The Impact of Gender Norms on Condom Use among HIV-Positive Adults in KwaZulu-Natal, South Africa

Critical to preventing the spread of HIV is promoting condom use among HIV-positive individuals. Previous studies suggest that gender norms (social and cultural constructions of the ways that women and men are expected to behave) may be an important determinant of condom use. However, the relationship has not been evaluated among HIV-positive women and men in South Africa. We examined gender norms and condom use at last sex among 550 partnerships reported by 530 sexually-active HIV-positive women (372) and men (158) who had sought care, but not yet initiated antiretroviral therapy in a high HIV-prevalence rural setting in KwaZulu-Natal, South Africa between January 2009 and March 2011. Participants enrolled in the cohort study completed a baseline questionnaire that detailed their socio-demographic characteristics, socio-economic circumstances, religion, HIV testing history and disclosure of HIV status, stigma, social capital, gender norms and self-efficacy. Gender norms did not statistically differ between women and men (p = 0.18). Overall, condoms were used at last sex in 58% of partnerships. Although participants disclosed their HIV status in 66% of the partnerships, 60% did not have knowledge of their partner’s HIV status. In multivariable logistic regression, run separately for each sex, women younger than 26 years with more equitable gender norms were significantly more likely to have used a condom at last sex than those of the same age group with inequitable gender norms (OR = 8.88, 95% CI 2.95–26.75); the association between condom use and gender norms among women aged 26+ years and men of all ages was not statistically significant. Strategies to address gender inequity should be integrated into positive prevention interventions, particularly for younger women, and supported by efforts at a societal level to decrease gender inequality.


Introduction
In 2013, over 6 million adults were living with HIV in South Africa [1]; numbers are increasing following the scale-up of antiretroviral therapy (ART) [2]. In South Africa, KwaZulu-Natal has the highest adult HIV prevalence at an estimated 29% among adults aged 15-49 years old in a rural area in 2011 [2]. Early initiation of ART reduces the risk of onward HIV transmission [3]. Currently in South Africa, treatment is initiated at <350 cells/μl [4] but the CD4 threshold will be raised to <500 cells//μl in January 2015 [5]. While an estimated 75% of eligible individuals received ART in South Africa in 2011, overall less than a third of HIV-positive adults are on treatment [6]; as such, condom use remains an important positive prevention strategy.
Reported condom use among HIV-positive women and men in South Africa is higher than in the general population [7][8][9][10][11][12]. By 2005, in rural KwaZulu-Natal, HIV-positive women were already significantly more likely to report using a condom with a regular partner at last sex than HIV-negative women [13]. In South Africa, higher levels of condom use among HIV-positive women and men have been associated with being male, younger age, higher education, and urban residency [7,10,14]. Condom use has also been linked to HIV-related factors including knowledge of being HIV-positive, longer duration since diagnosis, initiating ART, disclosure of HIV status to a partner and knowing a partner is HIV-negative [8,10,13,[15][16][17][18][19][20]. HIVpositive adults' lack of condom use has been associated with having a casual partner, sex with a positive partner, alcohol use ever or before sex, substance abuse in the past month or before sex, a history of forced sex (for women and men), and with coping strategies characterised by HIV denial and HIV-related stigma [7,9,11,14,19].
Despite this evidence, few studies have evaluated the impact of gender norms on condom use among HIV-positive women and men in South Africa. Gender norms, interpreted as social and cultural constructions of the ways that women and men are expected to behave, have been identified as important social drivers of the HIV epidemic [21][22][23][24][25], with implications for HIV prevention strategies for both women and men. For women, gender norms can create and reinforce their unequal position in relationships, families, societies and public domains [26,27]. A recent South African literature review demonstrated that women's relative disempowerment in relationships with men reduced their ability to refuse sexual advances and negotiate safer sexual practices including condom use [26]. Attempts to refuse sex or insist on safer sex can result in verbal, economic, psychological, physical, or sexual abuse [22,28,29]. For men, gender norms can exacerbate concepts of masculinity that promote sexual prowess, virility, and male control over women, and frame condom use and fidelity as unmasculine [30][31][32]. However, in the post-Apartheid era, there has been growing evidence of increasing expectations of gender equality that are reshaping the gender norms that inform HIV prevention behaviours [29,33].
We investigated the impact of gender norms on condom use at last sex among a cohort of HIV-positive, ART-naïve women and men seeking HIV care at three primary health care clinics in rural KwaZulu-Natal.

Study design
The Hlabisa HIV Treatment and Care Programme is a partnership between the Department of Health (in 17 Primary Health Care clinics, (PHCs)) and the Africa Centre for Health and Population Studies in rural KwaZulu-Natal, South Africa [34]. The programme began in late 2004 and by December 2011, 20,598 adults had initiated treatment, an estimated 31% of all HIV-infected adults aged 15-49 years [2]. The Africa Centre Demographic Surveillance Area (ACDSA) is contained completely within the catchment area of the programme (see www. africacentre.ac.za).
The analyses presented here use baseline data from 632 individuals enrolled between January 2009 and March 2011 in a cohort study investigating the impact of ART on family and partner relationships and sexual behaviour of HIV-positive individuals. Details of the study design and baseline characteristics of the cohort have been described elsewhere [35,36]. Men and non-pregnant women diagnosed as HIV-positive, accessing the HIV treatment and care programme in three of the 17 PHCs, aged 18 years or older and resident within the ACDSA, were screened for study eligibility when they returned to the clinic to receive their CD4 test result. Individuals with CD4<200 cells/μl or WHO Stage IV HIV disease, consistent with national guidelines for ART-eligibility in 2009, and those with CD4>500 cells/μl were eligible to enrol in the study. The CD4>500 cells/μl cut-off was chosen to identify a group of HIV-diagnosed individuals that could be expected to have repeated measures over time before becoming ARTeligible [35]. A general introduction to the study was given each morning by study staff in the clinic waiting room and interested individuals were invited to approach the staff member. All those who met the staff member and met the eligibility criteria were invited to participate and taken through the study information sheet and informed consent process. Condom use was not a requirement for ART eligibility. Analyses were conducted on all sexually-active participants, irrespective of ART-eligibility at baseline.
Detailed information was collected on socio-demographic characteristics, socio-economic circumstances, religion, HIV testing history and disclosure of HIV status, stigma, social capital, gender norms and self-efficacy. Information regarding each of the three most recent sexual partners in the past six months was also ascertained; if the participant had not had a sexual partner in the six previous months, questions were asked about their most recent sexual partner. Participants who reported an ongoing partnership at baseline were asked additional questions about their fertility intentions with their current main partner or partners and the quality of those relationships, including questions about communication, conflict, stability, identity, and commitment [35].
Gender norms were measured by a set of 19 questions (appendix A), adapted through focus group discussions from 24 questions developed by Pulerwitz et al [37]. Although the Pulerwitz gender norms scale was originally administered to men only, on review it was considered an appropriate measure of gender norms for both sexes and therefore administered to both women and men [38]. For gender norms, the Cronbach's alpha was 0.72 for women and 0.75 for men. HIV stigma was measured by a set of 24 questions (appendix A), adapted from Sayles et al.'s 28-item scale [39]. The questions assessed the individual's perceived HIV stigma in the community and internalised HIV stigma (referred throughout the rest of the paper as stigma). For stigma, the Cronbach's alpha was 0.75 for women and 0.76 for men. An ART knowledge score was created by summing the number of ART-related questions answered correctly from a set of 8 questions developed for the study (appendix A). The physical violence questions were adapted from the physical assault scale of the 'Revised Conflict Tactics Scale (CTS2)' [40], and the social support from their partner was measured as a marker of 'relationship quality' using a set of 10 questions adapted from the 24-item Social Provision Scale (appendix A) [41]. The relationship quality scale was asked for each of the participant's main partners. Participants with multiple partners were asked about the quality of each relationship for all partners they considered main partners. Among the 33 partnerships reported by participants with multiple partners, 11 (33%) partnerships did not have a relationship quality score because participants did not consider them main partnerships. The questionnaire was administered in a private room by study staff while participants waited to see ART clinic staff.
The questionnaire was translated into isiZulu, and reverse-translated to English independently to ensure integrity. In formative focus group discussions, it proved difficult in isiZulu to distinguish between answer options for some of the study questions, specifically between 'strongly agree' and 'agree' and similarly, 'strongly disagree' and 'disagree'. For this reason, response options were limited to 'agree', 'no opinion', or disagree' for questions with likertscale answers.
Ethics approval of the project was given by the Biomedical Research Ethics Committee at the University of KwaZulu-Natal, South Africa, and the London School of Hygiene and Tropical Medicine, and permission to conduct the study in government clinics was granted by the Provincial Department of Health in Pietermaritzburg, KwaZulu-Natal, South Africa. Individuals were enrolled after giving written informed consent.

Analysis
Study participants who reported having had sex in the year before enrolment were eligible for this analysis. Condom use at last sex within each partnership was measured in response to the question "Did you use a condom the last time you had sex with that partner". Condom use referred to male or female condom use. However, female condom use in this area is rare [42].
Scores for gender norms, ART knowledge, stigma and relationship quality were calculated by assigning a value of three for the answer 'agree', two for 'no opinion', and one for 'disagree' for questions that were designed to be affirmative, the reverse for non-affirmative questions. For gender norms, higher scores indicate more equitable norms and lower scores indicate male dominant norms. The highest score possible was 57; median score was 41, interquartile range 37 to 45. Initially, indicator variables representing gender norms quartiles were considered, but the upper two quartiles and lower two quartiles were not significantly different from each other in their estimated association with condom use at last sex (data not shown). Thus, a binary variable of gender norms, categorising scores of 41 vs. 42 was used in the models. Similarly, a higher stigma score represents greater HIV-related stigma. The maximum stigma score was 72, median score 42, interquartile range 24 to 62. Stigma was considered in the models as a binary variable, representing scores of 42 as greater stigma vs. scores of 41. Approximately half the study participants (51.7%) answered all the ART knowledge questions correctly, achieving the maximum score of 24. Thus, a binary indicator of ART knowledge, representing scores of 24 vs. 23 was used in the analyses. A higher relationship quality score represents greater social support within the relationship. The maximum relationship quality score was 30, median score 24, an interquartile range 22 to 28. A binary indicator of relationship quality, representing scores of 25 vs. 24 was used in the analyses. Participant age, age at first sex, time since HIV diagnosis, length of partnership, and partner age difference were assessed categorically. Partner age difference was calculated by subtracting the woman's age from the man's age. Some participants were not able to report whether a partner was older or younger; others were, but did not know by how many years. Only the 368 women and 168 men who reported partner age differences are included in the categorical variable of age difference.
We hypothesized that condom use at last sex would be positively associated with more equitable gender norms among both women and men, adjusting for other potential confounders of condom use and gender norms. All analyses were sex-specific. Associations with p-values less than 0.05 were considered significant. We initially compared participant-level and partnership-level characteristics according to the two groups defined by condom use at last sex, and across gender norm groups (male dominant vs. more equitable gender norms) using Chisquare tests for categorical variables and Wilcoxon rank sum tests for continuous variables. When examining participant-level characteristics, condom use at last sex for participants in multiple relationships was considered to be 'yes' if a condom was used at last sex in any partnership. A sensitivity analysis of the Chi-square associations of condom use at last sex was conducted by removing participants who responded differently for condom use at last sex for different partners.
Given that condom use may vary by partnership, logistic regression models of partnerships for women and men were used to explore participant-and partnership-level factors associated with condom use at last sex. The models included variance adjustment for correlation between participants' multiple ongoing partnerships using robust standard errors.
A priori, we hypothesized that age may modify the association between gender norms and condom use at last sex for both women and men [26,43]. Thus we initially tested for evidence of effect modification by including an interaction term between gender norms and age in logistic regression models of condom use at last sex that only included age and gender norms for each sex separately. Once effect modification by age had been established among women, age, gender norms, and their interaction were included in all multivariable models considered for women. This was not necessary for men given the lack of significant interactions. Factors identified in descriptive analysis as potential confounders of our association of interest or factors associated with the outcome were considered in the multivariable analyses. Wald tests were used to determine which variables remained in the final multivariable models.
Additional analyses focused on relationship quality, conflict, and fertility intentions as potential confounders of the association of interest using the subset of partnerships that were ongoing at baseline. All analyses were conducted using Stata, version 11.2 (StataCorp, College Station, Texas, USA).

Results
Among the 632 participants, 530 reported having had one or more sexual partners in the year pre-enrolment, for a total of 550 partnerships. The median partnership length was 5 years (IQR 2-10); median age difference within partnerships was 4 (IQR 1-7) for women and -4 (IQR -8-0) for men; condoms were used at last sex in 58% of partnerships. Seventy-two (of 530) participants reported to no longer be in a partnership at baseline, 418 were in monogamous sexually-active partnerships, 26 in an ongoing but not currently sexually active (participant-defined) partnership, and 14 reported being sexually active with more than one partner. Of the participants reporting more than one partnership, two were women and 12 were men. Of these 14 participants, four responded differently for condom use at last sex for their different partners. One of these four participants was a woman; the other three participants were men. In sensitivity analysis that dropped these four participants, the estimated associations of participant level characteristics with condom use at last sex, results did not materially change. Table 1 shows participant characteristics stratified by sex. Overall, 372 participants were women (70%), median age was 33 years (IQR 27-40): 31.5 years (IQR 26-38) among women, 37 years (IQR 31-44) among men (p<0.001). In addition to being statistically significantly older, men were less likely to have achieved secondary school or higher (56% vs. 71%), more likely to be ART-eligible (80% vs. 52%), be currently employed (34% vs. 19%), not have a current partner (18% vs. 12%), not always have lived within the ACDSA (65% vs. 51%), and have learned their HIV-positive status within the last year (78% vs. 52%), than women. The proportion with higher stigma did not differ significantly between women and men (p = 0.61), nor did the proportion with more equitable gender norms (p = 0.18). The reliability of the scales for stigma and gender norms for both sexes was good: for stigma, the Cronbach's alpha was 0.75 for women and 0.76 for men and for gender norms, 0.72 for women and 0.75 for men. Education was positively associated with more equitable gender norms among both women (p<0.001) and men (p = 0.002), while age and ART knowledge were negatively associated with more equitable gender norms (Table 2). Stigma (p = 0.003) and time since HIV diagnosis (p = 0.003) were negatively associated with condom use at last sex among women. More equitable gender norms were positively associated with condom use at last sex among men (p = 0.02), but not women (p = 0.27). Higher education (p = 0.01), being currently employed (p = 0.002), and currently having a partner (p = 0.02) were also positively associated with condom use at last sex among men. Among both sexes, ART-eligibility (measured by CD4 count and/or clinical status) was not associated with condom use at last sex and thus not a potential confounder of the association between gender norms and condom use (women p = 0.24, men p = 0.82). Table 3 stratifies partnership characteristics by sex. Women were significantly more likely than men to report alcohol use at the time of last sex by them or their partner (14% vs. 11%), have a partner older than themselves (92% vs. 18%), have had their partner insist on having sex when they didn't want to (27% vs. 13%), have ever refused to have sex with their partner (50% vs.18%), believe their partner had sex with someone else in the last six months (46% vs. 10%), have a currently employed partner (71% vs. 24%), and not be living with their partner (50% vs. 39%). There was no significant difference by sex in the proportion that had disclosed their HIV status to their partner (p = 0.55). However, significantly more men than women reported knowing their partner's HIV status (50% vs. 36%). Within partnerships, condom use (ever, at first sex, and at last sex) did not significantly differ between women and men. Table 4 shows partnership characteristics of women and men across gender norms groups and between groups defined by condom use at last sex. Among partnerships reported by women, condom use ever (p = 0.001) and at first sex (p<0.001), not disclosing HIV status to partner (p = 0.03), and knowing partner is HIV-negative (p = 0.003) were positively associated with more equitable gender norms. Condom use ever (p<0.001) and at first sex (p<0.001), no alcohol use at last sex (p = 0.04), never having unwanted sex with partner (p = 0.03), disclosure of HIV status to partner (p<0.001), and knowledge of partner's HIV status (p<0.001) were positively associated with condom use at last sex for women. Condom use at last sex among women was negatively associated with reports that their partner had definitely had sex with someone else in the last six months (p = 0.04). Partner age difference was not significantly associated with condom use at last sex among either sex.
Among partnerships reported by men, condom use ever (p = 0.003), at first sex (p = 0.001), and at last sex (p = 0.008), and no alcohol use at last sex (p = 0.05) were positively associated with more equitable norms. Condom use at last sex was positively associated with ever using a condom within the partnership (p<0.001), disclosure of HIV status to partner (p<0.001), and knowledge of partner's HIV status (p = 0.003) for men.
In models that included age, gender norms and their interaction, age was found to be an effect modifier of the association between gender norms and condom use at last sex for women (Wald p = 0.01), but not for men (Wald p = 0.90). In the final multivariable model for women, being sexually active with their current partner and disclosure of HIV status to their partner remained significantly associated with increased odds of a condom being used at last sex, while having higher levels of stigma, having unwanted sex with partner, and not knowing partner's HIV status remained associated with significantly lower odds of condom use at last sex ( Table 5). The interaction between age and gender norms remained significant in the final logistic regression model for women. However, the only significant odds ration (OR) estimate from the interaction was for women <26 years with equitable gender norms compared to the same age group with male dominant gender norms, (OR 8.88, 95% CI 2.95-26.75). Table 5 also reports the univariate OR estimates for the variables that remained in the final model and shows little confounding between the variables included in the final model. The multivariable Condom use at last sex for participants in multiple relationships was considered to be 'yes' if a condom was used at last sex in any partnership. c Sexually active / not sexually active is participant-defined.  In the final multivariable model for men, currently employed and disclosure of HIV status to their partner remained significantly associated with increased odds of condom use at last sex (Table 5). Education also remained in the model as a confounder. Gender norms were not significantly associated with condom use at last sex among men in the final multivariable model. Univariate OR estimates are also shown for the variables that remained in the final model (Table 5).
In the subset of main partnerships ongoing at baseline (N = 467, Table 4), women were significantly more likely than men to have had their partner ever use physical violence against them (14% vs. 4%) and to report lower relationship quality (58% vs. 46%); men were significantly more likely than women to have used physical violence against their partner (20% vs. 7%) and to want more children with their partner (40% vs. 30%). Among women, higher relationship quality was positively associated with condom use at last sex, but did not remain significant when added to the final multivariable model for women (Table 5). The reliability of the relationship scale for both sexes was good. The Cronbach's alpha for relationship quality was 0.76 for women and 0.74 for men. Among men, having recently argued (p = 0.01) and having ever used physical violence towards a partner (p = 0.03) were positively associated with condom use. However, neither variable provided significant additional contributions to the multivariable model for men in Table 5.

Discussion
We found younger HIV-positive women reporting more equitable gender norms were significantly more likely to have used a condom at last sex than those with male dominant gender  norms. Although the association between equitable gender norms and condom use is clear for women aged 25 or below it was less so in the older age groups of women and men of all ages. The lack of an association between gender norms and condom use in men and the over 25 year old women may reflect a difference in the way HIV-positive men and older women negotiate condom use.  While gender inequality is a recognised important driver of the HIV epidemic in Africa [44], few studies have measured its association with condom use among HIV-positive women and men. One intervention study in South Africa found that, irrespective of HIV status, condom use was significantly associated with condom negotiation, and condom use with a primary partner increased among women participating in a woman-focused HIV intervention [45]. A study of women aged 18-49 years in Botswana and KwaZulu-Natal used a large age difference within partnerships as a proxy for gender power imbalance. Condom use was associated with gender power imbalance for both sexes; women with partners more than 10 years older were less able to suggest using condoms to their partners, and men were more likely to refuse condoms when the age difference within a partnership was large [46]. This study did not adjust for participant age. In our cohort, very few (N = 75) partners had an age difference of 10 years or more. An evaluation of the Stepping Stones gender-focused project involving young adults aged 15-26 years did not observe any change in condom use at last sex among women or men in the intervention compared to the control group [47]. Identifying the key elements of gender transformative interventions remains critically important in terms of supporting positive prevention strategies.
Just over half of the participants in our study reported using a condom at their last sex act, consistent with data on condom use with regular partners from the population-based surveillance programme in the same area [13], and data collected among rural HIV-positive adults elsewhere in KwaZulu-Natal [7]. However, reported condom use in this cohort is lower than among HIV-positive individuals seeking treatment in urban areas of South Africa [9][10][11]. Our data demonstrates a negative association between condom use and previous experience of unwanted sex among women, consistent with literature on the association between intimate partner violence and low condom use in South Africa [48]. Stigma was also negatively associated with condom use among women. Interestingly, women's experience of stigma was no higher than men's in this cohort, in contrast to reports from other countries [49]. In Kenya, Mugoya et al. found that HIV-related knowledge was significantly inversely associated with stigma levels for both women and men; and it is possible that the lack of significant sex differences in stigma scores for our study participants reflects a high level of HIV-related knowledge.
Consistent with literature that has identified higher socioeconomic status as a predictor of increased condom use in KwaZulu-Natal, employment was a strong predictor of condom use for men [50]. Education was an important confounder of the relationship between gender norms and condom use among men, suggesting that education is key to reducing gender inequity and preventing HIV in rural South Africa. In contrast to previous studies, having recently argued and ever use of physical violence towards a partner were positively associated with condom use among the HIV-positive men in ongoing partnerships [23,47]. Further exploration is needed to determine if this finding is an artefact of the data.
Our findings demonstrate that both HIV-positive women and men who disclosed their status to their partner or who knew their partner's HIV status were significantly more likely to use a condom at last sex [7,9,10,29]. Interestingly, whereas the partner's HIV status was not significantly associated with condom use, knowledge of partner's HIV status was important, suggesting that communication between partners plays a critical role in determining condom use. It is encouraging that the majority of participants had disclosed their HIV status to their partner, but discouraging that the majority did not know their partners' status, typically because their partner had not tested for HIV. Interestingly, we found no association between a history of migration and condom use, nor between place of residence and condom use, despite reported condom use usually being higher among urban residents [51].
The proportion of women and men enrolled in the study are broadly representative of women and men attending the local ART programme and the sex-ratio of those on ART and in pre-ART care [13]. The trend of men accessing ART at much lower rates than women in South Africa [52] may reflect earlier diagnosis through antenatal testing among women as well as men avoiding treatment to prevent appearing 'weak' [52,53].
Strengths of this study include the cohort design, range of socio-behavioural variables that uniquely capture the behaviour and attitudes of HIV-positive individuals in a region of high HIV prevalence, large sample size, and sex-stratified analysis. In addition, the cohort is broadly representative of individuals with CD4>500 cells/ml and treatment-initiators in the local ART programme [36].
There are limitations to be considered when interpreting our results. Participants were not randomly selected: they were individuals who knew their HIV status, had chosen to seek care, and agreed to participate in a cohort study designed to assess sexual behaviour among HIVpositive adults. Although the ART eligible groups in our cohort were similar in age and sex distribution to the local ART programme [36], we need to be cautious in generalising our findings to all HIV-positive women and men. It is possible that women and men attending the ART programme may be more likely to have equitable gender norms than HIV-positive women and men not in the programme. Men accessing the clinic may be more likely to have equitable gender norms than men not accessing the clinic because they may be less likely to feel that their masculinity is threatened by seeking assistance than men not accessing the clinic. Women accessing the clinic may be more empowered and therefore also have more equitable gender norms than their counterparts not accessing the clinic. In addition, all data were self-reported, and thus susceptible to social desirability bias [54]. We would expect that social desirability would have less of an effect on reporting among adults who volunteered to participate in this cohort. However, it is unclear whether social desirability bias would impact differentially for individuals with equitable or male dominant gender norms.
It is important to note that there is no gold standard for measuring condom use [55,56]; reported condom use at last sex is more accurate than estimating the rate of condom use in the last month [54]. Condom use at last sex act has also been found to be a reasonable indicator of condom use in the last week [57]. However, condom use at last sex act may be a poor indicator of consistent condom use over time [58].

Conclusions
This study highlights the importance of gender equality for condom use among young HIVpositive women in KwaZulu-Natal and the need to more consistently collect data on gender equity in studies evaluating sexual behaviour. It also illustrates the need to promote communication about HIV status between couples and challenge HIV-related stigma. Strategies to address gender equity should be integrated into positive prevention interventions targeting both women and men, and supported by efforts at a societal level to decrease gender inequality.