma experiences, or of beliefs, because of social desirability bias [20,21]. However, our study used electronic data collection devices and sought to encourage honest reporting. Stigma reported by PLHIV might also have occurred outside the study communities or healthcare settings from which belief data were collected. Nevertheless, our results suggest caution in situations wherever reported fears and judgements are interpreted as a proxy for the experiences of PLHIV. It also underscores the role of internalized stigma in contributing to stigma experiences.
Stigma remains an important phenomenon in these study communities. Our results will inform ongoing work addressing the core hypotheses for our nested study: that the HPTN 071 (PopART) intervention may reduce levels of stigma in study communities, that stigma may undermine the effectiveness of efforts to scale up testing and treatment, or that the forms of HIV stigma may change over the period of the trial [15].