Spatial Variation and Factors Associated with Unsuppressed HIV Viral Load among Women in An HIV Hyperendemic Area of KwaZulu-Natal, South Africa

Abstract

New HIV infections among young women remains exceptionally high and to prevent onward transmission, UNAIDS set ambitious treatment targets. This study aimed to determine the prevalence, spatial variation and factors associated with unsuppressed HIV viral load at ≥400 copies per mL. This study analysed data from women aged 15–49 years from the HIV Incidence Provincial Surveillance System (HIPSS) enrolled in two sequential cross-sectional studies undertaken in 2014 and 2015 in rural and peri-urban KwaZulu-Natal, South Africa. Bayesian geoadditive model with spatial effect for a small enumeration area was adopted using Integrated Nested Laplace Approximation (INLA) function to analyze the findings. The overall prevalence of unsuppressed HIV viral load was 45.2% in 2014 and 38.1% in 2015. Factors associated with unsuppressed viral load were no prior knowledge of HIV status, had a moderate-to-low perception of acquiring HIV, not on antiretroviral therapy (ART), and having a low CD4 cell count. In 2014, women who ever consumed alcohol and in 2015, ever ran out of money, had two or more lifetime sexual partners, ever tested for tuberculosis, and ever diagnosed with sexually transmitted infection were at higher risk of being virally unsuppressed. The nonlinear effect showed that women aged 15 to 29 years, from smaller households and had fewer number of lifetime HIV tests, were more likely to be virally unsuppressed. High viral load risk areas were the north-east and south-west in 2014, with north and west in 2015. The findings provide guidance on identifying key populations and areas for targeted interventions.

1. Introduction

In 2014, the Joint United Nations Programme on HIV/AIDS (UNAIDS) set ambitious 90–90-90 HIV testing and treatment target to achieve the 73% composite viral suppression target by the year 2020 towards ending the epidemic by year 2030 [1]. While few countries like Australia and Botswana achieved this target [2, 3], the global public health community failed to achieve this target [4]. Therefore, in 2021 the UNAIDS Global AIDS strategy raised the targets to 95–95-95 with an overall viral suppression of 86% to be met by 2025 including prioritising sexual reproductive health and rights for women living with HIV (WLHIV), with the aim of controlling the epidemic by the year 2030 [5]. The “first 95” represents 95% of people living with HIV knowing their HIV status; the “second 95” represents 95% of people who know their HIV-positive status and are on antiretroviral therapy (ART); and the “third 95” represents 95% of HIV positive people who know their HIV status are on ART and are virally suppressed [1, 4]. At the country and global level, commitment, and resources to meet these indicators has been prioritised as the strategy was expected to prevent onward transmission of HIV and reduce HIV incidence [5, 6, 7].

In 2020, globally, 36 million adults over the age of 15 were living with HIV [4]. Out of these, 84% knew their status, 73% were accessing treatment and 66% were virally suppressed [4]. South Africa contributes approximately 22% of the global HIV burden [4, 8], and KwaZulu-Natal province is the epicentre [9, 10], where the UNAIDS targets has not been met [11, 12]. Whilst South Africa has substantially scaled-up ART provision, having the largest HIV treatment programme globally, has resulted in reducing number of HIV related death [8]. However, country level HIV prevalence of 14.0%, with an estimated 231,000 new infections remains persistently high [13], and almost a fourth of women in their reproductive ages (15–49) were HIV positive at the end of 2020 [8]. KwaZulu-Natal has the highest HIV burden with prevalence of 18.1% compared to Western Cape with a prevalence of 6.8% [14]. Heterosexual sex is the key path to HIV transmission and acquisition in this region [15], where women of reproductive age are disproportionately affected [16, 17], thus increasing the potential of mother to child transmission (MTCT) of HIV during pregnancy, childbirth, or breastfeeding [13, 18]. Thus, viral suppression is critically important among this key population for the prevention of mother-to-child transmission (PMTCT) of HIV [18, 19] and transmission to sexual partners.

Small area location-based approaches have been recommended for targeted interventions, scale up of treatment and identify spatially distributed structural and behavioural risk factors towards achieving the UNAIDS targets and to help to reduce the overall HIV burden [20]. Evidently, their exist geographic variation in the complexity of HIV epidemiological measures [21]. Therefore, spatial analysis and modelling accounting for the presence of spatial autocorrelation between observation and residual must be considered [20, 21]. Failure to account for spatial heterogeneity and possible causes could result in misleading epidemiologist, public health institutions, and policy makers. The national HIV prevalence survey among pregnant women that also examined socioeconomic factors associated with unsuppressed viral load did not account for the nonlinear effect of continuous covariates or mapped the spatial effect [22]. Therefore, the aim of this study was to determine factors associated with unsuppressed HIV viral load among women living with HIV while accounting for nonlinear effects of some continuous covariates and mapping spatial risk effect using Bayesian inference. Furthermore, the study assessed progress towards UNAIDS indicators, examined the prevalence and hotspots of unsuppressed HIV viral load among women in an HIV hyperendemic area of KwaZulu-Natal, South Africa. This study applied the Bayesian hierarchical Geoadditive model technique to identify risk factors associated with unsuppressed HIV viral load and mapping the spatial areas in KwaZulu-Natal, South Africa.

2. Methods

2.1 Sources of data, design, and procedures

This analysis was based on data from HIV Incidence Provincial Surveillance System (HIPSS) that monitored HIV related measures of HIV prevalence and incidence in association with the programmatic scale of HIV prevention and treatment efforts in a “real world” non-trial setting. The study undertook two sequential cross-sectional surveys with the first survey from June 2014 to 18 June 2015 (2014 Survey) and the second survey from 8 July 2015 to 7 June 2016 (2015 Survey). All study participants provided written informed consent and or assent, completed a face-to-face questionnaire to obtain socio-demographic, behavioural, knowledge of HIV testing, sexually transmitted infections (STI) and tuberculosis (TB) history and biological information. From a total of 600 Enumeration Areas (EAs), 591 EAs with more than 50 households were systematically selected at random, of which 221 were drawn for the 2014 Survey and 203 were drawn for the 2015 Survey. Households were randomly selected using multi-stage random sampling, were geo-referenced and one individual per household, within the age range 15–49 years old was randomly selected and invited to participate in the study. In the 2014 Survey a total of 9812 participants were enrolled, of whom 6265 were women, whilst in the 2015 Survey a total of 10,236 participants were enrolled, of whom 6341 were women. All enrolled participants had HIV antibody and viral load testing undertaken. In the 2014 Survey, 2955 were HIV positive and 2946 had viral load measurement, whilst 9 participants had missing viral load measurement. In the 2015 Survey, 2947 women were HIV positive and 2946 had viral load measurements, whilst 1 participant had missing viral load measurement.

HIPSS study was conducted in accordance with the approval by the Biomedical Research Ethics Committee of the University of KwaZulu-Natal (Reference number BF269/13), the KwaZulu-Natal Provincial Department of Health (HRKM 08/14), and the Associate Director of Science of the Centre for Global Health (CGH) at the United States Centre for Disease Control and Prevention (CDC) in Atlanta, United States of America (CGH 2014–080). Details about HIPSS study design, objectives and study and data collection procedures have been described elsewhere [10, 11].

2.2 Study population and geographic area

HIPSS was conducted in a geographically defined region of rural Vulindlela and peri urban Greater Edendale areas in the Msunduzi municipality, uMgungundlovu district of KwaZulu-Natal province in South Africa. Whilst this community has basic access to water, electricity and free health facilities, the area is characterised by high rates of unemployment, poverty, and HIV. The EAs are located between 29°39’ South and 30°17 East of KZN, covers a total of 33 wards in the Msunduzi and a part of uMngeni municipalities, in uMgungundlovu district.

2.3 Study variables

2.3.1 Dependent variable

The primary outcome variable was HIV viral load status among women living with HIV (WLHIV) in this community, which was categorised as binary outcome:

$${\mathrm{\pi }}_{\mathrm{ij}}=\left\{,\begin{array}{c}1\text{viral load}\ge 400\text{copies}/\mathrm{mL}\left(\text{unsuppressed}\right)\\ 0\text{viral load}<400\text{copies}/\mathrm{mL}\left(\text{suppressed}\right)\end{array}\right.$$E1

This threshold was used in accordance with the country revised ART treatment guideline [23, 24] as well as evidence from several studies on transmission potential at this cut off [25, 26]. Unsuppressed viral load calculation and definition was based on the composite viral suppression of all WLHIV irrespective of being on ART or not.

2.3.2 Explanatory variables

Initial data exploration to identify potential factors associated with unsuppressed viral load was established using multiple correspondence analysis and random forest analysis [27]. The explanatory variables considered in the study comprised of socio-demographic, behavioural, knowledge of HIV status and HIV testing, medical history, and biological variables. These included age, marital status, education level, community duration, migration history, monthly income, accessing health care, meal cut, income loss, place of residence, number of household members, had sex in the last 12 months, number of sexual partners in the last 12 months, number of total lifetime sex partners, forced first time sex, ever consumed alcohol, ever tested for HIV, number of lifetime HIV test, knowledge of HIV status, perceived risk of contracting HIV, exposed to TB last 12 months, ever diagnosed of TB, had any STI symptoms, ever diagnosed of STI, ever pregnant, currently on antiretrovirals (ARV) and current CD4 cell count. The variance inflation factors (VIF) was used to check for collinearity among continuous independent variables and all variables with VIF < 4 was assumed that multicollinearity was not significantly present. Also, non-linear effect of all continuous variables was also examined, of which only age, household size, number of lifetime HIV test and total number of children ever born displayed a significant nonlinear effect and were considered in the fitted model while the remaining independent variables were included as linear fixed effect.

2.4 Statistical data analysis

To account for the complex multilevel sampling design, weighted percentage and frequency were used to describe and summarise the study characteristics across both surveys. Progress towards each of the 95-95-95 indicators and composite viral suppression was estimated. Comparisons of weighted proportion of viral load status was estimated with associated 95% confidence intervals (CIs) and p values using Taylor series methods. Initial non-spatial bivariate survey logistic regression was used to test association between each background characteristics and the dependent variable using Rao-Scott chi-square test. Statistical analyses were performed using SAS (SAS Institute, Cary, North Carolina) version 9.4. Covariates with significant association at 5% significant level for each study year was included in the multivariate model.

Suppose ${\gamma }_{\mathit{\text{ijkl}}}$ denote the viral load status of women and $\mathrm{P}\left({\gamma }_{\mathit{\text{ijkl}}}=1\right)={\theta }_{\mathit{\text{ijkl}}}$ is the probability that woman $l$in household $k$ within cluster $j$ and district $i$ is unsuppressed and $\mathrm{P}\left({\gamma }_{\mathit{\text{ijkl}}}=0\right)=1-{\theta }_{\mathit{\text{ijkl}}}$ is the probability that the woman is suppressed. This assumes that the response variable ${\gamma }_{\mathit{\text{ijkl}}}$ is Bernoulli distributed. Thus, the hierarchical Geoadditive model is given as:

$$\mathit{\text{logit}}\left({\theta }_{\mathit{\text{ijkl}}}\right)=X_{\mathit{\text{ijkl}}}\beta +d_1{Y}_{\mathit{\text{ijkl}}1}+d_2{Y}_{\mathit{\text{ijkl}}2}+d_n{Y}_{\mathit{\text{ijkln}}}+d_{\mathit{\text{spatial}}}\left(g_h\right)$$E2

Eq. 2 is a semi-parametrical model, where $\mathit{\text{logit}}\left({\theta }_{\mathit{\text{ijkl}}}\right)$ is the logit link function, and $X_{\mathit{\text{ijkl}}}\beta +d_1{Y}_{\mathit{\text{ijkl}}1}+d_2{Y}_{\mathit{\text{ijkl}}2}+d_n{Y}_{\mathit{\text{ijkln}}}+d_{\mathit{\text{spatial}}}\left(g_h\right)$ is the Geoadditive predictor. Parameter $\beta$ is the vector of the linear fixed effects which we modelled parametrically. The unknown smooth function of the non-linear effect is denoted as $d_a\left(.\right),a=1,\dots .n$, which was modelled non-parametrically. $d_{\mathit{\text{spatial}}}\left(g_h\right)$is the spatial effect covariate of district $g_h$ in which a woman resides, which symbolises the unaccounted and unobserved effect that are not included in the model [28, 29, 30]. Thus, resulting in the partitioning of this spatial effect into a spatially correlated (structured) and uncorrelated (unstructured) effect, given as:

$$d_{\mathit{\text{spatial}}}\left(g_h\right)=d_{\mathit{\text{struct}}}\left(g_h\right)+d_{\mathit{\text{unstruc}}}\left(g_h\right)$$E3

The argument is that spatial effect is the proxy of most unobserved influence, under which spatial structure assumption must be followed. The structured spatial effect accounts for the assumption that location close in proximity are more likely to be correlated in respect of their outcome. While the unstructured spatial effect accounts for the spatial variation because of the effects of interminable district-level factors that are not related spatially [31, 32, 33].

The study utilised a fully Bayesian inference, hence all parameters and functions were considered as random variables and thus assigned with appropriate prior. Parameter $\beta$ was assigned vague Gaussian priors N (0, 1000). The Bayesian penalised spline (P-splines, second-order random walk smoothness prior and third-degree spline) was adopted for the unknown smooth function $d_a\left(.\right)$ [34, 35]. Borrowing strength from neighbouring locations, the intrinsic Gaussian Markov random field (IGMRF) prior as specified by Besag et al. [34] was used for the structured spatial effect $d_{\mathit{\text{struct}}}\left(g_h\right)\left(.\right)$ [36, 37]. Two regions $\left(g_h\right)\mathit{\text{and}}\left(g_i\right)$ are referred to as neighbours if they share common boundary, thus the spatial extension of random walk model was modelled by assuming the Besag-York-Mollie Conditional Autoregressive (CAR) prior given as:

$$d_{\mathit{\text{struct}}}\left(g_h\right)\mid d_{\mathit{\text{struct}}}\left(g_i\right),h\ne 1\sim N\left(\frac{1}{w_{g_h}}{\sum }_{g_i\in g_h}{d}_{\mathit{\text{struct}}}\left(g_i\right),\frac{1}{w_{g_h}{{\tau }^2}_{\mathit{\text{struct}}}}\right)$$E4

Where $w_{g_h}$ is the number of neighbours in district $g_h$, and $g_i\in g_h$ represents that $g_i$ is a neighbour of district $g_i$. Thus, the conditional mean of $d_{\mathit{\text{struct}}}\left(g_i\right)$is the average function of $d_{\mathit{\text{struct}}}\left(g_h\right)$ of neighbouring districts.

Independent and identically distributed random variable (i.i.d) Gaussian priors were assigned to the unstructured spatial effect to account for the unobserved covariates that are inherent within the districts, denoted as:

$$d_{\mathit{\text{unstruc}}}\left(g_h\right)\sim \mathrm{N}\left(0,\frac{1}{{{\tau }^2}_{\mathit{\text{struct}}}}\right)$$E5

where the variance ${{\tau }^2}_{\mathit{\text{struct}}}$ is the unknown parameter to be estimated. Hyperpriors defined as log-gamma $m,n$ distribution, where $m,n=1\mathit{\text{and}}n=0.001$ were assigned at the second stage of the hierarchy. Non-linear and spatial effect were imposed with a sum-to-zero limit in order to distinguish between the effects and intercepts.

Lastly, the posterior distributions of all the parameters $\pi \left(\theta \right)$and the likelihood function $L\left(x|\theta \right)$was estimated. The study then assumes that $\theta$ denotes vectors of the unknown parameters in the model and likelihood $L\left(.\right)$ is the product of individual likelihood. Thus, the posterior distribution is written as:

$$\pi \left(\theta |x\right)\alpha L\left(y|{\beta }_1,d_1,\dots ..,{\beta }_n{d}_n,\phi \right){\prod }_{h=1}^p\pi \left({\beta }_h|d^2\right)d_h^2$$E6

This is a high dimensional model and analysis which sometimes require good knowledge of advance mathematical and statistical computation. So, Markov chain Monte Carlo (MCMC) algorithm is required to generate samples from this distribution which comes with much computational difficulties. To circumvent this problem and difficulties, the Integrated Nested Laplace Approximation (INLA) was used to obtain the estimate [38, 39]. The outmost goal is to estimate marginal posterior distribution for the latent Gaussian model which was used to compute the summary statistics of interest like posterior mean, standard deviation, and 95% credible interval.

Three models were considered for comparison namely:

Model 1: Generalised Additive model (GAM): All categorical and some continuous variables were modelled as linear fixed effect, and nonlinear effects of covariates age, household size, total number of children ever born and number of lifetime HIV test.

Model 2: Structured Additive model (SAM), extension of GAM with the inclusion of CAR prior.

Model 3: Unstructured Geoadditive model (UGM), Model 2 with the inclusion of the spatial effect and modelled using i.i.d.

Deviance information criterion (DIC) of each model were compared. The final Geoadditive model was selected based on smallest DIC which was considered as good predictive performance and best fit model [40, 41]. The summary results give the posterior mean estimates with associated credible interval as well as the spatial effect map. The enumeration area shapefile was created in ArcGIS using the geographic attributes. Bayesian inference was analysed using INLA package in R software [37, 42].

3. Findings

3.1 Study characteristics

Table 1 shows the sample size and characteristics of HIV positive women in rural and peri urban areas of KwaZulu-Natal, South Africa. Almost half (45.2%) of the women had unsuppressed viral load in 2014 and about one third (38.1%) in 2015. Majority of WLHIV were aged between 20 and 44 years; 86.9% in 2014 and 85% in 2015 with median age and interquartile range (IQR) of 31(25–39) in 2014 and and 32(26-40) years in 2015. Majority of the women were never married; 84.6% in 2014 and 81% in 2015. More than half had incomplete high school education 53.5% in 2014 and 57.3% in 2015. Most women had always lived in the community; 76.5% in 2014 and 54.8% in 2015 whilst never being away from home in the last 12 months was 89.8% in 2014 and 92.7% in 2015. In 2014 75.5% and in 2015, 53.0% of women reported a monthly income of ≤R2500 More than half of women sampled (57.6%) in 2014 were from rural area whilst the majority (63.8%) in 2015 were from urban areas. Overall 77.2% in 2014 and 83.5% in 2015 had engaged in sex in the last 12 months, whilst 46.7% in 2014 and 22.1% in 2015 reported having had two or more number of sex partners in the last 12 months. Overall, the majority; 77.8% in 2014 and 84.3% in 2015 reported having had two or more lifetime sex partners. Almost all the women were not forced to have sex at their first-time sex encounter. Regarding their HIV testing knowledge and perception, 88.9% of women in 2014 and 98.9% in 2015 reported having had an HIV test with 62.7% in 2014 and 69.8% had HIV test more than twice in their lifetime. In 2014, 21.5% had a perception of not likely to contract HIV, while only 14.1% in 2015. Overall, 79.6% of women in 2014 and 88.2% in 2015 had reported having been pregnant in their lifetime. Less than half, 48.8% in 2014 reported to be on ART, though this increased to 59.8% in 2015. More than half of the women 55.8% in 2014 and 58.6% in 2015 had a current CD4 cell counts of ≥500 cells per μL and 23.1% in 2014 and 21.6% in 2015 had CD4 cell counts of <350 per μL.

Characteristics	2014 Survey^α	2015 Survey^β
Total	2955	2948
Age median (IQR)	31 [25–37]	32 [26–39]
Socio-demographic characteristics
	n (%)	n (%)
Age groups (in Years)
15–19	131 (4.6)	133 (4.9)
20–24	436 (14.3)	337 (11.2)
25–29	578 (20.3)	606 (20)
30–34	561 (20.7)	674 (21.1)
35–39	517 (18.4)	510 (18.8)
40–44	426(13.2)	431 (13.9)
45–49	306 (8.5)	257 (10.2)
Marital Status
Never married	2468 (84.6)	2364 (81)
Ever married	478 (15.4)	584 (19)
Level of Education
No Schooling	137 (2.9)	37 (1.4)
Incomplete High School	1525 (53.5)	1688 (57.3)
Complete high school	1293 (43.7)	1223 (41.2)
Duration in Community
Always	2290 (76.7)	1504 (54.8)
Moved here less than 1 year ago	88 (2.3)	122 (3.8)
Moved here more than 1 year ago	577 (21.0)	1322 (41.4)
Away from home last 12 months
Yes	311 (10.2)	217 (7.3)
No	2644 (89.9)	2731 (92.7)
Monthly Income^a
No income	602 (17)	50 (1.3)
≤ R2500	2177 (75.5)	1657 (53.0)
> R2500	169 (7.5)	1241 (45.8)
Run out of money^b
Yes	682 (24.5)	1477 (50.8)
No	2273 (75.5)	1469 (49.2)
Meal cut^c
Yes	606 (22.1)	1330 (46.2)
No	2342 (77.9)	1616 (53.8)
Accessing health care^d
Yes	1216 (44.9)	2168 (73.3)
No	1732 (55.1)	778 (26.7)
Place of Residence
Rural	1009 (57.6)	954 (36.2)
Urban	1946 (42.4)	1994 (63.8)
Behavioural characteristics
Had sex in the last 12 months
Yes	2218 (77.2)	2501 (83.5)
No	737 (22.8)	447 (16.5)
Number of sex partner in the last 12 months
1 partner	1178 (53.3)	1824 (72.9)
2 or more partners	1034 (46.7)	677 (27.1)
Number of lifetime sex partner
1 partner	533 (22.2)	455 (15.7)
2 or more partners	1844 (77.8)	2429 (84.3)
Forced first time sex
Yes	72 (2.5)	100 (3.3)
No	2831 (95.9)	2833 (96.1)
Do not remember	52 (1.6)	15 (0.5)
Ever consumed alcohol
Yes	390 (11.5)	536 (18.5)
Never	2564 (88.5)	2412 (81.5)
HIV knowledge and risk perception
Ever tested for HIV
Yes	2513 (88.9)	2868 (96.9)
No	442 (11.1)	80 (3.1)
Number of lifetime HIV test
Never	442 (11.1)	81 (3.1)
1 time	769 (26.2)	765 (27.1)
2 or more times	1744 (62.7)	2102 (69.8)
Knowledge of HIV status
Yes	1870 (65.6)	2219 (73.7)
No	1085 (34.4)	729 (25.3)
Perceived risk of contracting HIV
Likely to acquire HIV	573 (19.3)	461 (16)
Not likely to acquire HIV	686 (21.5)	405 (14.1)
I am already infected	1696 (59.2)	2082 (70)
TB/STI history
Exposed to TB last 12 months
Yes	103 (3.6)	162 (5.9)
No	2853 (96.6)	2786 (94.1)
Ever diagnosed with TB
Yes	251 (9.6)	363 (12.7)
No	2704 (90.4)	2585 (87.3)
On medication to prevent TB
Yes	219 (9.0)	449 (14.9)
No	2736 (91.0)	2499 (85.1)
Tested for TB
Yes	1245 (47.1)	1689 (57.4)
No	1710 (52.9)	1259 (42.6)
Ever had any STI symptoms
Yes	163(4.5)	80(2.9)
No	2792(95.5)	2868(97.1)
Ever diagnosed with STI
Yes	213 (9.1)	320 (11.3)
No	2742 (90.9)	2628 (88.7)
Clinical characteristics
Ever pregnant
Yes	2346 (79.6)	2595 (88.1)
No	600 (20.4)	353 (11.9)
On ARV
Yes	1346 (48.8)	1775 (59.8)
No	1600(51.2)	1172 (40.2)
ART dosage
Single/fixed	1079 (86.3)	1580 (88.5)
Multiple	172 (13.7)	196 (11.5)
Current CD4 cell count^f
<350 cells per μL	696 (23.1)	634 (21.7)
350–499 cells per μL	639 (21.1)	576 (19.7)
≥500 cells per μL	1593 (55.8)	1729 (58.6)

Table 1.

Characteristics of HIV positive women in Vulindlela and Greater Edendale, KwaZulu-Natal, South Africa, 2014–2015.

Participants missing for: a = 7, and f = 27 in 2014; b, c and d = 2, f = 9 in 2015. No response: e: = 879(64) for 2014(2015). Missing data were excluded from percentage calculation. ZAR = South African Rand (ZAR15 ∼ US$1). TB = tuberculosis, STI = sexually transmitted infections, ARV = antiretroviral drugs, ART = antiretroviral therapy, Ever had any STI symptoms = any symptoms of abnormal vaginal discharge, burning or pain when passing urine or presence of any genital warts/ulcers.

3.2 Progress towards UNAIDS 95-95-95 indicators

Figure 1 provides the status on the UNAIDS 95–95-95 indicators. Of the 2955 women in 2014 and 2948 in 2015 who tested positive for HIV, 9 and 1 participants respectively had no viral load measurement. Thus, 2946 women in 2014 and 2947 women in 2015 had viral load measurements. In 2014, to meet the “first 95”, 65.5% (95% CI, 62.9–68.2) (n = 1890/2955) were aware of their HIV positive status and for the “second 95”, 74.2% (95% CI, 71.6–76.8 (n = 1348/1870) had initiated ART and for the “third 95”, 82.9% (95% CI, 80.4–85.4) (n = 1105/1346) had achieved viral suppression, and overall viral suppression among all HIV positive women was 54.8% (95% CI, 52.0–57.5) (n = 1574/2946). While in 2015, progress towards 95–95-95 targets were: 74.7% (95% CI, 72.7–76.6) (n = 2219/2948) were aware of their HIV status; 80.0% (95% CI, 78.1–82.0) (n = 1777/2219) of these had initiated ART and 88.2% (95% CI, 86.6–89.9) (n = 1551/1777) of those on ART had achieved HIV viral suppression, resulting in the overall viral suppression among all HIV positives to be 61.9% (95% CI, 59.7–64.1) (n = 1828/2947).

Figure 1.

Progress of the UNAIDS 95–95-95 indicators by age group and overall, among HIV positive women (2014–2015). (A). First 95: Women living with HIV who knew they were HIV positive. (B). Second 95: Women who knew they were HIV positive and were taking ART. (C). Third 95: Women who knew they were HIV positive, were on ART and had achieved HIV viral suppression at HIV viral load <400 copies/ml. (D). UNAIDS composite measure towards achieving HIV viral suppression among all HIV positive women.

Disaggregated by age groups, Figure 1a shows the progress towards the “first 95” Knowledge of HIV status increased from 65.6% in 2014 to 74.7% in 2015, and across age groups, with highest achieved among 35–39 (86.5%), 40–44 (82.4%) and 45–49 (82.4%) in 2015. Highest increase in the knowledge of HIV positive status was in the age group 15–29, increasing from 25.8% in 2014 to 46.7% in 2015. Figure 1b shows the progress towards the “second 95”. Overall proportion of women who knew their HIV positive status and were on ART increased from 74.2% in 2014 to 80.0% in 2015. The uptake of ART varied across age groups, uptake was high in the 15–19 years age group at 74.8% in 2014 and 75.9% in 2015; in ages 30–34 uptake was 77.2% in 2014 and 80.5 in 2015; in ages 35–39 years uptake was 77.8% in 2014 and 85.6% in 2015; in ages 40–44 years uptake was 77.1% in 2014 and 84.6% in 2015 and in age 45–49 uptake was 79.1% in 2014 and 84.2% in 2015. However, ART uptake in the age group 20–24 years was lowest at 62.4% in 2014 and 62.8% in 2015. Figure 1c shows the progress towards the “third 95”, that is the proportion of HIV positive women who knew their HIV positive status, were on sustained ART and who had achieved viral suppression of <400 copies per mL. Proportion varied across ages group; HIV viral suppression was lowest at 66% among 20–24 years old in 2014 and increased to 74.4% in 2015. Viral suppression of 92.9% was achieved among 45–49 years old and 91.7% among 40–44 years old and 91.8% among 35–39 years old in 2015. Figure 1d shows the overall UNAIDS 95–95-95 composite measure of achieving viral suppression of 86% among all HIV positive women. Overall, 54.8% of women in 2014 and 61.9% in 2015 had achieved HIV viral suppression of <400 copies per mL. Substantial variation existed across the age groups, with 27% among 15–19 years in 2014 and increased to 46% in 2015. Highest achievement was observed with 76% among 45–49 years old.

3.3 Prevalence of unsuppressed HIV viral load

Table 2 shows that overall prevalence of unsuppressed HIV viral load was 45.2% (95 CI, 42.5–48.0), (n/N = 1372/2946) in 2014 and 38.1% (95% CI, 35.9–40.3), (n/N = 1119/2947) in 2015. Viral suppression increased by 7.1% over the study period. Unsuppressed viral load prevalence decreased as age increased and it was 72.9% (95% CI, 62.7–83.2), (n = 95/130) in 15–19 years age group, 68.2% (95% CI, 62.4–73.9), (n = 290/433) in the 20–24 years age group, 47.3% (95% CI, 41.9–52.7), (n = 299/577) in 25–29 years age group, 43.1% (95% CI, 37.9–48.3), (n = 248/561) in 30–34 years age group, 32.5% (95% CI, 26.6–38.4), (n = 185/513) in 35–39 years age group, 36.5% (95% CI, 30.6–42.4), (n = 153/426) in 40–44 years age group, 33.0% (95% CI, 26.6–39.3), (n = 102/306) in 45–49 years age group. In 2015, prevalence also decreased by age and it was 56.0% (95% CI, 43.8–64.1), (n = 74/133); 65.1 [59.5–70.7], (n = 210/337); 46.5 [41.4–51.5], (n = 279/606); 36.4% (95% CI, 32.1–40.8), (n = 244/674); 25.2% (95% CI, 20.9–29.4), (n = 125/509); 29.1% (95% CI, 24.0–34.2), (n = 120/431); 24.0% (95% CI, 18.3–29.8), (n = 67/257) in the 15–19, 20–24, 25–29, 30–34, 35–39, 40–44, and 45–49 years age categories (Χ² trend P < 0.001).

Characteristics	2014 Survey			2015 Survey
	n/N	% [95 CI]	P-value	n/N	% [95 CI]	P-value
Overall
≥400 copies per mL	1372/2946	45.2 [42.5–48.0]		1119/2947	38.1 [35.9–40.3]
<400 copies per mL	1574/2946	54.8 [52.0–57.5]		1828/2927	61.9 [59.7–64.1]
Age median (IQR)	31 [26–39]			32 [26–39]
Socio-demographic characteristics
Age groups (years)
15–19	95/131	72.9 [62.7–83.2]	<0.0001	74/133	56.0[43.8–64.1]	<0.0001
20–24	290/433	68.2 [62.4–73.9]		210/337	65.1 [59.5–70.7]
25–29	299/577	47.3 [41.9–52.7]		279/606	46.5 [41.4–51.5]
30–34	248/561	43.1 [37.9–48.3]		244/674	36.4 [32.1–40.8]
35–39	185/513	32.5 [26.6–38.4]		125/509	25.2 [20.9–29.4]
40–44	153/426	36.5 [30.6–42.4]		120/431	29.1 [24.0–34.2]
45–49	102/306	33.0 [26.6–39.3]		67/257	24.0 [18.3–29.8]
Marital status
Never married	1188/2468	46.4 [43.6–49.2]	0.03	948/2364	40.6 [38.1–43.2]	<0.0001
Ever married	184/478	39.0 [32.3–45.6]		171/583	27.4 [23.4–31.4]
Level of Education
No schooling	63/137	44.1 [34.8–53.4]	0.09	10/37	30.2 [12.8–47.6]	0.03
Incomplete High School	672/1521	43.2 [39.3–47.0]		605/1688	35.7 [32.8–38.6]
Complete High School	637/1288	47.9 [44.4–51.4]		504/1222	41.7 [38.0–45.3]
Duration in community
Always	1078/2282	45.5 [42.7–48.4]	0.53	585/1504	39.0 [35.8–42.1]	0.04
Moved here less than 1 year ago	44/88	48.4 [34.3–62.4]		60/122	48.4 [38.1–58.7]
Moved here more than 1 year ago	250/576	43.9 [38.3–49.5]		474/1321	36.0 [33.1–39.0]
Away from home last 12 months
Yes	152/311	50.1 [41.9–58.3]	0.21	113/217	50.2 [42.6–57.7]	0.01
No	1220/2635	44.7 [41.9–47.5]		1006/2730	37.2 [34.9–39.4]
Monthly Income^a
No income	285/601	46.4 [41.0–51.7]	0.29	17/50	34.4 [19.8–49.0]	0.89
≤R2500	1017/2170	45.6 [42.4–48.8]		627/1656	38.0 [35.3–40.7]
> R2500	67/168	38.8 [30.2–47.4]		475/1241	38.3 [34.9–41.7]
Run out of money^b
Yes	318/682	47.0 [41.2–52.7]	0.04	549/1477	37.7 [34.7–40.8]	0.02
No	1051/2257	44.7 [41.8–47.5]		569/1468	38.5 [35.4–41.6]
Meal cut^d
Yes	285/606	47.4 [41.5–53.3]	0.38	499/1330	37.3 [34.0–40.5]	0.45
No	1084/2333	44.6 [41.8–47.5]		619/1615	38.8 [35.9–41.7]
Accessing heath care^d
Yes	488/1212	41.7 [37.4–46.0]	0.03	769/2167	35.5 [33.0–38.1]	<0.0001
No	881/1727	48.1 [44.4–51.8]		349/778	45.2 [41.2–49.2]
Place of residence
Rural	457/1006	44.5 [40.3–48.7]	0.51	340/954	36.3[32.1–40.4]	0.25
Urban	915/1940	46.2 [43.4–49.1]		779/1993	39.1 [36.7–41.7]
Behavioural characteristics
Had sex last 12 months
Yes	1052/2212	45.8 [42.8–48.8]	0.29	965/2501	38.8[36.4–41.3]	0.14
No	320/734	43.3 [38.9–47.7]		154/446	34.4[28.9–39.8]
Number of sex partner last 12 months
1 partner	908/1178	46.1 [43.0–49.2]	0.02	860/1824	38.2[35.6–40.7]	0.03
2 or more partners	464/1034	43.6 [39.1–47.4]		259/677	37.9[33.6–42.3]
Total number of lifetime sex partners
1 partner	265/532	48.3 [42.0–54.6]	0.01	185/455	39.4[34.7–44.1]	0.05
2 or more partners	828/1839	43.7 [40.7–46.8]		910/2428	37.9[354–40.2]
Forced first time sex
Yes	31/72	38.8 [26.1–51.5]	0.28	33/100	34.7[23.7–45.7]	0.74
No	1314/2822	45.2 [42.4–48.0]		1082/2832	38.3[36.0–40.6]
Do not remember	27/52	56.1 [39.4–72.9]		04/15	30.4[1.1–59.7]
Ever consumed alcohol
Yes	229/390	58.8 [51.7–65.9]	<0.0001	249/536	44.4[39.4–49.5]	<0.0001
No	1143/2556	43.9 [40.6–46.3]		870/2411	36.6[34.2–39.1]
HIV knowledge and risk perception
Ever tested for HIV
Yes	1086/2505	42.4 [39.5–45.3]	<0.0001	1066/2867	37.3[35.1–39.5]	<0.0001
No	286/441	68.1 [63.2–73.0]		53/80	63.4[51.4–75.4]
Knowledge of HIV status
Yes	619/1865	31.2 [28.1–34.3]	<0.0001	591/2218	25.8[23.6–28.0]	<0.0001
No	753/1081	72.0 [68.5–75.6]		528/729	74.3[70.7–77.8]
Perceived risk of contracting HIV
Likely to Acquire HIV	364/572	67.4 [62.1–72.6]	<0.0001	309/461	69.0[64.0–74.0]	<0.0001
Not likely to Acquire HIV	478/682	70.7 [66.1–74.7]		281/405	70.2 [64.8–75.5]
I am already infected	530/1692	28.9 [25.8–31.9]		529/2081	24.6[22.3–26.9]
Number of lifetime HIV tests
Never	286/441	68.1 [63.2–73.0]	<0.0001	54/81	63.7[51.8–75.6]	<0.0001
1 time	330/764	41.6 [37.1–46.2]		225/765	29.1[25.2–33.1]
2 or more times	756/1741	42.7 [39.1–46.3]		840/2101	40.4[37.7–43.2]
TB/STI history
Exposed to TB in the last 12 months
Yes	41/102	41.2 [27.8–54.6]	0.56	44/159	22.4[14.9–29.9]	<0.0001
No	1331/2844	45.4 [42.6–48.2]		1075/2786	39.0[36.8–41.3]
Ever diagnosed with TB
Yes	77/251	30.9 [23.2–38.6]	<0.0001	82/362	19.7[15.6–23.9]	<0.0001
No	1295/1295	46.8 [44.0–49.5]		1037/2585	40.8[38.4–43.2]
On medication to prevent TB
Yes	48/217	20.6 [13.6–27.6]	<0.0001	84/449	16.7[12.7–20.7]	<0.0001
No	1324/2729	47.6 [45.0–50.3]		1035/2498	41.8[39.4–44.3]
Tested for TB
Yes	417/1242	32.6 [28.8–36.4]	<0.0001	463/1688	27.1[24.7–29.5]	<0.0001
No	955/1704	56.5 [53.4–59.6]		656/1259	52.9[49.5–56.3]
Had any STI symptoms
Yes	65/162	37.3 [28.1–46.4]	0.08	30/80	37.4[23.0–51.9]	0.92
No	1307/2784	45.6 [42.9–48.4]		1089/2867	38.1[35.8–40.4]
Ever diagnosed with STI
Yes	98/213	43.7 [35.0–52.4]	0.71	155/320	47.7[41.5–53.9]	0.001
No	1274/2733	45.4 [42.5–48.3]		964/2627	36.9[34.5–39.3]
Clinical characteristics
Ever pregnant
Yes	1029/2346	42.5 [39.6–45.5]	<0.0001	947/2595	36.6[34.3–38.9.5]	<0.0001
No	343/596	59.4 [53.7–65.1]		169/352	49.3[42.6–55.9]
Currently on ARV
Yes	241/1346	17.1 [14.6–19.6]	<0.0001	220/1775	11.8[10.2–13.5]	<0.0001
No	1131/1600	72.1 [68.9–75.3]		899/1172	77.3[74.4–80.2]
ART dosage
Single/fixed	127/1077	11.5 [8.9–14.0]	<0.0001	170/1579	24.4(17.1–31.7)	<0.0001
Multiple	36/172	21.2 [12.3–30.1]		50/196	10.2(8.9–11.7)
Current CD4 cell count^e
<350 per μL	493/695	69.3 [64.8–73.8]	<0.0001	430/633	68.5[64.4–72.5]	<0.0001
350–499 per μL	315/638	49.1 [43.2–55.0]		247/576	42.0[37.4–46.6]
≥500 per μL	546/1591	33.2 [29.9–36.5]		437/1729	25.6[23.2–28.0]

Table 2.

Prevalence of unsuppressed viral load by study characteristics among women in Vulindlela and Greater Edendale, KwaZulu-Natal, South Africa, 2014–2015.

A total of nine women in 2014 survey and one woman in 2015 survey were missing viral load data. Participants missing for: a = 7, and e = 27 in 2014; b, c and d = 2, e = 9 in 2015.

Whilst unsuppressed viral load prevalence was similar across most variables, decrease in the trends over the study years was observed. In 2014 unsuppressed viral load prevalence was 50.1%, (n = 152/311) and declined to 20.2%, n = 113/217) in 2015, among women that were away from home in the last 12 months (compared to those that were never away from home; 44.7%, (n = 1220/2635) in 2014 and 37.2%, n = 1006/2730) in 2015. Among those that ever-consumed alcohol 58.8%, (n = 229/390) in 2014 and declined to 44.4%, n = 249/536) in 2015 compared to those that never consumed alcohol and 43.9%, (n = 1143/2556) (36.6%, n = 870/2411) also among those that never had HIV test 68.1%, (n = 286/441) (63.4%, n = 53/80). Among those that ever had an HIV test 42.4%, (n = 1086/2505) in 2014 and 37.3%, n = 1066/2867) in 2015 had unsuppressed viral load. Similarly, among women who did not know their HIV status 72.0%, (n = 753/1081) in 2014 and 74.3%, (n = 528/729) in 2015 compared to those who knew their status 31.2%, (n = 619/1875) in 2014 and 25.8%, n = 591/2218) in 2015 had unsuppressed viral load. Women who perceived they are not likely to contact HIV 70.7%, (n = 478/682) in 2014 and 70.2% (n = 281/405) in 2015 compared to those who already perceived they had been infected 28.9%, (n = 530/1692) (24.6%, n = 529/2081), also women who have ever been diagnosed of STI 43.7%, (n = 98/213) (47.7%, n = 155/320), among women who had never been pregnant 59.4%, (n = 343/600) (49.3%, n = 169/352) compared to those that has ever been pregnant 42.5%, (n = 1029/2346) (36.6%, n = 947/2595), likewise among WLHIV and not on ART 72.1% (n = 1131/1600) (77.3%, n = 899/1172) in comparison with those on ART 17.1% (n = 241/1346) (11.8%, n = 220/1775). Prevalence was higher among women whose current CD4 cell count were < 350 count per μ/L, 69.3%, (n = 493/695) (68.5%, n = 430/633), and those with CD4 cell count of between 350 and 499 count per μ/L 49.1%, (n = 315/638) (42.0%, n = 247/576) compared to those 500 count per μ/L 33.2%, (n = 546/1591) (25.6%, n = 437/1729) in 2014(2015) respectively.

Figure 2 shows the observed prevalence map of unsuppressed viral load. Highest prevalence was observed in the north and south of Vulindlela and east part of Greater Edendale in 2014, while in the north part of Vulindlela and the south part of Greater Edendale in 2015. The north area (Mpophomeni) showed a consistently high prevalence across both surveys.

Figure 2.

Observed prevalence maps of unsuppressed viral load among women (a) 2014 and (b) 2015 in Vulindlela and Greater Edendale area in uMgungundlovu district, KwaZulu-Natal province, South Africa.

3.4 Model diagnostic measures

Table 3 shows values of the deviance information criterion (DIC) and effective numbers of parameters (pD) for each of the fitted model. Unstructured model has the minimum values (DIC = 2593.26 and 2087.70) for 2014 and 2015 respectively, thus attesting as the best fit model for the data sets, while GAM model offers the least fit. Besides, the unstructured model is of actual interest because it contains all the variables considered, and account for spatial autocorrelation and between clusters heterogeneity, failure to do so would have produced misleading and overfitting results. Thus, further results of this study are based on the unstructured model.

	2014 Survey			2015 Survey
Parameters	GAM	Structured	Unstructured	GAM	Structured	Unstructured
DIC	2768.42	2597.64	2593.26	2097.27	2089.95	2087.70
DIC saturated	2998.34	2992.52	3004.69	2992.61	2987.95	2998.66
pD	44.92	48.88	49.23	46.53	46.50	47.55

Table 3.

Model diagnostic.

DIC: Deviance Information Criteria. pD: effective numbers of parameters.

3.5 Non-linear effect of continuous covariates on women

Figure 3 shows the non-linear effect of continuous covariates after accounting for other variables. The results shows that current age, number of household members, total number of children ever born and total number of lifetime HIV test, had a non-linear significant effect on women being virally unsuppressed in this study area. Furthermore, in Figure 3a and e shows a slight increase in effect among ages 15 to 20 in 2014 and sharp increase in 20 to 25 in 2015, after which the effect declined. Younger age 15 to 29 have higher risk of being virally unsuppressed compared to ages 30 above. Figure 3b and f shows that risk of unsuppressed viral load decreases with higher number of household members from 5 members. Also Figure 3c and h shows that the effect of total number of children ever born decreases the risk of being virally unsuppressed in 2014 but increases in 2015. Similarly, Figure 3d and g showed that the risk of unsuppressed viral load increased as the number of lifetime HIV tests increased in 2014, whilst in contrast unsuppressed viral load decreased as the number of lifetime HIV tests increased in 2015.

Figure 3.

Nonlinear effect of continuous covariate.

3.6 Fixed effect model

Table 4 displays the adjusted posterior mean estimates with their 95% credible intervals of the linear fixed effect from the multivariable model. If these intervals contain the number zero (0), then the parameter (estimate of the mean beta) is not significant; otherwise, it is significant. Factors associated with unsuppressed viral load across both years were knowledge of HIV status, low perceived risk of contracting HIV, ARV treatment and current CD4 cell counts. Women with no prior knowledge of their HIV status were more likely to be virally unsuppressed than those that knew their status. Women with either unlikely or likely perception of contracting HIV, not on ARV, and for those on ARV having multiple tablets of ARV had the highest risk of being virally unsuppressed compared to their reference categories. Additionally, in 2014 those that ever consumed alcohol were also at higher risk of having unsuppressed viral load. While in 2015, we also found that women that reported being away from home in the last 12 months, had a meal cut, being with two or more sexual partners in one’s lifetime, ever tested with TB and ever diagnosed with STI had the highest risk of being virally unsuppressed compared to their counterparts.

	2014 Survey			2015 Survey
Variables	Posterior mean	Posterior SD	95 Credible intervals	Posterior mean	Posterior SD	95% credible intervals
intercept	0.405**	0.075	(0.255, 0.551)	0.407**	0.076	(0.257, 0.557)
Marital status (ref: Ever married)
Never married	0.032	0.017	(−0.002, 0.067)	0.005	0.021	(−0.035, 0.045)
Education status (ref: Complete high school)
Incomplete high school	0.015	0.021	(−0.015, 0.045)	−0.005	0.014	(−0.032, 0.022)
Duration in community (ref: Always)
Moved here less than 1 year ago	−0.016	0.042	(−0.099,0.068)	0.044	0.033	(−0.022, 0.109)
Moved here more than 1 year ago	0.002	0.019	(−0.034, 0.039)	0.002	0.014	(−0.025, 0.029)
Away from home last 12 month (ref: No)
Yes	0.002	0.024	(−0.045, 0.048)	0.056 **	0.025	(0.007, 0.105)
Run out of money (ref: No)
Yes	0.016	0.026	(−0.035, 0.067)	0.011 **	0.018	(0.024, 0.045)
Accessing healthcare (ref: Yes)
No	0.018	0.016	(−0.013, 0.049)	0.014	0.015	(−0.016, 0.044)
Total number of sex partners last 12 months (ref: 1 partner)
2 or more partners/no res	0.024	0.019	(−0.013, 0.061)	0.002	0.019	(−0.036, 0.039)
Total number of lifetime sex partners (ref: 1 partner)
2 or more partners	0.037	0.027	(−0.016, 0.091)	0.049**	0.058	(0.102, 0.162)
Ever had alcohol (ref: No)
Yes	0.058 **	0.022	(0.016, 0.101)	0.029	0.017	(−0.005, 0.063)
Ever tested for HIV (ref: Yes)
No	−0.058	0.034	(−0.120, 0.014)	−0.022	0.045	(−0.110, 0.066)
Knowledge of HIV status (ref: Yes)
No	−0.142 **	0.030	(−0.201, −0.084)	−0.200 **	0.030	(−0.259, −0.142)
Perceived risk of contracting HIV (ref: Already infected)
Likely	0.095 **	0.025	(0.045, 0.144)	0.062**	0.028	(0.008, 0.116)
Not Likely	0.103 **	0.027	(0.050, 0.156)	0.074 **	0.029	(0.017, 0.131)
Ever tested for TB (ref: No)
Yes	−0.034	0.018	(−0.069, 0.001)	−0.070**	0.015	(−0.099, −0.041)
Exposed to TB last 12 months (ref: No)
Yes	0.061	0.041	(−0.019, 0.142)	−0.022	0.029	(−0.079, 0.035)
Diagnosed of TB (ref: No)
Yes	0.008	0.028	(−0.048, 0.064)	0.005	0.022	(−0.038, 0.048)
On medication to prevent TB (ref: No)
Yes	−0.052	0.029	(−0.108, 0.004)	−0.027	0.019	(−0.064, 0.010)
Ever had any STI symptoms (ref: No)
Yes	0.034	0.033	-(0.031, 0.100)	−0.077	0.039	(−0.084, 0.069)
Ever Diagnosed with STI (Ref: No)
Yes	0.042	0.028	(−0.014, 0.097)	0.059**	0.021	(0.018, 0.099)
Ever Pregnant (ref: Yes)
No	−0.065	0.019	(−0.103, 0.028)	0.001	0.018	(−0.156, 0.059)
on ART (ref: Yes)
No	0.321**	0.030	(0.262, 0.379)	0.511**	0.031	(0.451, 0.571)
ARV dosage (ref: fixed/single)
Multiple	0.251**	0.027	(0.199, 0.303)	0.242**	0.018	(−0.163, −0.093)
Current CD4 Cell count (cells per μl) (ref: < 350)
350–499	−0.184 **	0.021	(−0.226, −0.143)	−0.157**	0.020	(−0.197, −0.118)
≥ 500	−0.319**	0.018	(−0.354, −0.285)	−0.287**	0.016	(−0.319, −0.254)

Table 4.

Adjusted posterior means, standard deviation (SD) and 95% credible intervals for the best fitted model.

**

Significant at 5% level of significance.

3.7 Spatial effect map

Figure 4 shows the chloropleth spatial effect maps based on model 3, shows both positive and negative effects with predicted high and low risk areas of unsuppressed viral load. The colours on the chloropleth maps show the log-odds scale, indicating each area contribution to the odds of unsuppressed viral load in women. Predicted high risk areas are shaded in yellow and gold brown (0.00015 to 0.00020), in 2014, two distinct locations were in the north-east and south-west, while 2015 shows a clustered area in the south-east. Predicted lower risk areas are shaded in royal to dark blue (−0.00015 to −0.00025), the south-west in 2014, with both north and west in 2015. Evidently their exist spatial variation of unsuppressed viral load in this hyperendemic community.

Figure 4.

Estimated posterior mean of the unstructured spatial effect map on the log-odds of unsuppressed viral load among women in uMgungundlovu district, KwaZulu-Natal province, South Africa (2014–2015). (a) 2014, (b) 2015.

4. Discussion

This analysis examined factors associated with unsuppressed viral load among women ages 15–49 years in peri-urban Greater Edendale and rural Vulindlela areas in the uMgungundlovu district, KwaZulu-Natal, South Africa between 2014 to 2015 while accounting for possible nonlinear effect of some continuous variables and mapping the unstructured spatial effects. We fitted hierarchical Bayesian Geoadditive multivariate model while controlling for the confounding effects of the explanatory variables. Bayesian spatial approach have numerous advantages over frequentist statistics, such as ability to account for and measure uncertainty in a model, minimise bias in complex data, ability to produce smoothed risk map, increased prediction accuracy, just to name a few [39, 43, 44]. Due to the strength of this approach many studies have emanated in investigating risk factors of anaemia in Sub Saharan Africa [33, 45], of HIV variation in Kenya [32], viral suppression [46] and other infectious disease globally [47]. Application of Bayesian spatial modelling therefore helped in identifying predictors and high-risk location of unsuppressed viral load among women in a small enumeration area. This enhancement of strategically identifying areas of key population is highly recommended as part of the global AIDS strategy to end inequality in resources allocation and provide localised HIV intervention in hyperendemic communities.

Our study found that knowledge of HIV status, perceived risk of contracting HIV, not on ART, and ART dosage were consistent significant factors associated with higher odds of being virally unsuppressed across both years. Having a CD4 cell count of >350 cells per μL was more likely to be associated with viral load <400 copies per mL Additionally, alcohol consumption was significant in 2014 while meal cut, total number of lifetime sex partner, ever tested for TB and ever diagnosed with STI were factors associated with unsuppressed viral load in 2015. These revealed the heterogeneity and need for continuous surveillance of HIV and its measures, as the predictors of this outcome are dynamic. Although fewer studies on women have been conducted in the country and other developing countries. Similar findings on the association of higher number of sexual partners were also reported among women in uMkhanyakude district of north KZN [48]. The use of ART and dosage of ARV was also found to be significant which is similar to past studies [49, 50]. Furthermore, similar studies have found higher CD4 cell count of >350 cells per μL to be predictive of being virally unsuppressed [51, 52]. We also found similar results on alcohol [53] from Western cape, South Africa and history of TB or STI in Kenya and Uganda [49, 54]. Alcohol use has been found to be associated with non-adherence to treatment in people living with HIV [55] and prevalence of which leads to high risk of transmission [56]. Several studies have shown relationship between TB and virological non-suppression [51, 57]. Similarly, concurrent ART and TB/STI treatment has been shown to increase the risk of virological non-suppression due to impaired treatment adherence and pharmacokinetics drug interaction [49].

The nonlinear effects of age, household size, total number of children ever born and total lifetime HIV test were considered. Across both years, risk of being virally unsuppressed decreases as age increases, with younger age 15–20 having a higher risk and being older associated with reduced risk of unsuppressed viral load. This is similar to past studies [48, 58]. Also risk of being virally unsuppressed decreases with increasing size of the family. Having 5 or lower number of family member showed a higher risk of being unsuppressed. This revealed that larger family members could bring more support to WLHIV. In contrast unsuppressed viral load decreases as number of children ever born increases in 2014 while the inverse was observed in 2015 (risk increases as number of birth increases). Recent study among pregnant women revealed significant association of currently breastfeeding with increase odd of viral load non-suppression [22, 59].

The unstructured spatial effect and observed prevalence map revealed the existence of localised positive spatial variation of unsuppressed viral load among women of reproductive age in this hyperendemic community. While higher prevalence was observed in the north area from both surveys and southern area in 2015. The predicted risk map revealed that in 2014 north-east and south-west as well as south-west in 2015 have an increase likelihood of being virally unsuppressed. This evidently shows that there are regional/district specific factors contributing to unsuppressed viral load with substantial spatial variation. Spatial variation in HIV and its measures has been reported by previous studies [16].

Among women in this community progress on 95–95-95 target was 65.5%, 74.2%, 82.9% in 2014 and 74.7%, 80.0%, 86.6% in 2015. The largest shortfall was in the first target, which is the entry point to health care system. None of the UNAIDS targets were met among this key population. Although, the country has made significant progress but has not achieved the UNAIDS 95–95-95 and 86% composite viral suppression target [14]. However, significant increase in viral suppression of 7.1% over a one year period was seen in this study, while ages 35 to 49 contributed to this increase, which could be attributed to the country commitment and effort in ART scaleup and intervention [27]. However, judging by the 90–90-90 indicator, our findings showed that the “third 90” target was met among age group 35–39, 40–44, 45–49 (91.8%, 91.7%, 92.9%) in 2015.

The key strengths of our study were the robustness of the study design, high participation rate, available of spatial variables and conducting the survey in a real time setting.

5. Limitation

This was a cross sectional population based study and not a randomised clinical trial with limited ART data available. Therefore, no causal effect could be established between unsuppressed viral load and women characteristics. The results are not generalisable to older individuals or children as study only accessed men and women aged 15–49 years.

6. Conclusion

Spatial effects in the model act as a representative of the unobserved predictors which strengthen the result. Identifying high risk areas could help policy maker, epidemiologist, and public health institutions to develop develop strategies and interventions that are suitable for women in the area, thus increasing the impact of allocated resources as well as effective monitoring to improve the health status of women in the community. Increase in progress of the 95–95-95 targets over time showed that the target is achievable in this community among this key population, with intensive HIV testing service, eradication of stigmatisation, ending inequality and increasing uptake of ART treatment. Knowledge of HIV status is a proxy and entry point to achieving the other indicators, generally women are more likely to test than men and receive optimum health care especially during pregnancy.

The likelihood of being virally unsuppressed was higher among younger age group, highlighting public health implication of sustained risk of HIV transmission. Aside clinical factors, family support cannot be underestimated as part of the factors that could help in achieving undetectable viral load among women of reproductive age. Right perception and knowledge of HIV positive status, being on ART and having a higher CD4 cell count contributed to achieving viral suppression. Thus, these remain multi-factorial and important public health priority to attain viral suppression towards the goal to end the epidemic by 2030.