Menu
Menu
Menu
Menu
Background: Despite recent progress in improving access to maternal health services, the utilization of these services remains inequitable among women in developing countries, and rural women are particularly disadvantaged. This study sought to measure i) disparities in the rates of institutional births between rural and urban women in Ghana, ii) the extent to which existing disparities are due to differences in the distribution of the determinants of institutional delivery between rural and urban women, and iii) the extent to which existing disparities are due to discrimination in resource availability. Methods: Using Demographic and Health Survey data from 2003, 2008, and 2014, this study decomposed inequalities in institutional delivery rates among urban and rural Ghanaian woman using the Oaxaca, the Blinder, and related decompositions for non-linear models. The determinants of the observed inequalities were also analyzed. Results: Institutional delivery rates in urban areas exceeded those of rural areas by 32.4 percentage points due to differences in distribution of the determinants of institutional delivery between the two areas. The main determinants driving the observed disparities were wealth, which contributed to about 16.1% of the gap, followed by education level, and number of antenatal visits. Conclusion: Relative to urban women, rural women have lower rates of institutional deliveries due primarily to lower levels of wealth, which results in financial barriers in accessing maternal health services. Economic empowerment of rural women is crucial in order to close the gap in institutional delivery between urban and rural women.
This study analyzed data from the three most recent rounds of the Ghana Demographic and Health Survey (DHS): a large, nationally representative household survey that collected information on fertility and family planning, infant and child mortality, maternal and child health, nutrition, malaria, HIV/AIDs, and a number of other household characteristics using questionnaires administered by trained interviewers. Additional details about the survey methodology and sampling procedures can be found elsewhere [13, 15]. For the purposes of this study, our sample was restricted to women of reproductive age (15–49 years), and focused on their most recent births during the 5 years preceding each of the 2003, 2008, and 2014 DHS surveys. Thus, the final sample spanned the years 1999–2014 and included 13,802 births; of which 5672 (41.10%) occurred in urban areas and 8130 (58.90%) occurred in rural areas. Institutional delivery was the primary outcome of interest and was assessed using self-reported data on the location of delivery of all births that occurred within 5 years of the dates of the surveys. For analyses, it was considered a binary variable and was classified as one if a woman delivered in a public or private healthcare institution and zero otherwise. Area of residency was the key independent variable and was a binary; classified as urban or rural based on the definition used in the 2010 Population and Housing Census. Under this definition, communities were considered urban if they had a population of 5000 persons or greater and rural if they had a population of fewer than 5000 persons [12]. Additional data were extracted from the DHS surveys and were treated as covariates in all analyses. These included factors that are known to be associated with institutional delivery and/or area of residency, such as wealth quintile (ordinal categorical variable), education level (ordinal categorical variable with categories ‘no education’, ‘primary education’, ‘secondary education’, and ‘tertiary education’), parity (continuous variable representing the number of children born to a woman) and distance from a health facility (binary variable classified as one if the woman considered distance to be a barrier to accessing care and zero otherwise), as well as more general sociodemographic factors such as age (ordinal categorical variable with categories ‘15–19’, ‘20–24’, ‘25–29’, ‘30–34’, ‘35–39’, ‘40–44’, and ‘45–49’), ethnicity (nominal categorical variable with categories ‘Ga’, ‘Akan’, ‘Ewe’, and ‘a tribe from the three northern regions’), and religion (nominal categorical variable with categories ‘Christian’, ‘Muslim’, ‘Traditionalist’, ‘other’, and ‘no religion’) and pregnancy-related factors such as pregnancy complications (dummy variable classified as one if a woman had complications and zero otherwise), antenatal care visits attended (continuous variable), contraceptive use (binary variable classified as one if a woman practiced family planning and zero otherwise), and birth year (categorical variable with categories for each year from 1999 to 2014). A variable denoting the year of the survey (categorical variable with response options ‘2003’, ‘2008’, and ‘2014’) was also included. Given that the Ghanaian government has previously introduced reforms in an attempt to increase maternal healthcare utilization and improve outcomes, we also chose to include a categorical variable denoting the reform period in which the birth occurred for all analyses, so as to control for the effects these may have had on service usage. This variable took on a value of one for births that took place outside of the reform periods, two for the first reform period, which was characterised by the provision of free maternal care in public facilities and was in place from 2003 until 2007, when it was integrated into the already functional National Health Insurance Scheme (NHIS) [16], and three for the second reform period, which was introduced in 2008 and provides free NHIS enrolment for all pregnant women, who are enrolled automatically at their first antenatal care visit for a period ending 3 months after their delivery. Under this scheme, all maternity care is covered free of charge in all healthcare facilities, including antenatal care, delivery (vaginal or caesarean), and emergency care [16]. Statistical analyses were conducted using STATA. Descriptive statistics including means for continuous variables and percentages for categorical variables were calculated to describe the characteristics of the study sample by area of residency. The Oaxaca, Blinder, Reimers, and Cotton decomposition methods for non-linear models were used to explain the gap in rates of institutional delivery between urban and rural women. According to the Oaxaca decomposition theory, differences in the mean of an outcome for two groups can be explained by differences in the level or distribution of the determinants of the outcome (explained component), differences in the impact of these determinants on the outcome (unexplained component), and/or the interaction of the two [21]. Assume a regression model that links Y, the outcome variable, to a set of covariates, X with a vector of coefficients, β. Yj = βjXj where j = rural, urban. The difference between Y¯ urban and Y¯ rural (where Y¯ represents average) can be written in two ways: Where ∆X¯ is the difference between X¯ urban and X¯ rural and the similarly for ∆β. Equations (1) and (2) are equivalent and describe the decomposition of the difference between the outcomes of the groups. The three terms on the right-hand side represent the three components of the difference between the outcomes. The first two components (the explained and unexplained components) are the average differences between the Xs of the urban and rural women and that of the βs, with each multiplied by weights. Typically, the first component is weighted by coefficients while the second component is weighted by covariates. The Oaxaca [20] decomposition (1) uses the high group (urban women in this study) as the reference group, weighting differences in characteristics by the coefficients of urban women and differences in coefficients by the covariates of rural women. The Blinder (1973) decomposition (2) does the opposite, using the low group as the reference group (rural women in this study), and weighting differences in characteristics by the coefficients of the rural women and differences in coefficients by the covariates of the urban women. Therefore, the Oaxaca decomposition assumes that the outcome of the high group is in accordance with their characteristics, and that of the low group is due to discrimination against them while the Blinder decomposition assumes that the outcome of the low group is in accordance with their characteristics and that of the high group is the result of societal favoritism. Thus, while the unexplained component of the Oaxaca decomposition focuses on discrimination against the low group, the Blinder’s focuses on favoritism of the high group [21]. Other studies propose using the weighted averages of the two groups as weights. Hence, the weight of the differences in covariates is equal to the weighted mean of the coefficients of the urban and rural groups. According to Reimers [24], the weighted mean should be computed as 0.5 (equal weights for the two groups), while Cotton [6] believes it should be the proportions of the two groups in the sample. Because the outcome of the decomposition is sensitive to the weighting method used [17], this study ran a different decomposition for each method: Oaxaca, Blinder, Reimers, and Cotton. Regressions were performed for urban and rural women separately and then the estimated coefficients and covariates were used to compute the decompositions. Consistent results using the different weights were thought to represent robustness of the study outcome. The equation for the decomposition analysis conducted in this study is specified as follows: j = U, R U = urban, and R = rural. Where y represents the outcome variable, institutional delivery, X2 represents age, X3 represents education, X4 represents household wealth quintile, X5 represents parity, X6 represents pregnancy complications, X7 represents distance from a healthcare facility as a perceived barrier, X8 represents the maternal care reform period, X9 represents the number of antenatal visits the mother attended, X10 represents ethnicity, X11 represents religion, and X12 represents survey year. Four decompositions were estimated: one pooling the data from the three survey years used in the study, and one for each individual survey year. The Heckman Selection Model approach was used to account for possible selection bias wherein only women who had already given birth could be selected for inclusion. The Heckman approach is a two-stage procedure involving the estimation of the selection equation using the larger sample by probit and then computing the inverse mills ratio variable using the predicted outcome in the first stage, and the estimation of the equation of interest with the inclusion of the inverse Mills ratio variable in the second stage. In this study, the dependent variable of the selection equation was a previous delivery and the independent variables were individual-level characteristics including the mother’s age, education level, wealth status, religion, ethnicity, marital status, location of residency, and use of contraceptives. The Mills ratio computed was then added to the response eq. (3) for estimating the decompositions. Overall decompositions were computed to determine the contribution of the explained component, along with that of the unexplained component plus the contribution of the interaction to the gap in the outcome. Additionally, detailed decompositions were computed to determine the contribution of each independent variable to the explained and unexplained decompositions, as described by Kaiser [17]. Following the methods of Kaiser [17], the detailed decomposition analyses focused only on the explained decomposition.
N/A
