HIV-Prevalence in South Africa by settlement type: A repeat population-based cross-sectional analysis of men and women

To assess i) whether there is an independent association between HIV-prevalence and settlement types (urban formal, urban informal, rural formal, rural informal), and, ii) whether this changes over time, in South Africa. We draw on four (2002; 2005; 2008; 2012) cross-sectional South African household surveys. Data is analysed by sex (male/female), and for women by age categories (15–49; and 15–24; 25–49) at all-time points, for men in 2012 data is analysed by age categories (15–24; 25–49). By settlement type and sex/age combinations, we descriptively assess the association between socio-demographic and HIV-risk factors; HIV-prevalence; and trends in HIV-prevalence by time. Relative risk ratios assess unadjusted and adjusted risk for HIV-prevalence by settlement type. All estimates are weighted, and account for survey design. In all survey years, and combinations of sex/age categorisations, HIV-prevalence is highest in urban informal settlements. For men (15–49) an increasing HIV-prevalence over time in rural informal settlements was seen (p = 0.001). For women (15–49) HIV-prevalence increases over time for urban informal, rural informal, rural formal, and women (15–24) decreases in urban formal and urban informal, and women (25–49) increases urban informal and rural informal settlements. In analyses adjusting for potential socio-demographic and risk factors, compared to urban formal settlements, urban informal settlements had consistently higher relative risk of HIV for women, in all age categorisations, for instance in 2012 this was RR1.89 (1.50, 2.40) for all women (15–49), for 15–24 (RR1.79, 1.17–2.73), and women 25–49 (RR1.91, 1.47–2.48). For men, in the overall age categorization, urban informal settlements had a higher relative risk for HIV in all years. In 2012, when this was disaggregated by age, for men 15–24 rural informal (IRR2.69, 1.28–5.67), and rural formal (RR3.59, 1.49–8.64), and for men 25–49 it was urban informal settlements with the highest (RR1.68, 1.11–2.54). In 2012, rural informal settlements also had higher adjusted relative risk for HIV-prevalence for men (15–49) and women (15–49; 15–24; 25–49). In South Africa, HIV-prevalence is patterned geographically, with urban informal settlements having a particularly high burden. Geographical targeting of responses is critical for the HIV-response.

Introduction HIV-incidence and HIV-prevalence is spatially distributed globally, nationally, and subnationally. The spatial patterning of HIV across settings reflects inequalities in access to resources, healthcare services, and power differentials, particularly along lines of poverty, sexuality, gender and race [1][2][3]. Across Africa (excluding South Africa) a limited body of work has looked at this by settlement type. In Kenya, HIV-prevalence was assessed comparing urban slums, with urban-non slum settlements, with the HIV-prevalence 12% and 5% respectively [4]. While in Namibia's capital city, Windhoek, hotspot mapping identified particularly high HIV-incidence in informal settlement areas [5]. Understanding the spatial patterning of HIVprevalence globally, and nationally, is critical to ensure limited resources are targeted most effectively, particularly as donor funding is declining [6].
In South Africa, a number of studies have explored the spatial patterning of HIV. Studies have, for instance, sought to highlight HIV 'hot-spots'. For example, Tanser et al [7] found wide variation in HIV-prevalence (from 6-39%) in a rural demographic health surveillance area in KwaZulu-Natal, with clustering of HIV in homesteads [8], nearer national roads [7], and by transport corridors [9]. Amongst women screened for enrollment in clinical trials, also in KwaZulu-Natal, but in more urban settings, HIV-prevalence clustered in 'hot spots' with HIV-prevalence in those ranging from 56.0% to 39.0%; with distinguishing factors including being in peri-urban communities (typically including informal settlements), and reporting more lifetime sexual partners [10]. Finally, one nationally representative youth (15)(16)(17)(18)(19)(20)(21)(22)(23)(24) survey, found that HIV-prevalence was pattern by rural/urban status, with higher HIV-prevalence in urban settings, and was highest for young women in urban informal settlements (but not men) [11].
In each of these data on settlement type is collected and analysed in the main report. The 2017 data is not yet publicly available, and in the report that has been produced, settlement types have been coded differently to previous studies, combing urban informal and urban formal to create one category of 'urban'. All four previous (2002; 2005; 2008; 2012) nationally representative studies analysed data by urban formal, urban informal, rural formal and rural informal, and these analyses have described HIV-prevalence being substantially higher in urban informal settlements, compared to other settlement types [13][14][15].
However these nationally representative data have not be disaggregated settlement type by age and sex, which are important factors in the patterning of the HIV-epidemic, with younger (15)(16)(17)(18)(19)(20)(21)(22)(23)(24) women at greater risk of acquiring HIV than younger men, and women of all ages more likely to be living with HIV, compared to men [17]. Additionally, the prior descriptive work provided only prevalence estimates, without adjustment for the distribution of potential socio-demographic differences that might account these spatial differences. Nor have changes in HIV-prevalence and settlement-type been explored by time.
Drawing on representative population-based data in South Africa of four surveys (2002; 2005; 2008 and 2012), this paper has three objectives. First, to describe HIV-risk acquisition factors by settlement type and age and sex. Second, to describe HIV-prevalence by settlement-type and assess trends over time by settlement-type. Third, to assess whether settlement-type independently predicts HIV-prevalence after accounting for risk factors.

Data collection
We used four nationally representative population-based cross-sectional household surveys from 2002,2005,2008 and 2012 in South Africa. The National Prevalence Surveys employed a stratified multistage cluster sampling approach. In the first stage 1000 census enumeration areas (EAs) were selected proportional to size and stratified by province, geotype(settlement type) and race; after which a fixed number of households were selected per EA in the second stage. Sampling weights were calculated to correct for unequal probabilities of selection and non-response at the EA, household and individual level. The individual weights for each survey were benchmarked to relevant mid-year population estimates by age, race, sex, and province. This ensured that the sample for each survey was representative of the population distribution in South Africa for sex, age, race, settlement type, and province. Response rates for HIV-testing varied, with lower response rates consistently by men, compared to women (e.g. in 2012, for those older than 15, the testing response rate for females was 57.7%, compared to 42.3% in males) [14]. Detailed information about survey design, sampling methods, refusal rates, and survey administration is available in individual reports [12][13][14][15]. The comparability of design makes these four surveys nationally representative data sets, enabling assessment of longitudinal change across them.
All studies received ethical approval from the Human and Social Research Council's (HSRC) ethics committee before starting, and participants provided informed consent for questionnaires, and separately for HIV-testing. Information on this can be found in the original reports.

Outcome
The primary outcome for analysis is a binary HIV-status (positive or negative). Testing strategy varied by data collection year. In 2002 and 2005 oral saliva samples were obtained and tested using ELISA kits, with confirmatory testing on positive tests only in 2005 [12,15]. The sensitivity and specificity of the ELISA test are 99% and 99% respectively [12,15]. In 2008 and 2012, dried blood spots were collected and tested, and confirmatory tests conducted on all positive samples (and a 10% sub-set of negative samples) using a combination of three HIV-1 enzyme immunoassays, with 99% specificity and sensitivity [13,14].
Classification of households into settlement type was done through coding of Enumeration Areas (EA) using census data and classifications provided by Statistics South Africa. Four options were available: urban formal, urban informal, rural formal (including commercial farms), and rural informal, which included tribal authority areas [12][13][14][15]. According to Statistics South Africa, urban areas are continuously built-up areas, and either urban formaldefined as land proclaimed as residential where settlements are structured and organized. In addition, services are provided through local government and development is controlled. Or, urban informal, which are unplanned settlements on land not defined as residential, and comprising of informal dwellings. Rural areas are any area not deemed as urban, and comprises of rural informal, which areas are controlled by tribal authorities and informal settlements outside of urban areas. In earlier surveys this was just termed tribal authority areas. And rural formal settlements, which are formal settlements outside of urban areas, which do not fall under tribal authority areas, including commercial farms [20].

Covariates
In addition to age and sex, we identified comparable measures across all four surveys as potential co-variates, based on known risk factors for HIV-acquisition. All studies assessed current work status [employed (formal or informal), unemployed, or student]. Education level was assessed as either: none through to secondary not completed, else completed secondary or higher. For people reporting ever having had sex, age of sexual debut, and number of past year sexual partners were assessed.
We first calculated the distribution of weighted proportions of socio-demographic and HIV-acquisition risk factors by settlement type, sex, age, and survey wave, presenting percentages and 95 percent confidence intervals (95%CI). We used chi-square tests to assess variation. We then calculated weighted HIV-prevalence by settlement type for each survey by sex, and age groups. Differences are assessed through chi-squared tests. We also assessed trends in HIV-prevalence by settlement type using chi-square tests.
To assess the potential impact of settlement type on prevalent HIV infection, we used individual level poisson regression reporting relative risk with robust standard errors, taking into account sampling weights, stratification by province, and clustering within EA. We then adjusted for education, employment status, age at first sex, number of partners in past year. We report both the unadjusted and adjusted relative risk ratios. All analyses were performed in STATA15, using the survey commands, and weighted estimations.
HIV-prevalence by settlement type for men  showed consistent patterns (Table 3). HIV-prevalence in all surveys was highest in urban informal settlements, where it was almost twice that of urban formal settlements, and for 2005, 2008 and 2012, and these were significantly different. There was no difference between other settlement types, and urban formal settlements. Rural informal settlements showed an increasing HIV-prevalence over time (p = 0.001). In the age disaggregated tables (Table 3B and 3C) HIV prevalence was highest among 15-24 year olds residing in rural forma areas, followed by urban informal areas and rural informal areas. Among men 25-49 years old, the highest HIV-prevalence was among those in urban informal settlements, and lowest among those in urban formal settlements, though 95%CI did overlap. Table 4 presents the unadjusted and adjusted relative risk ratios for all men by settlement type and survey year. In multivariable regressions (Table 4A) In 2012, where it was possible to disaggregate men by age (Table 4), for men aged 15-24, compared to urban informal settlements, the relative risk of HIV-prevalence was higher in rural informal (RR2.69, 1.28-5.67) and rural formal (RR3.59, 1.49-8.64), compared to urban Table 2. 2012 Survey for men socio-demographic and risk factors, by settlement type and age disaggregation.   Table 5A). The largest proportion who had not completed secondary school was in urban informal and rural informal for all age combinations, and the smallest proportion was always in urban formal settlements (Table 5A, 5B  and 5C). Consistently, over the three age categorization the largest proportion reporting sexual debut �18 years, was in urban informal settlements, and smallest in urban formal settlements. For the combined group  and older women  this was significantly different. For women aged 15-24 (Table 5C), a greater proportion in urban informal settlements reported �2 past year sexual partners (11.0%) and this was significantly different to other settlement types, where it was <5.0%.   Urban formal n/a n/a n/a 2.1 [1.3, 3.5] Urban informal n/a n/a n/a 3.2 [1.6, 6.4] Rural informal n/a n/a n/a 3.2 [1.9, 5.3] Rural formal n/a n/a n/a 8.

PLOS ONE
HIV-Prevalence by settlement type in South Africa       there were no differences with education levels, but among 15-24 nearly three-quarters in urban informal (73.7%), and rural informal (74.0%) had not completed secondary school, compared to half (53.5%) in urban formal. For HIV-risk factors in the 2012 survey, amongst those 15-24 (Table 8B), the greatest proportion reporting two or more past year sexual partners were those in urban informal settlements (9.0%) and this was significantly different to other settlement types (4.8% urban formal; 3.3% rural informal; 1.2% rural formal; p = 0.008). For all age combinations, the greatest proportion reporting sexual debut of �18 years was in urban informal settlements, and lowest proportion in urban formal settlements, and these were all significantly different.
HIV-prevalence by settlement types, over surveys for all age categorisations for women (Table 9) showed consistent patterns. The highest HIV-prevalence was among urban informal settlements. For women   (Table 9A), compared to urban formal settlements these differences were significant at all-time points, and in 2008 and 2012, compared to urban formal settlements, rural informal settlements were also significantly higher. Similarly for women 15-24 (Table 9B), compared to urban formal settlements, urban informal settlements had a significantly higher HIV-prevalence for 2005, 2008 and 2012. And for women   (Table 9C), urban informal settlements had a higher proportion of HIV-positive women, in all years, and in 2008 and 2012, in rural informal areas also had a higher proportion of HIV-positive women, compared to urban formal areas.
In the combined age group for women  ; Table 9A) urban informal, rural informal and rural formal, all saw significantly increasing HIV-prevalence trends. In younger (15-24; Table 9B) women, there were decreasing HIV-prevalence trends for urban formal and urban informal. And for older  ; Table 9C) women, increasing HIV-prevalence trends for urban informal, rural informal and rural formal settlement types.
In adjusted analyses for women (Table 10), for all survey years, and in all age categorisations (15-49; 15-24; 25-49), compared to urban formal settlements, women in urban informal settlements had significantly higher relative risks of living with HIV. In other settlement types, compared to urban formal settlements, there was no consistent patterning, but in 2012, women in rural informal settlements had a higher relative risk for HIV-prevalence, for all age combinations (15-49; 15-24; 25-49).

Discussion
Our analysis of four representative household surveys in South Africa from 2002 to 2012, show consistent patterning around the distribution of HIV, by settlement types, with urban informal settlements showing the highest prevalence. This finding was consistent in multivariate models accounting for potential socio-demographic cofounders, and in multivariate models for all women and men aged 15-49, and for women across different age stratifications (15-24; 25-49). For men, stratification by age was only possible for the most recent survey (2012), and in adjusted models urban informal settlements had the highest relative risk of HIV-prevalence for older men , but for younger men (15)(16)(17)(18)(19)(20)(21)(22)(23)(24) adjusted relative risk HIV-prevalence was highest among rural formal settlements.
The markedly high HIV-prevalence, and consistent association between urban informal settlements and HIV-prevalence, even in adjusted models may have a number of underlying

PLOS ONE
HIV-Prevalence by settlement type in South Africa    explanations. HIV-risk behaviour showed clear patterning by settlement type, with early sexual debut more common amongst both women and men residing in urban informal settlements across all surveys (and age categorisations). Similarly, young women (15)(16)(17)(18)(19)(20)(21)(22)(23)(24) in all survey waves more often reported two or more past year sexual partners in urban informal settlements compared to other settlement types. Early sexual debut and multiple sexual partners, both of which are often driven by poverty, inequitable gender norms and violence, are key risks for HIV-acquisition [21][22][23]. Interventions targeting these factors may be important strategies.
Structural factors may also shape the disproportionate burden of HIV in urban informal settlements. Across surveys, women and men in urban informal settlements had lower educational attainment than those in formal settlements. Low education, since the early 2000s, has been recognised as a risk factor for HIV [24]. Similarly, urban informal settlements had high unemployment rates, and unemployment, via increased poverty, may be a risk factor for HIVacquisition [1]. Structural interventions focused on strengthening educational attainment, and increasing employment/reducing poverty, are potentially important interventions.
Even after adjustment for potential co-founders, informal settlements consistently had higher HIV-prevalence than other settlement types, suggesting there remained unmeasured factors. Housing in informal settlements can be unstable, and studies from North America have highlighted how housing instability is an independent risk factor for HIV and STIs [25]. It may also be that settlement type is a proxy for sexual networks; studies in North America have highlighted how dense sexual networks, shaped by poverty and racism, are crucial for HIV-transmission patterns [26]. Finally, it may also be that people acquiring HIV are transitioning into urban informal settlements as they lose access to work and formal housing, and this explains the high HIV-prevalence. We could not, however, find a consistent measure of mobility across surveys to include this as a potential cofounder.
Our analysis also highlighted two trends in HIV-prevalence by settlement type. First, increasing HIV-prevalence amongst older  women in urban informal, rural informal and rural formal, and men  in rural informal. These changes may be indicative of the rollout of ARVs, with people surviving for longer, and HIV becoming a chronic disease [27]. The changing distribution of HIV also suggests that there may be important migration patterns as people age, with movement from urban to rural areas. The decreasing HIV-prevalence amongst younger (15)(16)(17)(18)(19)(20)(21)(22)(23)(24) women in urban formal and urban informal settlements is likely indicative of the impact of improved prevention of mother-to-child transmission in South Africa [28], and also potentially the impact of HIV-prevention programming, reducing HIVincidence [16]. The lack of similar trends in rural formal and informal settings highlights important gaps in impact for both programmes.
For men in 2012 the age stratified adjusted analysis showed that among young men those residing in rural formal areas had significantly higher HIV-prevalence than urban formal. Given that men typically acquire HIV at a later age [29], it may be the failure of prevention of mother-to-child transmission interventions in rural formal areas that leads to this patterning, paralleling what is seen among young women. For older men the sustained high HIV-prevalence in urban informal settlements, similarly parallels that seen among women of all ages, and     is likely due to the rollout of ART. Importantly, given that one important pathway for HIVtransmission is often intergenerational [29], the high HIV-prevalence in older men in urban informal settlements could explain the continued high HIV-prevalence among young women in urban informal settings.

Limitations
This study has a number of limitations. In all surveys, there was relatively high refusal rates for the main survey and for HIV-testing, particularly amongst men, and while data were weighted to account for this, there may be bias. In addition, the low levels of HIV-testing prevented agedisaggregated analysis of the men's data, apart from for the 2012 data, and even in 2012 the large confidence intervals in adjusted analyses highlight the lack of precision. South Africa has high levels of mobility and internal migration, and the assumption that all participants were regular residents of the settlement type where data were collected cannot be tested. Similarly, there was variation in the timing of studies, and seasonal migration may have played a greater, or lesser role in depending on the months of data collection. In addition, only a few potential cofounders were consistently assessed in all four surveys, and key risk factors such as intimate partner violence, condom use, and migration were not included. We were unable to include the 2017 survey data as the data set had not been made public, and the categorization of  Table 9. Weighted HIV-prevalence for women by settlement type over time.   settlement type was changed, with urban formal and urban informal combined into one category, 'urban', limiting comparability across time. Finally, only HIV-prevalence rather than incidence was assessed, and as such, we cannot determine whether the pattern of HIV-incidence is the same as prevalence. Nonetheless, analysis is nationally generalizable, as the surveys were all population based samples, and analyses were weighted to census data and to account for response refusal.

Conclusions
The analyses presented have a number of implications for programming in South Africa. First, in terms of ensuring ARV coverage, and achieving 90-90-90, settlement types with the highest, and increasing HIV-prevalence (urban informal settlements, and rural informal and formal settlements) remain most poorly served. The historical legacies in South Africa remain around access to healthcare, despite attempts to address these via government. Spatially targeted access to ARVs is therefore an important priority. Second, given the concentration of urban informal settlements, in relatively geographically bounded, and dense situations, a significant effort is required around HIV-prevention programming. Current programmes such as DREAMS and She-Conquers, do recognize and target urban informal settlements [19], however, the specific challenges of interventions in these settings, and lack of well evaluated interventions delivered here, remains a major challenge to intervention programming. More widely, this analysis suggests that settlement type may be important in understanding the spatial distribution of HIV in South Africa, and elsewhere. Such spatial patterning of HIV reflects patterns of wider marginalization of those who live there, including race, gender and poverty [1,25,26]. Further investigation into how these factors intersect to shape high HIVincidence and prevalence is critical to understand how best to prevent HIV, and ensure HIVtreatment is available to all.