Towards universal health coverage for reproductive health services in Ethiopia: Two policy recommendations

Reproductive health services are crucial for maternal and child health, but universal health coverage is still not within reach in most societies. Ethiopia’s goal of universal health coverage promises access to all necessary services for everyone while providing protection against financial risk. When moving towards universal health coverage, health plans and policies require contextualized knowledge about baseline indicators and their distributions. To understand more about the factors that explain coverage, we study the relationship between socioeconomic and geographic factors and the use of reproductive health services in Ethiopia, and further explore inequalities in reproductive health coverage. Based on these findings, we discuss the normative implications of these findings for health policy. Using population-level data from the Ethiopian Demographic and Health Survey (2011) in a multivariate logistic model, we find that family planning and use of antenatal care are associated with higher wealth, higher education and being employed. Skilled attendance at birth is associated with higher wealth, higher education, and urban location. There is large variation between Addis Ababa (the capital) and other administrative regions. Concentration indices show substantial inequalities in the use of reproductive health services. Decomposition of the concentration indices indicates that difference in wealth is the most important explanatory factor for inequality in reproductive health coverage, but other factors, such as urban setting and previous health care use, are also associated with inequalities. When aiming for universal health coverage, this study shows that different socioeconomic factors as well as health-sector factors should be addressed. Our study re-confirms the importance of a broader approach to reproductive health, and in particular the importance of inequality in wealth and geography. Poor, non-educated, non-employed women in rural areas are multidimensionally worse off. The needs of these women should be addressed through elimination of out-of-pocket costs and revision of the formula for resource allocation between regions as Ethiopia moves towards universal health coverage.

Survey data have the greatest potential in the analysis of health equity [27]. We used data from the most recent Ethiopian Demographic and Health Survey (EDHS 2011), conducted by the Ethiopian Central Statistical Agency between December 2010 and June 2011 [21]. This household-level survey is a nationally representative sample of 17,817 households selected on the basis of the Population and Housing Census from 2007 (Ethiopian Central Statistical Agency). The sample was selected by a stratified cluster sampling design and consisted of 16,515 women (15–49 years of age) and 14,100 men (15–59 years of age). Data design and collection is fully described in the Ethiopia Demographic and Health Survey 2011 final report [21]. Ethical clearance for the EDHS was provided by the Ethiopian Health and Nutrition Research Institute Review Board, the National Research Ethics Review Committee at the Ethiopian Ministry of Science and Technology, the Institutional Review Board of ICF International, and the U.S. Centers for Disease Control and Prevention. The current study was exempted from full review by the Regional Committee for Medical and Health Research Ethics in West Norway, as the study is based on anonymous data with no identifiable information. As the overall reproductive health coverage is low in Ethiopia [21], we studied individual-level indicators proposed by the WHO to monitor reproductive health [28]. The following indicators for reproductive health coverage have been identified as high-priority interventions in the Ethiopian Health Sector and Development Plan IV [17]: family planning (FP), antenatal care (ANC), and skilled birth attendance (SBA) (see web-Additional file 1). In the analysis we explanatory variables were based upon descriptive data (Table 1) and recommendations from the current literature on factors that have been associated with reproductive health coverage and mortality, and factors that have been recognised as relevant in inequality analysis [26, 29, 30]. We included a range of possible explanatory variables that have been shown to be associated with reproductive health services: socioeconomic variables at the household level, barriers reported at the household level, geography, and use of other health care services. Maternal age and birth order of child were included in the analysis as potential confounding factors [23]. We used the wealth index from the EDHS as a proxy for socioeconomic status. The index was created using principal component analysis, where the index is a continuous variable based on household assets and living standard (for further details, see the DHS website [31]). Based on the wealth index, five wealth quintiles were used in the multivariate analysis, as our primary interest was the difference between poor and less-poor groups. We included additional socioeconomic factors as dummy variables (for further description, see the web-Additional file 1). To further understand the barriers to health-service use [26], we included reported problem(s) of getting medical help for self in the model. Although we cannot assume a causal relationship between the reported problem(s) of “getting medical help for self” and health coverage; studying the reported problems can add information about less understood household level barriers and demand factors (Fig. 1) [26]. We included the following reported problems in our analysis (0 = not a problem, 1 = a significant problem): permission to go, money needed for treatment, distance to health facility, having to take transportation, not wanting to go alone, concern over no female provider, concern over no provider, concern over no drugs being available, and workload inside and outside the home. These factors may explain reproductive health coverage and inequalities in reproductive health coverage. To determine if identified religious beliefs and related traditions were associated with health coverage, we included information related to religious view (Islam, Orthodox Christianity, Protestant Christianity, and other religions). We also included administrative region (nine regions and two cities) as independent variables to determine if they would be associated with coverage. We used Addis Ababa as a reference region, as this is the region that is closest to reaching full coverage of services (Table 1). Previous use of antenatal care and skilled attendance at birth were included in the models, as the literature indicates that previous health-services utilisation is a predictor for successive use of health services (see web-Additional file 1) [23]. The analysis was conducted using STATA statistical software (STATA 13.1). To explore possible associations between explanatory variables and binary outcomes, other factors held equal, we performed multivariate logistic regression [32]. The data material is from a household survey, and standard sample weights (provided in the DHS data set) were used to correct for potential over-and under-sampling. Further, we adjusted for the clusters (the primary sampling units). The analysis was based on women in their reproductive age (15–49 years); 11,654 women, and their 7764 last pregnancies. As previous health care use and use of antenatal care was included in the model, the analysis was limited to 7422; 7708; and 7702 women in the final regression analysis of family planning, antenatal care and skilled attendance at birth, respectively. Modifying the outcome of the logit model, we present the exponential coefficients as adjusted odds ratios (OR) to give the reader an approximation of how a 1-unit change in the explanatory variables will affect the dependent variable(s); If the OR is higher than one, exposure associated with higher odds of the outcome. If the OR is lower than one, exposure is associated with lower odds of the outcome. Based on the current literature and Table 1, we hypothesised that higher education, higher wealth, urban residence, being employed, and having health insurance would be associated with higher use of reproductive health services [19–21, 26, 29, 33, 34]. We further hypothesised that female headed household and problems reported with getting medical help for self would be factors associated with a lower chance of using reproductive services. It is difficult to predict how religion and geography affect outcomes, but the descriptive data indicate that they may have an impact (Table 1). The concentration index has been used to quantify health and health service coverage inequalities when seeking to understand how coverage indicators of interest vary across income or wealth distributions [27]. Recent discussions illustrate that none of the inequality measures available are perfect [35]. We chose the Erreygers corrected concentration index (CCI), as it corrects for several problems in the standard concentration index as noted in the literature [7, 35]. For the reproductive health coverage variables of interest (y), the Erreygers CCI can be calculated as: where yi is reproductive health coverage (dependent variable) of the individual i and Ri is her fractional rank in the wealth distribution, with i = 1 for the poorest individual and i = N for the least-poor individual in the sample. A positive CCI will indicate that the better off have disproportionately higher service coverage, and the opposite is true for a negative CCI. We hypothesise that the CCI will be positive when looking at FP, ANC, and SBA, as the literature has described that the better off make more use of services [1, 7, 36–38]). To further explore which factors contribute to inequalities, the concentration index can be decomposed by relating health outcomes to their potential socioeconomic determinants [27, 35, 39]. Hereby, we can investigate underlying inequalities that may explain the variation in health coverage. The concentration index can be decomposed to the contributions of the individual factors to wealth-related health inequality, where each factor’s contribution is the product of its sensitivity and the degree of wealth-related inequality of the given factors [27, 35, 39]. The concentration index of a given dependent variable of interest, y, can be written as where x−k is the mean of xk (reproductive health coverage), CIk is the CI of xk, and GCϵ is the generalised CI of the error term (ε). CCI is then equal to a weighted sum of the CIs of the k regressors. The residual expresses the inequality that cannot be explained due to systematic variation in the regressors included in the analysis. The closer the residual goes towards 0, the better the fit of the model. We use the wealth index as a continuous variable creating the fractional rank, but look at the contribution of the different wealth quintiles in the decomposition analysis. The decomposition of the dependent variable is based on a linear regression model. Though logistic regression was used in the multivariate analysis, Gravelle et al. have shown that the decomposition analysis can also be extended for binary outcomes [40]. Only explanatory factors that showed P < 0.05 significance in the multivariate regression analysis were included in the decomposition analysis.

• onarheim, kristine husøy
• tolla, mieraf taddesse
• norheim, ole frithjof
• abdullah, muna
• miljeteig, ingrid