Data
The study employed secondary data from three rounds of the Ghana Demographic and Health Survey (GDHS) conducted between 2003 and 2014. In Ghana, the survey was conducted by the Ghana Statistical Service (GSS) in collaboration with the National Public Health and Reference Laboratory (NPHRL) and the Ghana Health Service (GHS).
The survey collects comprehensive information on health care utilization. It also collects information on household asset and ownership. This serves as an important source to assess economic status of the household. We used the wealth index as the measure of socioeconomic status to rank women from the lowest to highest in the inequality analysis. In this study, we used three rounds of the GDHS survey (2003, 2008 and 2014). The Ghana Demographic and Health Survey (GDHS) followed a two-stage sample design. The first stage involved selecting sample points (clusters) consisting of enumeration areas (EAs). The second stage involved the systematic sampling of households. The households included in the survey were randomly selected from each cluster to constitute the total sample size of households. In deriving the data, focus was on women aged 15 to 49 who were permanent residents of the household or visitors who had stayed in the household being interviewed the night preceding the Survey (GSS, 2015). A total of 5691 eligible women participated in the 2003 survey whilst 4916 women as well as 9396 women were interviewed in the 2008 and 2014 survey respectively. The survey year 2003 represents the period before the introduction of the NHIS in 2004 while 2008 and 2014 represent two periods after the implementation of the scheme.
Analytical approach
The analytical approach for this study was in three stages. The first stage used concentration curves (CCs) to examine the trend and pattern of socioeconomic inequalities in health care utilization (measured by antenatal care and delivery by trained attendants). In the second stage, concentration indices (CIs) were computed for each outcome variable across the years. The final stage decomposed the concentration indices to understand the contribution of various factors to inequality. The methods are discussed in detail as follows.
Concentration curves and concentration indices
To examine the trend in inequalities in health care utilization, we constructed CC for each of the health care utilization measures. The CC gives a graphical view of the pattern and extent of inequalities in ANC utilization and DTA. A CC is a plot of the cumulative percentage of the outcome variable on the y-axis against the cumulative percentage of the population ranked by household socioeconomic status (starting from the poorest) on the x-axis [15]. The 450 line or the diagonal in the CC graph represents equality in healthcare utilization. If the CC lies above the diagonal, outcome variable is concentrated among poorer people. When it is concentrated among richer people, the CC lies below the line of equality. There is no inequality when the CC lies on the 45° line. The extent of inequality is shown by how far the CC lies away from the line of equality (45° line). The further the CC is from the line of equality, the greater the extent of inequality [15]. The CIs were also estimated to determine the degree and nature of inequalities in ANC and DTA.
The CI is defined as “two times the area between the concentration curve and the line of equality” ([15], p. 95). The CI was calculated using the following formulae.
$$ \mathrm{CI}=\frac{2}{\mu }c\; ov\;\left({y}_i{r}_i\right) $$
(1)
Where y is a set of health utilization variables, ri is fractional rank of individual in the wealth score distribution, cov is covariance and μ represents the mean of the healthcare variable. The CI can either be positive or negative. The sign of the CI explains the relationship that exists between the healthcare variable and position in the wealth score distribution. If the CI is zero, it means that there is no inequality in the distribution of healthcare use by wealth and hence the CC will coincide with the line of equality. A negative value of the CI is obtained if the healthcare variable is disproportionately concentrated among the poorest whilst a positive value of CI suggest inequality concentrated among the richest. The value of the CI ranges between − 1 and + 1 (i.e., − 1 ≤ CI ≤1) and the CI gives information about the strength of the relationship and the extent of variability in the dependent variables. The closer the absolute value of the CI to one, the greater the level of inequality.
Decomposition analysis
The decomposition of the CI was performed to estimate the individual contribution of explanatory variables to inequalities in the outcome variables. The contribution of every individual characteristic is defined as the product of how sensitive that characteristic is to health and the extent of inequality in that factor [21].
Decomposition of the healthcare inequality relies on the assumption that the healthcare is a linear function of the outcome variables. This is important because in decomposition analysis the concentration indices are calculated using the predictions from a linear regression model [21].
The starting point was to express a linear function of the outcome variables in relation to the NHIS variable as well as other demographic and socioeconomic control variables. This is given as:
$$ y=\mathrm{a}+{\sum}_k{\beta}_k{x}_k+\upvarepsilon $$
(2)
Where x represents the vector of explanatory variables, including NHIS status. Following Wagstaff, Doorslaer, & Watanabe [21], the standard concentration index (CI) for outcome variable y can be written as
$$ \mathrm{CI}\left(\mathrm{y}\right)={\sum}_k\left({\beta}_k{\overline{x}}_k/\upmu \right)\;{c}_k+\mathrm{G}{C}_{\varepsilon }/\upmu $$
(3)
Where CI(y) is the standard concentration index, \( {\overline{x}}_k \) is the mean of xk, ck is the CI for xk, μ is the mean of y, G Cε is the generalized CI for the error term (ε). From equation (3), two important grouping can be made; (i) the first term on the right-hand side of the equation expresses a weighted sum of the CI of k regressions, where the weight \( {\overline{x}}_k \) is the elasticity of y with respect to xk (ηk = \( {\beta}_k{\overline{x}}_k \) / μ). (ii) the second term on the right-hand side is the residual element which expresses the portion of inequality that cannot be explained by the contributing variables. Statistical significance of the CIs as well as the decomposition analysis was calculated using the bootstrapping technique with robust standard errors [7].
Description of variables
In this study, we focused on pregnancy-related maternal health care utilization indicators. Specifically, we used, at least four ANC visits, and delivery by skilled attendants. Outcomes were measured as dummy variables that take the value of one if a woman had utilized the service and 0 otherwise.
In terms of socioeconomic indicators, the GDHS collects household asset information that is used to compute a wealth index. This wealth index has been shown to be strongly correlated with the economic and social status of the household [18]. Other variables included community, household and individual characteristics. The GDHS collects demographic and socioeconomic variables such as education, place of residence, age, gender, health facility in community, health insurance coverage, region of residence, sanitation, family size, among others.