Measuring socioeconomic inequalities in prenatal HIV test service uptake for prevention of mother to child transmission of HIV in East Africa: A decomposition analysis

Background Despite efforts made towards the elimination of mother-to-child HIV transmission, socioeconomic inequality in prenatal HIV test uptake in East Africa is not well understood. Therefore, this study aimed at measuring socioeconomic inequalities in prenatal HIV test uptake and explaining its main determinants in East Africa Method We analysed a total weighted sample of 45,476 women aged 15-49 years who birthed in the two years preceding the survey. The study used the most recent DHS data from ten East African countries (Burundi, Comoros, Ethiopia, Kenya, Malawi, Mozambique, Rwanda, Uganda, Zambia, and Zimbabwe). The socioeconomic inequality in prenatal HIV test uptake was measured by the concentration index and illustrated by the concentration curve. Then, regression based Erreygers decomposition method was applied to quantify the contribution of socioeconomic factors to inequalities of prenatal HIV test uptake in East Africa. Results The concentration index for prenatal HIV test uptake indicates that utilization of this service was concentrated in higher socio-economic groups with it being 15.94% higher among these groups in entire East Africa (p <0.001), 40.33% higher in Ethiopia (p <0.001) which was the highest and only 1.87% higher in Rwanda (p <0.01) which was the lowest. The decomposition analysis revealed that household wealth index (38.99%) followed by maternal education (13.69%), place of residence (11.78%), partner education (8.24%), watching television (7.32%), listening to the radio (7.11%) and reading newsletters (2.90%) made the largest contribution to socioeconomic inequality in prenatal HIV test in East Africa. Conclusion In this study, pro-rich inequality in the utilization of prenatal HIV tests was evident. The decomposition analysis findings suggest that policymakers should focus on improving household wealth, educational attainment, and awareness of mother-to-child transmission of HIV (MTCT) through various media outlets targeting disadvantaged sub-groups. Data were extracted from the most recent Demographic and Health Surveys (DHS), consisting ten East Africa countries with complete data for the period from 2011 to 2018 [40, 41]. Countries were Burundi (DHS,2016–17), Comoros (DHS,2012), Ethiopia (DHS,2016), Kenya (DHS,2014), Malawi (DHS,2015–16), Mozambique (DHS,2011), Rwanda (DHS,2014–15), Uganda (DHS,2016), Zambia (DHS,2018), and Zimbabwe (DHS,2015). The DHS are based on nationally representative samples that provide data for a wide range of monitoring and impact evaluation indicators in the areas of population, health, and nutrition [42] and therefore, its an important source of data on health of families in developing countries. The DHS employed a cross-sectional study design with a stratified two-stage sampling strategy. First, each country was divided into enumeration areas (clusters) based on the census frames in the country, and then, households were randomly selected within each cluster. Furthermore, since the DHS surveys were intended to address household-based health issues, strata for urban and rural households were used for the selection of respondents. The DHS follows a standard procedure of data collection and presentation (similar questionnaires) and uses the same definition of terms in each country. The DHS data were collected by the country-specific department of health and population, in collaboration with Inner City Fund (ICF) International using standardized household questionnaires. The detailed methodology of the survey design, sample selection, survey tools, and data collection are described elsewhere [43, 44]. Pooled DHS datasets from ten East African countries were structured by creating a country-specific cluster and country-specific strata. Hence, in this study, we used a total weighted sample of 45,476 women aged 15–49 years who birthed in the two years preceding the survey. The information on the HIV testing during antenatal care and birth in the two years preceding the survey for each woman in the sample was found in the women’s individual records of DHS to measure socioeconomic inequalities in prenatal HIV test service uptake for PMTCT This study used data from the Measure DHS/ ICF International which is secondary in nature. The data were anonymous and are available to apply for online. Therefore, Approval was received from DHS/ ICF International to use. The outcome variable in this study included socioeconomic inequality in prenatal HIV test service uptake. Prenatal HIV testing was defined as the proportion of women who tested for HIV and received the test result during antenatal care or before birth [45]. The variable was coded as “1” if a woman was tested for HIV and also received the HIV test result otherwise coded as “0” if the woman did not test for HIV or tested but did not receive the test result during antenatal care or before birth. Based on previously published literature from low and middle-income countries [18, 19, 24], the following socioeconomic and demographic factors were selected for this study and examined: country of participants, place of residence (urban and rural), maternal age (categorized as 15–24 years, 25–34 years, and 35–49 years), and maternal and partner educational level (categorized as no education, primary education, secondary and higher education). The sex of the household head was coded as ‘male’ if the participants lived in the male-dominated household, or ‘female’ if otherwise. The household size was classified as 1–3, 4–6, or ≥7 members and employment status were categorized as not working, formal employment, and non-formal employment. Women’s access to mass media was categorized as yes or no in relation to access to the radio, watching television, and reading magazines/newspapers. The calculation of concentration indices (CI) requires a single indicator to capture respondents’ SES. Hence, the DHS household wealth index was used as a measure of the SES of mothers in East Africa [46]. The household wealth index stands out to be the most appropriate measure of SES for national surveys in comparison with direct measures of living standards such as income, consumption, or expenditure [47]. The wealth index was constructed via principal components analysis (PCA) [48] using data such as household assets (e.g. radio, televisions, refrigerators, farmland, farm animals) [48, 49], housing characteristics (e.g. type of water access, type of flooring), and access to basic services (e.g. electricity supply, source of drinking water and sanitation facilities) [46]. PCA is a multivariate statistical method that is widely used as a data reduction technique [48]. The detailed methodology of the wealth index construction is described elsewhere [46, 49]. In this study, wealth index score was divided into quintiles, each category comprising 20% of the population. The lowest 20% quintile was assigned to the poorest households, the next 20% quintile to generally poor households, followed by another 20% quintile for the middle-class households, and finally the top 40% quintile for the wealthier and wealthiest households. The analytical approach for this study was undertaken in three stages. Firstly, sample characteristics and prevalence of prenatal HIV test service uptake were described using frequencies and percentages. Secondly, concentration indexes (CIs) were determined to examine the extent of socioeconomic inequalities in prenatal HIV test service uptake in East Africa. Thirdly, we decomposed the concentration indices to understand the contribution of various factors to inequality. The methods are discussed in detail as follows. Concentration index (CI) has become a popular tool to measure health and health care inequality in the field of health policy and health economics research [50]. It assumes values between -1 and +1 [38]. When the concentration index is positive, it reflects higher uptake of HIV test services tend to the rich. If it is negative, it reflects higher uptake of HIV test services tend to the poor. In the absence of any inequalities, the value of the concentration index is zero [38]. It can be estimated as: Where CI is the concentration index for health service use (prenatal HIV test service uptake in this case), yi is health service use for individual i, μ is the mean of health service use, and Ri is the fractional rank or asset score of the ith individual in the living standards distribution/socio-economic rank with i  =  1 for the poorest and i  =  n for the wealthiest. The concentration index depends only on the relationship between the health variable (yi) and the rank of the living standard variable (Ri) and not on the variation of the living standard variable itself. When the health variable of interest is dichotomous, the concentration index is not bounded within the range of (–1,1) [51]. The lower bound is then equal to μ−1+(1n), the upper bound is equal to 1−μ+(1n) [38]. For large samples, the (1n) terms vanish, and the maximum and the minimum values tend to μ−1 and 1−μ respectively [51]. To consider the bounded nature of prenatal HIV test service uptake, Erreygers (2009) proposed a modified version of the concentration index called Erreygers normalized concentration index (ECI) [52]. This is defined as: where ECI is Erreygers concentration index, CI is the generalized concentration index and μ is the mean of prenatal HIV test service uptake. The Erreygers normalized concentration index (ECI) method was also applied to decompose prenatal HIV test service utilization inequality in this study [39, 52]. Furthermore, with concentration curve, the cumulative proportion of women ranked by SES (on the x-axis) against the cumulative proportion of prenatal HIV test service uptake (on the y-axis) were plotted. If the concentration curve lays below the 45-degree line, it suggests prenatal HIV test service use is concentrated more amongst women from rich households and vice versa. Decomposition of the healthcare inequality relies on the assumption that healthcare is a linear function of the outcome variables [38]. However, since the outcome variable of this study is binary, an appropriate statistical technique for non-linear settings is needed. According to O’Donnell, et al. [38], a linear approximation when dealing with a discrete change from 0 to 1 is to use marginal or partial effects (dh/dx) estimates as follows: Where βkm is the marginal effects (dy/dx) of each xk; e indicates the error term generated by the linear approximation. In this study, we performed the decomposition analysis using a generalized linear model (GLM) with a binomial distribution and a logit link function as a linear approximation to capture the partial effects of socioeconomic factors on prenatal HIV test service uptake [53]. GLM is a suitable regression to provide consistent results for the decomposition of binary outcomes regardless of the choice of the reference category [53]. Therefore, given the relationship between y and xk in Eq [3], we decomposed the concentration index of prenatal HIV test service uptake y(CI) into its contributory factors as follows In this expression, x¯k is the means of explanatory variables, βkm is the patrial effect on explanatory variable xk (dy/dxk), CIk is the concentration index for determinant xk, and GCIe is the generalized concentration index for the error term. In summary, the contribution of the determinants is calculated in four steps. Firstly, the regression model of the health outcome variable is performed for all xk to obtain the marginal effects of determinants (βkm), which demonstrate the associations between the determinants and prenatal HIV test service uptake (positive signs indicate positive associations whilst negative signs indicate negative associations). Secondly, the elasticity of the health variables was calculated for each x (xk), which is the sensitivity of prenatal HIV test service uptake to changes in the determinants (βkmx¯kμ). It denotes the change in the dependent variable (socioeconomic inequality in prenatal HIV test service uptake in this case) associated with a one-unit change in the explanatory variables. Thirdly, the CIs are calculated for the prenatal HIV test uptake and each explanatory variable (CIk). Fourthly, the contribution of each explanatory variable to the overall CI is calculated by multiplying the elasticity of each determinant by its concentration index ((βkmx¯kμ)CIk). In this study, we accounted for the multistage survey design during descriptive, regression, and decomposition analyses by creating country-specific clustering, country-specific strata, and population-level weight. All statistical analyses were conducted using STATA version 14.2 (Stata Corp, College Station, TX, USA).