Wealth-related inequality in early uptake of HIV testing among pregnant women: An analysis of data from a national cross-sectional survey, South Africa

Objectives Wealth-related inequality across the South African antenatal HIV care cascade has not been considered in detail as a potential hindrance to eliminating mother-to-child HIV transmission (EMTCT). We aimed to measure wealth-related inequality in early (before enrolling into antenatal care) uptake of HIV testing and identify the contributing determinants. Design Cross-sectional survey. Settings South African primary public health facilities in 2012. Participants A national-level sample of 8618 pregnant women. Outcome measures Wealth-related inequality in early uptake of HIV testing was measured using the Erreygers concentration index (CI) further adjusted for inequality introduced by predicted healthcare need (ie, need-standardised). Determinants contributing to the observed inequality were identified using the Erreygers and Wagstaff decomposition methods. Results Participants were aged 13 to 49 years. Antenatal HIV prevalence was 33.2%, of which 43.7% came from the lowest 40% wealth group. A pro-poor wealth-related inequality in early HIV testing was observed. The need-standardised concentration index was â ‘0.030 (95% confidence interval â ‘0.038 to â ‘0.022). The proportion of early HIV testing was significantly better in the lower 40% wealth group compared with the higher 40% wealth group (p value=0.040). The largest contributions to the observed inequality were from underlying inequalities in province (contribution, 65.27%), age (â ‘44.38%), wealth group (24.73%) and transport means (21.61%). Conclusions Our results on better early uptake of HIV testing among the poorer subpopulation compared with the richer highlights inequity in uptake of HIV testing in South Africa. This socioeconomic difference could contribute to fast-tracking EMTCT given the high HIV prevalence among the lower wealth group. The high contribution of provinces and age to inequality highlights the need to shift from reliance on national-level estimates alone but identify subregional-specific and age-specific bottlenecks. Future interventions need to be context specific and tailored for specific subpopulations and subregional settings. A secondary analysis of data from a national cross-sectional survey conducted in 2012 to evaluate the South African PMTCT programme was conducted.20 The methods have been explained in detail elsewhere.21 In summary, the survey was conducted at public primary healthcare clinics and community health centres offering immunisation services countrywide. The primary aim was to measure national and provincial-levels MTCT among infants attending public health facilities for their 6-week immunisation. Infants with known and unknown HIV exposure were eligible for inclusion. The 6-week postpartum point was chosen because it has a 99% infant coverage for immunisation.22 Antenatal HIV prevalence and presumed PMTCT coverage were used to estimate the sample size needed for each province at precisions of 30% to 50% and a design effect of 2. The national target sample size was 12 200, ranging between 700 and 1800 per province, proportional to provincial 6-week immunisation coverage. A two-stage probability proportional to size sampling approach was used. The first stage was at provincial level. In each province, health facilities were stratified into medium (130–300 immunisations per year) and large (300 immunisations or more per year) facilities. Large facilities were further stratified into two groups—facilities in districts with antenatal HIV prevalence <29% or≥29%, which was the 2009 national average antenatal HIV prevalence. Therefore, facilities were grouped into three strata. The second stage was at health facility level: 580 facilities selected proportional to target facility sample size were needed to achieve the desired provincial and national sample sizes. The target number of infants per facility was taken as the median number of infants expected in each facility within each stratum over a 3-week data collection period. Finally, caregiver–infant pairs were invited to enrol into the study during the 6-week immunisation visit using either random or consecutive selection depending on facility size. Ultimately, 10 533 infants were screened and 9120 provided both interview and infant blood data to measure MTCT. With respect to the data analysis for the primary outcome (6-week MTCT), sampling weights were calculated as the inverse of the realised sample size, accounting for South African live births, relative to the target sample size for each facility. Consent to enrol into the study, to be interviewed and to take infant blood for laboratory HIV tests was sought from infant caregivers. Ethics approval was granted by the South African Medical Research Council (MRC) Ethics Committee in 2009 (institutional review board identifier—FWA00002753). Information about sociodemographic characteristics and uptake of antenatal and PMTCT programmes was collected through interviews. Two HIV tests were performed on the infants: (1) an ELISA for passively transferred maternal anti-HIV antibodies to confirm maternal HIV infection and infant HIV exposure and (2) an HIV total nucleic acid PCR to confirm infant HIV infection. The ELISA results for infant HIV exposure were used here as a proxy for antenatal HIV prevalence. Data from 8618 out of 9120 consented caregiver–infant pairs were used for analysis; the rest had missing information to establish SES. The main outcome variable was binary: early uptake of HIV testing, that is, self-initiated HIV testing before enrolment to antenatal care versus PMTCT programme-influenced testing after enrolling into antenatal care during pregnancy. Independent variables with potential to influence inequality in the outcome were chosen, that is, variables which can influence or be influenced by socioeconomic background and at the same time can influence at least one of the outcomes: education level, dichotomised as primary school and lower or high school and above was selected as education could influence attitudes towards the importance of healthcare; marital status, dichotomised into single women (ie, not married, not in a relationship, widows, divorced) and married (or cohabiting) women, was included as spousal support is likely to encourage uptake of healthcare; transport to health facility categorised into own car, public transport and walking was included as a marker of ease of healthcare access, affecting the frequency and timing of uptake; prior knowledge about PMTCT as either ‘yes’ or ‘no’ was included as prior knowledge can influence timing of HIV testing in relation to pregnancy; a categorical variable of the nine South African provinces was included as provincial differences in healthcare management and in cultural behavioural norms has been observed; lastly, source of income with four categories of women namely employed, dependent on extended family, dependent on spouse or partner and fourthly those with irregular sources of income such as government grants. The latter is not a good measure of household income but is a common structural division in South Africa, and it will be important to know whether and how it impacts on the primary outcome variables. Three healthcare need-based variables included were  maternal age, a positive syphilis diagnosis result during pregnancy and a positive tuberculosis (TB) diagnosis result during pregnancy. These were used to predict and adjust for inequality due to differences in need for ill-health-related healthcare, therefore allowing for a better prediction of inequality under equal needs. Age is not ill health itself but different age groups have pre-existing differences in risk of ill health which thus introduces inequity in need for healthcare. The wealth scores to measure SES were generated from household living conditions and household assets (ie, house building material, sanitation, water, domestic fuel source and household appliances) using principal component analyses.23 The wealth scores are only based on household assets because information on actual value of household income was not available. However, these assets in the current South African context do give a good indication of wealth status. Wealth-related inequality measures were performed in R Statistical package v3.1.0 and in STATA version SE 2013. Wealth-related inequalities were determined using the concentration index measure which has been described in detail elsewhere.24 25 Briefly, the concentration index is used to measure wealth-related inequality and ranges from −1 to 1. It is calculated from twice the area under a curve (which is a relative measure of the covariation between the health outcome and the SES ranking, formula shown in equation 1), the concentration curve, which deviates from a line of equality (the diagonal straight line). Along this diagonal line, CI=0, meaning that there is no inequality caused by wealth differences, that is, the distribution of the variable of interest across the SES groups is not influenced by wealth. in which h is the health outcome of interest, r the SES ranking and µ the mean of the health outcome. In this study, for example, h would be either ‘early uptake of HIV testing’ or ‘infant HIV exposure’. A positive CI (and curve below the diagonal line) indicates that a variable is favourable among the higher wealth groups (the wealthy), otherwise it is more prevalent among the lower wealth groups (the poor, when the curve is above the diagonal line). Contribution of determinant variables to wealth-related inequality can be calculated using a regression-based decomposition analyses shown in equation 2. where for 1 to k determinant variables, βk is the coefficient of a determinant variable, x¯k is the mean of the determinant, µ is the mean of the health outcome and Ck is the concentration index of the determinant. An error term would also be included in equations 1 and 2 for continuous outcomes.24 In this study, k would represent the six independent variables described earlier. The concentration index formulas were initially designed for continuous variables, therefore are limited in handling the bounded nature of binary variables. Since the outcome variable of this study was binary, we applied the commonly used Erreygers correction26 on the CI (equation 3) to correct for the linearity assumptions in the above equations. where h is the health outcome of interest and µ mean of the health outcome. Therefore in Erreygers correction, the concentration index of the health outcome is multiplied by four times the mean of the outcome, then divided by the range of the outcome. Similarly the wealth-related inequality decomposition by contributing determinant factors was adjusted using the Erreygers method (equation 4).9 The target strata sample sizes for the survey were not all fully attained; hence, all analyses were adjusted using appropriate sampling weights. To accurately measure horizontal wealth-related inequality under equal needs, we used two approaches to adjust for need-based inequality measure. First, we included the healthcare need-defining variables—age, syphilis diagnosis during pregnancy and TB diagnosis during pregnancy in the decomposition analyses together with non-need variables to generate a need-standardised concentration index.24 27 Second, we subtracted the concentration index defined by need variables alone (inequality due to need-predicted uptake) from the standard concentration index.