Abbreviations

AIC: Akaike information criterion

ANC: Antenatal Care

AOR: Adjusted Odd Ratio

BIC: Bayesian information criterion

CI: Confidence Interval

COR: Crude Odd Ratio

CSA: Central Statistics Agency

EA: Enumeration Area

EDHS: Ethiopian Demographic and Health Survey

MCH: Maternal and Child Health Service

PCA: Principal Component Analysis

PNC: Post-Natal Care

SD: Standard deviation

VIF: Variance inflation Factor

WHO: World Health Organization

Background

Maternal health refers to the health of women during pregnancy, childbirth, and the postpartum period. Maternal health is an im¬portant indicator of wellbeing of women, and has significant impli¬cations for the survival of the neonates and children [1]. The most commonly known maternal health interventions includes antenatal care service (ANC), delivery at health facilities, and postnatal care services (PNC) [1]. Antenatal care (ANC) is a service given to pregnant women by a skilled health care provider. This service is given during pregnancy to ensure better health outcomes to both the mother and the newborn [2].

Currently, the World Health Organization (WHO) recommends, under normal circumstances, a pregnant woman is expected to get ANC services at least eight times during her pregnancy. The first visit is recommended during the 12 weeks of pregnancy [3]. In this regard, insufficient ANC visits or delays in visits or reduced number of ANC visits may lead to poor pregnancy outcomes [4].

Another key maternal and child health intervention is health facili-ty delivery. Giving birth in a healthcare facility with the assistance of skilled birth attendants can significantly reduce the rates of maternal morbidity and mortality [5]. Using health facility during birth is also playing a crucial role in prevention of stillbirths and enhance the survival of a new-born child [6]. Postnatal care (PNC) must be provided in the first two months after delivery and usually considered as a golden opportunity for health service providers to train mothers on healthy breastfeeding practices, to screen mothers for a postpartum depression, monitor the infant growth and overall health status, to observe and treat childbirth-related complications, advice those women about the existing family planning options and refer/ link the mother and baby for specific care if needed, among other services [7,8]. According to a WHO recommended guideline, women and the newborn should get at least three post¬natal contacts with the health service provider after childbirth. The first visit should be made between 48 and 72 hours, the second visit between day 7 and 14th and the third at six weeks postpartum [7].

Globally, an estimated 800 women die every day from compli-cations related to pregnancy, childbirth and postpartum period in 2020 [9]. In the same year, nearly 70% maternal deaths occurred in sub- Saharan Africa [9]. In Ethiopia, Maternal Mortality Ratio (MMR) was estimated as 420 per 100,000 live births in 2016 [10]. Globally 87% of pregnant women get the antenatal care services given by skilled health personnel at least once during their preg¬nancy period [11].

The lowest levels of antenatal care service provision rates are observed in Sub-Saharan Africa and South Asia. Dickson et al in 2022 done a pooled analysis of the ANC utilization from a data collected from year 2000 to 2018 from 32 countries from Sub-Sa-haran Africa using DHS surveys. The result revealed that 76% of pregnant women had at least one ANC check-up. Among coun¬tries in this region, Gambia (99%) and Burundi (8%) had relatively higher and lower prevalence of ANC respectively [12]. Evidence from the 2019 EMDHS indicated that, 43% of pregnant women in Ethiopia had at least four ANC visits in 2019, 59% and 37% for urban and rural areas, respectively [13].

In 2020, birth assisted by a skilled attendant in the world reached 83% [14,15]. Similarly, in Ethiopia, birth assisted by a skilled attendant reached 50%, increasing from 26% in 2016 to 48% in 2019, with prevalence of 70% for urban and 40% for rural areas [13]. Only 34% of women in Ethiopia received a postnatal check within two days of delivery, and with prevalence of 48% for urban and 29% for rural areas 2019 [13]. Previous studies conducted in Ethiopia focused on the determinants of maternal health care ser¬vice utilization (i.e., which includes ANC, health facility delivery, and PNC).

For instance, some of the most commonly cited predictors of ANC included: availability, educational level, history of antenatal care use, wealth index, husband’s attitude, the home visits by health ex¬tension workers, level of understanding the health complications related to pregnancy, health related training given to household and place of residence [16-21]. Most frequently reported predic¬tors of delivery at facility were wealth/income, women and their husbands’ education level, place of residence, and the utilization of ANC [14-16,18,20]. Moreover, the predictors of PNC service utili¬zation included; wealth, place of residence, model family, prenatal stay at maternity waiting homes, antenatal care in the first trimes-ter, complete antenatal care service, family conversation during pregnancy, cesarean delivery (C-section), the notification of birth to health extension workers, maternal health education, informa¬tion on postnatal care service utilization, and women autonomy [16,18,20,22].

This being the case, very little attention has been given to the study of MCH in the four study regions concerned, even though service utilization in these dominantly pastoral regions is unacceptably low [10]. Previous studies on the subject have also typically em¬ployed individual based cross-sectional data or a single indicator (such as ANC or PNC or delivery care) to estimate prevalence and determinants [16-19,22-26]. In addition, none of them quantified the contributions of observed and unobserved heterogeneity at the individual, household and community level through decomposition techniques. To the best of our knowledge, very few studies were conducted in Ethiopia that examined the cluster level inequality in the three key maternal health care service utilizations. Therefore, the aim of this study is to assess the urban-rural inequality of ma¬ternal health care utilization in four regions of Ethiopia.

Methods

Study Setting

The study is done in four more or less geographically marginalized regions of Ethiopia, namely, Afar, Benishangul Gumuz, Gambela and Somali regions. They all have similar physical characteristic i.e.; all the four regions have a border line with the neighboring countries with Ethiopia. In terms of economic status, these regions are deemed disadvantageous for access to basic health services which includes maternal and child health (MCH) i.e., prevalence of most MCH service utilization is much lower compare to the re¬maining regions of Ethiopia [10]. According to the 2020 projected population size by Central Statistical Agency (CSA) of Ethiopia,the total women population size in Afar region was 885 thousand (46% of the total population of the region), Somali region had 2,893 thousand (48% of the total population of the region), Ben-ishangul-Gumuz had 570 thousand (49% of the total population), and Gambella had 229 thousand (48% of the total regional popu¬lation) [27].

Data Source

This study used the 2016 Ethiopian Demographic and Health Sur¬vey (EDHS) data sets which are freely available online. The 2016 EDHS sample was designed to provide estimates for each indica¬tor for the entire country. In addition, sample size for urban and ru¬ral areas was calculated separately. The survey was carried out by the Ethiopian Central Statistics Agency (CSA) and ICF. The data collection was done in the period from March 2016 to July 2016. Therefore, we extracted the required data for this particular study for the four regions of Ethiopia (i.e., Afar, Benishangul Gumzu, Gambela and Somali regions) from the existing data set [10].

Study Design

In this study, combinations of cross-sectional and partial ecologi-cal designs were used. It is cross-sectional because the 2016 EDHS collected the date at a specific point in time in the lives of respon-dents. This makes it difficult to assess the cause-effect relationship between the outcomes and explanatory variables of interest. On the other hand, the analysis followed a partial ecological design. A partial ecological study design involves a mixture of individual level and ecological variables (group level). In ecological analysis, the unit of observation is a group, not separate individuals for one or more study variables. To make ecologic inferences about effects on group rates the study subjects were grouped by place (cluster) [28].

Sampling

The EDHS 2016 survey used a two-stage stratified sample. In the first stage a total of 645 Enumeration Areas (EAs) were selected with a probability proportional to the size of the Enumeration Area (EA) and in the second stage sampling a fixed 28 households per EA were selected. Then, the data collection for maternal health care service utilization indicators were done by interviewing all women in the reproductive age group (i.e., women in the age range 15 to 49) in those fixed 28 selected households. The present analy-sis considered 2563 women who gave birth in five years and 1555 women who gave birth in the two years preceding the survey from 220 clusters located in those four regions.

Study Variables

Outcome Variables

Maternal health care utilization: the three key maternal health ser¬vices (ANC, delivery, and postnatal care) were used as outcome variables and all of them were recoded as dummy variables (0 =ANC utilization <4, 1=ANC utilization >=4; 0 =home delivery (with no assistance health professional), 1 = delivery takes place at health facility; 0 =postnatal care not utilized, 1 = postnatal care utilized in the first 2 day after birth).

Exposure Variables

Cluster level variables: Year of woman education, woman age, total children ever born, wealth index, Media access (generated from TV, Radio, internet and magazines by using PCA), women working status, husband education, and women autonomy for own health care utilization were found by polling individual data in cluster level. To create the cluster level variable values, the study used a mean value/ proportion of the individual level data by using an aggregation technique in STATA.

Individual Level: Marital Status, Type Place of Resi-dence and Region

For numerical variables this study measured by their mean and standard deviations. The categorical variables of the study clas¬sified as follows; Wealth index classified in to 5 categories i.e. poorest, poorer, middle, richer, and richest. Women autonomy for own health care utilization answered as (women decision making autonomy for own health care coded as “yes” and women consul¬tation with husband or others, and decision making by her husband or others coded as “no”. Marital status was classified in to “others” and “married”, Place of residence as “urban” and “rural”, Region was as Afar, Benishangul-Gumuz, Gambella and Somali.

Statistical Analysis

Data cleaning and analysis were done using STATA -v17. The analysis began with aggregating the individual level variables by Enumeration Areas (EAs). The variables aggregated were wom¬an education, woman age, total children ever born, wealth index, and media access, husband education, and women autonomy for own health care utilization and women work status. The remain¬ing variables such as a marital status, type place of residence, and region were not aggregated and were considered as individual lev¬el variables. We used ArcGIS 10.6.1 to display the utilization of maternal health care services by each EA. The Moran’s Index was calculated to assess this distribution.

Spatial autocorrelation (Global Moran’s I) statistic measures to see whether the optimal maternal health care services utilization pat¬terns were dispersed, clustered, or randomly distributed in the four regions of Ethiopia. Moran’s I is a spatial statistic which is valid to estimate the spatial autocorrelation by considering the entire data set but it produces a single output value which usually ranged from − 1 to + 1. The output of Moran’s I values is interpreted as; when Moran’s I close to − 1 indicates the maternal health care service utilization is dispersed, whereas I close to + 1 indicate the maternal health care service utilization clustered; the maternal health care service utilization is said to be distributed randomly if Moran’s I value is zero. A statistically significant Moran’s I (p < 0.05) is lead¬ing to the rejection of the null hypothesis (example facility deliv¬ery is randomly distributed) and indicates the presence of spatial autocorrelation [29].

For the regression analysis, bivariate logistic regression was used to select the potential variables using a cut off p< 0.2. Those vari¬ables with p<0.2 were further assessed in the multivariable logis- tic regression analysis to see their net effect on the three outcome variables. Multicollinearity among the independent variables was checked using VIF and values >2.5 were removed. The prelimi-nary analysis indicated that region, total children ever born, and women education had significant multicollinearity with Variable Inflation Factor (VIF) >2.5. Therefore, the study excluded those three predictors from the logistic regression analysis. The Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) were used to select the model. In addition, Blinder Oaxaca decomposition technique was used to assess rural urban disparity of MCH service utilization.

Blinder Oaxaca Decomposition Technique is particularly useful for identifying and quantify the individual contribution of group differences in measurable characteristics such as education, work experience, marital status, and geographic location, gender and ethnic differences in outcomes. Although not used as often, this technique is also useful to identify reasons for geographic differ-ences (e.g., urban/rural), time-period difference, or other categori-cal differences [30]. In recent years, we have found studies show-ing how the Blinder-Oaxaca decomposition technique was used to explain inequalities in health outcomes between two groups. The technique identifies the factors that made the difference, and it is necessary to allocate the mean difference between the two groups (group 1 and group 2) for some ongoing health outcomes [29].

For this study, the outcome variable of interest is key maternal health care service utilization (ANC, place of delivery, and PNC). To explain the rural-urban disparities in maternal health care ser-vice utilization among mothers residing in those four regions, we used the Blinder-Oaxaca decomposition. This technique decom-poses the gap in maternal health care utilization between urban and rural areas into two parts; a part that is due to difference in the distribution of the determinants of maternal health care utilization (covariates effect) between those two areas, and another part that is due to the difference in the effect of these determinants (coeffi¬cients effect) between urban and rural areas.

For example, if yi, our outcome variable, is affected by a single variable, x, and we have two groups, urban and rural, then mater¬nal health care utilization for the rural, and urban mothers are giv¬en by Eqs. (1) and (2) respectively. The superscripts in the below equations are r is stands to describe rural and u stands to describe urban

yi r = βr xi + εir (Eq.-1)

yi u = βu xi + εi u (Eq.-2)

therefore, the urban-rural gap in the mean maternal health care uti¬lization in the four regions of Ethiopia (yiu− yir), is given as in Eq. (3).

yu − yr= βu xu - βr xr (Eq.-3)

Where xu and xr are the independent variable at their means for the urban and rural. The overall urban- rural gap could be decomposed into a gap attributable to the difference in the level of the covari¬ates, X's, and a gap that is attributable to difference in coefficients, β's as in Eqs. (4) and (5):

yu − yr = Δx βr−Δβxu................................. (Eq.-4)

yu − yr = Δx βu−Δβxr.................................. (Eq.-5)

Where Δx = xu − xr and Δβ = βu − βr. Here the study decomposi-tion equation can be re-written as in Eq. (6):

yu − yr = Δx βr+ Δβxr+ ΔβΔx

= E +C+CE …………………………….……………………….

(Eq.-6)

Therefore, in this study the overall urban-rural gap in maternal health care utilization in four regions of Ethiopia consists of the gap in endowment (E), and the gap between coefficients(C), and the interactions (CE).

Result

Background Characteristics of Study Participant

For the maternal health care services, a total of 2563 women were interviewed in 220 clusters (EAs) which comprises mothers who gave birth within five years before the survey. The cluster /commu-nity/ level background characteristics of those women who were included in this study from the four regions of Ethiopia is summa¬rized and presented in Table 1 below.

The result indicates that the mean age of women included in the study was 28.6 (SD =2.38, 95% CI (28.53- 28.71)), the mean chil- dren ever born was 4.2 (SD =1.09, 95% CI (4.15- 4.24)), the mean year of school attendance was 1.51 years (SD= 1.9, 95% CI (1.45-1.58)) and on average respondents of the study had at least 1 media exposure ( SD= 2.16, 95% CI (0.94-1.11) (Table 1). The individual level background characteristics of the respondents are summa¬rized in table-2. It is seen that 93% of them were married, and 84% of them lived in rural area Table 2.

Table 1: Community Level Background Characteristics of Women in Four Regions of Ethiopia

Table 2: Individual Level Background Characteristics of Women in Four Regions of Ethiopia

For PNC service utilization, the data were collected from a total of 1501 women residing in 220 clusters who delivered a baby with¬in two years before the survey. The community level background characteristics of those women who utilized the PNC service and included in this study is summarized in Table 1. It is seen that the mean age of the women was 28.48 (SD =2.33, 95% CI (28.37-28.6)), the mean children ever born was 4.24 (SD= 1.07,(4.18-4.29)), the mean year of school attendance by women was 1.46 years (SD= 1.66, 95%CI (1.37-1.54)) and on average respondents of the study had at least 1 media exposure ( SD= 2.16, 95%CI (0.89-1.11) . In addition, of those who utilized the PNC, about 81% of them were not working during the survey (Table 1). The individual level background characteristics of the women across the four study regions is summarized in Table 2 also indicates 95% of them married, and 84% of them lived in rural areas (Table 2).

Spatial Patterns in the Utilization of Maternal Health Care in Four Regions of Ethiopia

Figure 1 shows a geographic distribution of maternal health ser¬vice utilization among women in the age group 15–49 in 220 clusters situated in the four regions of Ethiopia respectively. The results indicate the utilization of a maternal health services (ANC, delivery at health facility and PNC) ranges from 0% to 100%. Uti¬lization of maternal health services is lowest in Somali and Afar regions as compared to Benishangule-Gumuz, and Gambela re¬gions. In Afar region, there are few clusters which have highest utilization of some of the maternal health services but the presence of geographical clustering was analyzed for all selected outcome indicators i.e., maternal healthcare services (ANC, place of de-livery and PNC) using global spatial auto-correlation (Moran’s’ Index) at cluster levels as presented in figure 2. Maternal health- care services considered in this study there were an indication of geographic clustering in the utilization of all maternal healthcare services considered in this study i.e., ANC (Moran’s I = 0.185031, p = 0.011056), health facility delivery (Moran’s I = 0.271865, p = 0.000209) and PNC (Moran’s I = 0.315410, p = 0.000015).

Figure 1: Maps Showing Geographic Distribution of Maternal Health Service Utilization in 220 Clusters of Four Regions of Ethiopia

Figure 2: Spatial patterns of maternal health care utilization in 220 clusters in the four regions of Ethiopia. From the graphs show A) Antenatal care visits, B) Delivery at Health facility, and C) Postnatal cheek up visit.

Bivariate Logistic Regression

The results in Table 3 presented the bivariate logistic regression for the selected potential predictors. Based on the review of re-lated literature, the study used a p-value < 0.2 as cut off value to consider the significant predictors as potential variables for fur-ther analysis in the multivariable logistic regression. Therefore, in ANC utilization for both urban and rural residences; except a variable called “autonomy health care utilization” the remain¬ing variables considered in this bivariate logistic regression were taken for further analysis in multivariable regression analysis. In place of delivery, for urban residence except marital status all the remaining variables were considered in a multivariable regression analysis. But in rural settings; except women age and household size, all the remaining predictor variables in the bivariate logistic regression were taken for further analysis in a multivariable re¬gression analysis. In PNC service utilization, for urban residence a predictor variables wealth index, husband education level, women working status, marital status, media exposure, ANC service uti¬lization and place of delivery was considered for further analysis in the multivariable regression analysis. But for the rural setting; a predictor variables household wealth, husband education level, women working status, media exposure, ANC service utilization and place of deliveries were considered for further analysis in the multivariable regression analysis.

RC: Reference Category

Table 3: Bivariate Logistic Regression for Selected Predictors of Maternal Health Care Utilization in Urban and Rural Ethiopia

Predictors of Maternal Health Care Service Utilization

The study results in Table 4 revealed that odds of ANC service utilization increased by 1.09 times when the Media exposure in-creased (AOR = 1.09, 95% CI (1.02, 1.16)). In addition, rural set-tings women living in households with middle wealth index (AOR = 3.36 95% CI (2.05–54.9)), richer (AOR = 6.07, 95% CI (3.73, 9.87)), and richest (AOR =8.90, 95% CI (4.80, 16.48)) were more likely to utilize ANC service compared to those poorest wealth quintiles. The odd ratio of utilizing health facility delivery service were 3.19 times higher for women who had at least 4 Antenatal care (ANC) in urban settings compared to those who have < 4 ANC during their pregnancy in similar settings (AOR =3.19, 95% CI (1.99–5.13)). The result in rural settings also showed a positive association i.e., the odds of utilizing health facility delivery were 4.56 times higher for women who had ANC during their pregnan-cy compared to those who had not have ANC (AOR = 4.56, 95% CI (3.48–5.98)). Further, the odds of health facility delivery also increased by 1.13 times when the media exposure increased (AOR = 1.13, 95% CI (1.04–1.22)) in urban settings and this increase also revealed in rural settings i.e., we have found a positive asso- ciation that indicates an increase of health facility delivery utiliza¬tion by 1.18 times when the husband years of education increased (AOR = 1.18, 95% CI (1.03–1.37)).

The odd ratio of utilizing a postnatal care (PNC) were 47.6times for women who gave birth in health facilities compared to those who did not give birth in health facilities in urban settings (AOR=47.6, 95% CI (6.32-359.03)). But the result in the rural settings revealed a little bit increase i.e., the odd ratio of utilizing a PNC service were 49.87 times for women who gave birth in health facilities compared to those who did not give birth in health facilities in rural areas (AOR=49.87, 95% CI (26.54,92.72)). Additionally, the result in table-4 also indicates that the odds of utilizing a PNC service were 2.3 times for women who used ANC service during their pregnancy as compared to those who did not use ANC ser¬vice in urban settings (AOR=2.3, 95% CI (1.15-4.60)) .Regarding PNC service utilization, wealth index middle (AOR = 3.36 95% CI (2.05–54.9)), richer (AOR = 6.07, 95% CI (3.73, 9.87)), and rich- est (AOR =8.90, 95% CI (4.80, 16.48)) were more likely to utilize ANC service compared to those poorest wealth quintile.