STRENGTHS AND LIMITATIONS OF THIS STUDY
-
The intersectionality framework allowed the researcher to unveil subpopulations of women who were at the crossroads of multiple axes of marginalised social identities.
-
A more finely-grained examination of disparity using this framework would help introduce interventions that target the most underprivileged groups to ensure equitable access to maternal health care services.
-
Under the principle of intersectionality, interaction-based studies of health care equity do require large number of observations to ensure that CIs do not overlap. However, even though the Ethiopia Demographic and Health Surveys offfer the luxury of large samples, 95% CIs of many intersections overlap, making it challenging to compare results in such intersections.
-
Besides, the number of intersections and main effects included in models was determined by the sample size. As a result, a large number of intersections did not find their way into the study, potentially influencing the findings. Finally, in this study, the investigator attempted to make associations, and the findings should not be interpreted as causations.
Introduction
Globally, coverage of skilled birth attendance (SBA) and postnatal care (PNC) services has gone up over time. However, SBA remains one of the most unequally distributed services in the world,1 and the noticeable gaps in these services were observed across different social identities such as location, wealth, women’s educational status, empowerment of women, ethnicity and race.2–4 It is found in the literature that disparities in the utilisation of maternal health care services are the result of gaps in access to the services to the disadvantage of marginalised women.5
Disparities in the utilisation of health care contribute to poor maternal health outcomes,6 and reducing gaps in poor maternal health outcomes involves extensive work on improving quality of maternal health care,7 where equity is a part.8 Literature suggests that disparities in access to maternal health care services have direct implications on disparities in maternal deaths both within and between countries and continents. In 2020 alone, maternal mortality ratios of 477–654 deaths per 100 000 live births were reported from low-income countries compared with a tiny ratio of just 12 maternal deaths per 100 000 live births in developed countries.9 10 Nearly 9 in 10 maternal deaths globally are found in sub-Saharan Africa (SSA) and Southern Asia, with the SSA contributing 70% of the deaths.9 Despite efforts, Ethiopia continues to face alarmingly high maternal deaths. It is not just that Ethiopia is home to one of the highest burdens of maternal mortality globally10—roughly 4% of all maternal deaths—but that there is also large within-country variation to the disadvantage of underserved populations.11 Literature showed that the lifetime risk of maternal death dropped from 4.6% in 2000 to 1.9% in 2016 in Ethiopia,11 reflecting a ‘maternal mortality transition’ over time. However, theory-driven equity studies are needed on maternal health care services to inform the development of targeted interventions for underserved populations, thereby reducing gaps in maternal mortality to the point of no concern.
Attainment of Sustainable Development Goal (SDG) 3 requires, among other things, tracking of progress on process indicators such as SBA.12 However, monitoring of progress in equity requires data on these indicators for all subgroups of women including the most underserved ones. Currently, there exist works examining inequality in maternal health care using traditional unitary epidemiological approach. However, Hancock, 2007, as cited in Bauer,13 pointed out that this approach conceals health care use experiences of women who are disadvantaged on multiple social identities simultaneously. Yet, without producing data on maternal health care indicators for these women, universal access to these services would not come true. To this end, previous inquiry suggests using intersectionality theory to promote population health research and improve health equity.13
Intersectionality theory had its origin in black ‘feminist scholarship’ by the African-American scholar Kimberlé Crenshaw (Crenshaw, 1998, as cited in Bauer, 2014).13 The theory argues that various disadvantages are ‘mutually constituted’ and that these disadvantages could not be comprehended by research techniques that view gender and race as separate areas of investigation. It contends that upstream social contextual forces such as classism and racism interact with each other to give rise to health injustices. Interestingly, the framework extends to studies that investigate variables other than race and gender, such as socioeconomic status.13 14
Although qualitative investigation is the ‘most developed’ approach for intersectionality,15 the theory has lately gained traction in quantitative inquiry spanning many fields of study including public health.13 16 Evidence shows that most published works17 adopting this theory use an ‘intercategorical’ analysis approach described by McCall.18 Practically, the approach is commonly applied via, for example, computation of what is known in the language of statistics as ‘interaction effect’.17 In literature, researchers use different ways to integrate intersectionality into their studies, such as regression using intersection variables, regression using main effects, multilevel modelling, structural equation modelling and decomposition.17
There exists evidence on inequalities in skilled maternal health care services in Ethiopia.19–22 However, these studies mostly rely on the traditional single risk factor approach, and there is a lack of intersectionality driven studies. The unidimensional analysis omit the ‘complex ways’ in which numerous social identities intersect with discrimination to lead to disparities.16 A study, despite not explicitly integrating intersectionality into its method, examined use of skilled maternal health services at the intersection of wealth and maternal education.23 However, investigation of disparity by relevant intersections, in addition to those created by education and wealth status, is essential to expose underserved women experiencing multiple marginalisations simultaneously and is helpful to craft targeted interventions.
The 2016–2030 Global strategy for women’s, children’s and adolescents’ health calls for research and innovation to achieve the three goals of the strategy—survive, flourish and transform.24 Additionally, one of the objectives of the global roadmap, Ending Preventable Maternal Mortality strategy, is to address disparities in access to maternal and newborn health care as well as to ensure the Universal Health Coverage (UHC), the achievement of which requires analysis of health care gaps based on the principles of human rights.25 The current study intends to apply the new approach to comprehend disparities in the utilisation of SBA and PNC in Ethiopia. The study would assist in identifying women who are at several axes of disadvantage at the same time. In the context of social justice approach to social inequity-driven disparities, priority needs to be placed on subgroups of people who are at multiple marginalisations.13 Therefore, this study aims to investigate intersectional disparities in the receipt of skilled maternal health care services using the intersectionality theory as a guiding framework.
Methods
Data sources and setting
Data for the study came from the Ethiopia Demographic and Health Surveys (EDHSs). Between the 2000 and 2016, four major surveys were undertaken. Data from all four surveys were used for the analyses of SBA. However, for PNC, the analysis was restricted to the data in the last wave of the EDHS, i.e., 2016, since the data from the other surveys were too small to allow intersectional analysis for some variables. The EDHSs are part of the global DHS project and collected data nationally on many areas including maternal health, women’s empowerment, child health, domestic violence, nutrition, HIV/AIDS, malaria, and maternal and child mortality. The primary focus of the surveys is reproductive-age women and children younger than 5 years though data are also collected on men.26 27
Sampling procedure and samples
The sampling process and samples of the surveys are discussed elsewhere.28–31 Briefly, the EDHSs follow a stratified two-stage cluster sampling technique. The two-stage sampling approach was shown to be more efficient, and precise and yield samples that are representative at the national, urban versus rural and regional levels.27 32 The right number of observations is determined by a compromise between the budget at hand and the needed survey precision. Size of a cluster (enumeration area) cannot have much effect on sampling error if the second stage sample size is fixed, and the optimal ‘sample take’ hinges on the cost ratio and intracluster correlation (ICC), and calculating the ICC ultimately amounts to determining the ideal sample size.27 32
The surveys apply a two-level stratification, which divides residents into urban and rural areas within each of the nine regions and administrations of the two cities. After stratification, primary sampling units—referred to as enumeration areas in the context of EDHS—were chosen systematically using the probability proportional to size technique, where large clusters have a higher likelihood of being included in the sample. This constitutes the first step of sample selection. In the second step, a predefined number of households were systematically chosen from each enumeration area. A detailed account of the DHS sampling process can be found elsewhere.27
Data on maternal health care services were sought from women who had live births 5 years preceding the surveys. Eligible for the surveys were women in the reproductive period who slept in the selected households the night before the surveys. From the households recruited for the surveys, 15 367, 14 070, 16 515 and 15 683 women were interviewed in 2000, 2005, 2011 and 2016, respectively. The corresponding response rates were 97.8%, 95.6%, 95.0% and 94.6%. Data on SBA included 10 873, 9861, 11 654 and 10 641 children born 5 years preceding the surveys in 2000, 2005, 2011 and 2016, respectively. All four rounds of the EDHSs were used for the trend analysis for SBA, and a separate model was fitted for each of the four surveys. Additionally, the 2016 survey was used for a more in-depth examination of intersections. However, for PNC, only the 2016 EDHS was used since the other surveys did not allow for a more in-depth study of intersections. The women’s recode dataset (IR dataset) for PNC and children’s dataset (KR dataset) for SBA were used for the analyses.33 Following DHS’s practice, the analysis for PNC in this study was restricted to women who had live births 2 years preceding the survey (2016 EDHS), resulting in the final observations of 4081. For SBA, however, analyses were done on live births born 5 years preceding the surveys.
Measures
The outcome variables were SBA and PNC for mothers within the first 2 days following delivery. The PNC does not include care for the newborn. Both outcome variables were collected from the participant’s self-report. SBA and PNC constitute services given to mothers during and after delivery.
The outcome variables’ definitions given in this study conform to the definitions of the same variables provided by the WHO. They refer to services rendered by skilled or competent health care workers, consisting of physicians, nurses, midwives and health officers. A skilled health professional is a member of the medical community who has the required training, according to the WHO, to attend a typical pregnancy, delivery and post partum as well as identify, treat and refer complications.34 All EDHSs follow the WHO’s definitions of skilled health professionals except that the 2016 EDHS also counts health extension workers as skilled health professionals. However, since health extension workers do not satisfy the WHO’s definition of competency, they are not included in this study. The outcomes are binary measures, with 1 indicating that one or more skilled health workers provided the services and 0 indicating otherwise.
Exposure factors thought to be connected to these two maternal health care services were included in the study.2 5 35 Variables appear to be mediators were excluded.36
No education, primary, secondary or higher education were the three categories used to group maternal and paternal education. Husband education had missing values, with the highest being 27 (0.28%) unweighted observations in 2005 and were coded together with the ‘no education’ category. However, in 2000 EDHS, its missing on 11 (0.1%) unweighted observations were found merged with the system missing in the dataset and were therefore dropped from the analysis because there is no way to differentiate them from missing values that are not due to errors. Urban versus rural was used as a dichotomy for place of residence. Utilisation of maternal health care services has been shown to vary according to age at birth, with adolescent women having a lower chance of obtaining the services.37 In this study, therefore, age at birth was classified as adolescent (age <20 years) vs not adolescent (≥20 years). Media exposure was divided into two categories: not exposed to any media at least once a week and exposed to at least one of the three media, namely, radio, television, newspaper or magazine at least once a week. These three media each had missing values, with the highest being 18 (0.18%) unweighted observations in 2005 and these observations were coded with the ‘not exposed’ category. Women’s and partners’ occupations each had missing observations, with the highest being 113 (0.97%) and 100 (0.86%) unweighted observations in 2011, respectively. After missing observations were coded with the ‘no occupation’ category, occupations were divided into two groups: no occupation and occupation.
There were three categories of pregnancy intentions: wanted, mistimed and unwanted. Birth order was divided into four categories: 1, 2–3, 4–5 and 6+. Orthodox, Protestant, Muslim and other were the four categories used to classify religion. For the PNC, ‘others’ was merged with Muslim due to sample size issue. To measure women’s empowerment, the study used the SWPER Global women’s empowerment index, developed by Ewerling et al (2019).38 In low-and middle-income countries, the SWPER Global index assesses married women’s empowerment in three areas: decision-making, social independence and attitude towards abuse. The study used wealth index that the DHS developed applying assets, materials for dwelling construction, water and sanitation facilities and other things. It has five categories: poorest, poorer, middle, richer and richest. The construction of the wealth index has been described elsewhere.39 Region is divided into nine divisions and two city administrations. Missing and unknown replies were dealt with in compliance with the 2018 DHS statistics guide.40
Statistical analysis
Analyses are guided by the principle of intersectionality which contends that multiple social locations are mutually constitutive and that methodology that regards individual social positions as discrete subjects of research cannot comprehend them.13 Literature indacted that computations of absolute measures for prevalence of health indicators for all possible intersections would be a new way to appreciate their distribution across intersections13 and this can be done using the intersectional analytic framework. The intersectional analysis was performed using statistical interactions. One of the quantitative strategies used in the literature to integrate intersectionality into research is regression with interactions.13 14 41 Interactions were constructed between selected exposure variables, and all constituent elements of the interactions were included in model to assist in the appropriate estimation and interpretation of the interactions.42 Adequacy of observations for each intersectional category was judged following practices in the literature. In this regard, the study employed a rule of 10 Events Per Parameter (EPP) to choose the right number of main effects and interactions in models and by extension, to prevent models from becoming overly complex.43 That is, each intersection created beforehand was cross-tabulated with the outcome variables to see whether there are sufficient observations in each cell, that is, a minimum of 10 observations. Once the sample adequacy was ensured, then full regression model was built, with all the main effects and interactions included in models simultaneously. After running the regression models, the e(sample) Stata’s function was run to mark observations used to produce estimates. Using observations in the models in the ‘yes’ category of the outcome variables, I reinvestigated the sample adequacy using cross-tabulation, and only intersections found to have the minimum number of observations were included in the final models. The procedure has been documented in a Stata do file and is available as online supplemental files 1; 2 so that researchers can see what the model-building process looks like in the study.
Supplemental material
Supplemental material
This step by step approach helps prevent overfitting and produce good estimates. Building good model is a science as well as an art that makes use of theoretical perspectives, statistics and working knowledge44 and I followed this practice for my model building. For the statistical approach to model building, the Stata command svylogitgof is used to evaluate performance of the models.
Research suggests that studies place heavy emphasis on the statistical importance of results and their signs, giving little attention to substantive and practical values.45 However, researchers can make their research findings more palpable by generating predictions after fitting a model (Long and Freese, 2006, as cited in Williams, 2012).45 Furthermore, analysts ought to adhere to established procedures for the analysis and interpretation of interactions that differ from conventional methods that do not include interactions.42 Following this recommendation, the study used the margins Stata postestimation command to compute predictive probabilities and Average Marginal Effects (AMEs). AME for the selected interactions was plotted using the coefplot Stata command.46 Additionally, for the sake of transparency and thorough reporting, exponentiated coefficients were presented. According to Karaca-Mandic et al (2012), exponentiating regression coefficients of an interaction term in a logit model results in a ratio of Odds Ratio (OR) rather than an OR,47 which readers should interpret as such.
The study follows current recommendations on scientific investigations. Specifically, I rely heavily on point and CI estimations and avoided the term ‘statistical significance’ since it is a misguided statistical concept that needs to be abandoned.48–51 Because the outcome variables are binary, the logit model was used. All analyses used the svy Stata module to account for clustering, stratification and unequal selection probability of observations. Stata v.16 was used for all analyses. Results of the study were reported in conformity with the Strengthening the Reporting of Observational Studies in Epidemiology reporting standard for cross-sectional study.52
Patient and public involvement
Because the study used publicly accessible data, the researcher did not interact with any patients or members of the general public while doing the study.
Results
Descriptive statistics
There were 10 873, 9861, 11 654 and 10 641 live births in 2000, 2005, 2011 and 2016, respectively, of which 9867, 9075, 10 592 and 9915 had complete data allowing for fitting models on SBA. Overall, 1140 (5.7%; 95% CI 4.8% to 6.6%), 1032 (5.7%; 95% CI 4.9% to 6.6%), 1581 (10.0%; 95% CI 8.7% to 11.5%) and 3393 (25.9%; 95% CI 23.2% to 28.7%) live births 5 years before the surveys were attended by skilled health workers, respectively, in 2000, 2005, 2011 and 2016. The 2016 survey had 4081 women with live births 2 years preceding the survey, of which 3843 had complete data allowing for fitting the model on PNC. It was found that 815 (15.9%; 95% CI 13.9% to 18.0%) women received PNC within 2 days of birth in 2016.
Analyses of the surveyed women’s characteristics revealed large differences both within and between the surveys (online supplemental table 1). Adolescent women made up about 1/10 of the sample, and their share did not improve over time. Approximately, 8 out of 10 women in the surveys lived in rural areas. The percentage of women who attended higher education climbed from less than 1% in 2000 to approximately 6% in 2016, coinciding with a drop in the number of women who did not have an education from 75% in 2000 to just under 50% in 2016. For more information, see online supplemental table 1.
Trend of intersectional disparity in SBA
Predictive probabilities were computed for each intersection for the four surveys and are shown in online supplemental table 2. The exponentiated regression coefficients are presented in online supplemental table 3.
The trend analysis showed that interaction between maternal education and residence occurred in all rounds of the surveys, and that coverage of SBA improved over time in all intersections, especially between the first and the last surveys. However, the pace of change varied noticeably, with better-performing intersections showing higher improvements.
Among all the intersections created by interacting women’s education with a place of residence, the predictive probability of SBA was the highest among secondary-schooling women who dwelt in urban settings. For example, the prediction in this intersectional group was 0.255; 95% CI 0.113 to 0.397 in 2000 and 0.589; 95% CI 0.359 to 0.819 in 2016. On the other hand, the lowest prediction was found among non-educated women who lived in rural areas with values of 0.0236; 95% CI 0.0154 to 0.0317 in 2000 and 0.203; 95% CI 0.177 to 0.229 in 2016. In terms of percentages (predicted probabilities can be presented in percentages), this means that, for example, educated women living in urban areas had a predicted point estimate of 26% in 2000, with the interval ranging between 11% and 40%. This stands in stark contrast to the coverage among non-educated women living in rural areas, which was about 2% in 2000, with intervals ranging between 2% and 3% and was 20% in 2016, with intervals ranging between 18% and 23%. These findings demonstrated that multiply advantaged women, women at the intersection of multiple advantageous positions such as being educated and living in urban areas, had better chance to use the service than the multiply disadvantaged women, such as non-educated women who lived in rural areas. Figure 1 depicts the AME of rural residence (compared with urban) for each category of maternal education for each survey. The figure indicated that in each category of education, there was rural-urban gap, and the disparity grew between the first and the last surveys, at least according to point estimates.
Average marginal effects on skilled birth attendance of residence across women’s education, 2000–2016 EDHS. EDHS, Ethiopia Demographic and Health Survey.
Age at birth had a varying degree of influence on the receipt of SBA in urban and rural places, suggesting that interaction happened between them. For example, adolescent women living in rural areas had the lowest predicted coverage with 0.024 (95% CI 0.011 to 0.038) in 2000 and 0.213 (95% CI 0.171 to 0.256) in 2016, followed by non-adolescent women living in rural areas with 0.041 (95% CI 0.031 to 0.050) in 2000 and 0.236 (95% CI 0.214 to 0.258) in 2016. On the other hand, non-adolescent women who resided in urban areas had the highest prediction of 0.121 (95% CI 0.070 to 0.171) in 2000 and 0.466 (95% CI 0.344 to 0.588) in 2016.
In terms of intersections created using media exposure and residence, the highest probability of SBA took place among women who lived in urban areas and the lowest was among women who lived in rural areas irrespective of their media exposure, suggesting that media exposure seemed to not affect the uptake of the service (see online supplemental table 2 for the details).
Intersectional disparity in SBA based on the 2016 EDHS
Based on the analysis of the 2016 data, coverage of SBA differed among the intersections. Birth order’s influence on the probability of births attended by competent health personnel was different between urban and rural places. The highest probability of uptake was predicted among women having first-order children who lived in urban areas (0.703; 95% CI 0.548 to 0.858) and the lowest was among rural women with a birth order of 6+ (0.183; 95% CI 0.153 to 0.213), resulting in 0.52, or 52 percentage points difference. Among the intersections created by interacting women’s schooling with wealth index, the intersection with the highest predicted probability comprised of women who completed secondary or higher education and were in the richest wealth quintile (0.405; 95% CI 0.187 to 0.623). It was followed by the intersections that included primary schooling women who lived in the richest category of wealth (0.374; 95% CI 0.274 to 0.475) and women who completed secondary or higher education and lived in the poorer household (0.359; 95% CI 0.161 to 0.556) though the CIs of these two intersections overlap. On the other hand, the poorest women had the lowest predicted probability, irrespective of their educational status. For example, non-educated and poorest women had a probability of 0.120 (95% CI 0.089 to 0.151), which almost coincided with 0.117 (95% CI 0.039 to 0.195) for women who completed secondary or higher education and were the poorest.
Women at the intersection of rural residence and non-education had a prediction of 0.204 (95% CI 0.178 to 0.229). This was the lowest compared with that of the other intersections created between maternal education and residence. The AME indicated that rural residence had, based on the point estimates, 17–25 percentage points lower predicted probability than urban areas, with the rural–urban disparity falling as the education status of women increased. This finding signified that the highest rural–urban disparity occurs among women who were not educated (figure 2). The other intersections that had low predictions included non-educated adolescent women (0.166; 95% CI 0.109 to 0.222), adolescent women who were the poorest (0.125; 95% CI 0.073 to 0.176) and non-educated adolescent women who lived in a rural area (0.147; 95% CI 0.088 to 0.207). Online supplemental table 4 shows the predicted probabilities of SBA for all the intersections in 2016, and the exponentiated regression coefficients are presented in online supplemental table 5.
Predicted probabilities of skilled birth attendance for the intersections created by maternal education and residence and average marginal effects of residence, 2016 EDHS. AME Average Marginal Effect; EDHS, Ethiopia Demographic and Health Survey.
Intersectional disparity in PNC based on 2016 EDHS
Due to low samples for a certain number of intersections, CIs overlap and they did differ mainly by their point estimates. Maternal age at birth and maternal education had an interaction, with the result that non-educated adolescent women had the lowest prediction of 0.122 (95% CI 0.057 to 0.187) while non-adolescent women who completed secondary or higher had the highest predicted coverage of 0.205 (95% CI 0.092 to 0.318). Although the two intervals overlap, point estimates gave rise to a difference of more than 8 percentage points. Among the six intersection groups created by the interaction of maternal education and residence, women who completed secondary education and lived in urban locations had the highest prediction of 0.277 (95% CI 0.011 to 0.542). On the other hand, the lowest coverage was predicted among non-educated women who lived in rural places with 0.119 (95% CI 0.087 to 0.152). This showed, according to the point estimates, an absolute difference of 16 percentage points between the doubly advantaged (being educated and living in urban area) and doubly disadvantaged (being not educated and living in rural areas) women. Additionally, the AME of rural residence (compared with urban) was the highest among non-educated women with 13 percentage points, signifying the varying role of rural areas across maternal education levels (figure 3).
Predicted probabilities of skilled postnatal care for the intersections created by maternal education and residence and average marginal effects of residence, 2016 EDHS. AME Average Marginal Effect; EDHS, Ethiopia Demographic and Health Survey.
The interaction between residence and birth order revealed that the intersections had predictions whose CIs overlap. According to the point estimates, however, the highest PNC coverage was predicted for urban women with a birth order of 6+ with 0.295 (95% CI 0.115 to 0.476) and the lowest was among rural women with a birth order of 6+ with 0.147 (95% CI 0.105 to 0.190). As birth order increased, the predicted probability increased among women who lived in urban areas. For women in rural locations, it increased slightly up to the fifth birth order, after which it decreased.
However, there were intersections whose interval estimates did not overlap. For example, among the 10 intersection groups in the interaction between wealth and age at birth, the highest prediction was observed among adolescent women who lived in the richest households with 0.496 (95% CI 0.175 to 0.816), followed by non-adolescent women in the richest households with 0.246 (95% CI 0.160 to 0.332). Online supplemental table 6 offers detailed information on the predictive probabilities of the PNC. The exponentiated regression coefficients of the factors associated with PNC are found in online supplemental table 7.
Discussion
Using intersectionality as a guiding theoretical tenet, disparities in the utilisation of SBA and PNC services were examined in Ethiopia. Overall, the results showed some intriguing inequality patterns across intersections, with women in a certain multiply disadvantaged positions having poorer chances of obtaining the services. Improvements in SBA coverage were noted in all intersections over time. However, women in some intersectional positions lagged behind others, suggesting that these groups of women would require targeted interventions to ensure that all women use the services.
The study uncovered that predicted coverage of the services varied between intersections and that some intersections constituted by multiple disadvantages had the lowest predicted probabilities, mainly based on point estimates. For example, pertaining to intersections between women’s education and their place of residence, women who were not educated and lived in rural locations had the lowest prediction of SBA. Similarly, interaction between women’s and partner’s education resulted in SBA being the lowest among uneducated women whose husbands were either not educated or completed primary schooling. Women who completed either primary or secondary/higher schooling and were partnered with someone who completed secondary or higher education had the highest prediction of obtaining SBA. These findings illustrate how disadvantageous social locations such as, for example, lack of education and rural residence, intersect to put some women at higher risk of non-utilisation of maternal health care services. The results imply that interlocking women’s social identities result in disparity that could have been obscured by the ‘single determinant’ approach.53
Research revealed that in Ethiopia, 2.3% of births were attended by skilled professionals in rural and 34.5% in urban areas in 2000, and 21.2% and 80.1% of births in rural and urban areas, respectively, in 2016.54 Intersections of rural residence and women’s education had predictive probabilities of SBA, where some of these intersections had lower and others had higher probabilities than the average coverage of rural area. On the other hand, intersections of urban location and women’s education each had coverage far lower than the aggregate coverage for urban place. The finding supports the idea that rural and urban areas are not homogeneous55 and that intersection of residence and other factors like wealth56 is important to comprehend unique experiences of disadvantaged subpopulations.
The inverse association of maternal health care and birth order has been documented in literature.57 Our study demonstrated that intersections of birth order and residence had different predicted probabilities of SBA, with the probabilities decreasing as birth order increases in both urban and rural areas which is consistent with the published literature.57 The highest coverage was found among women who lived in urban setting and had a birth order of 1 (70.3%) and the lowest was among women who lived in rural area having a birth order of 6+ (18.3%), resulting in a difference of 52 percentage points. Further, in urban areas, the disparity between women with birth orders of 1 and 6+ was 40 percentage points. This difference is 16.4 points in rural area. This suggests that not only do birth order and SBA have an inverse relationship, but that this inverse relationship is stronger in urban areas than in rural places. Ultimately, the findings could support the notion that urban and rural places are heterogeneous communities,58 suggesting the need for context specific strategies that would help narrow disparities. However, the inequality pattern was different for PNC, where the predicted percentages climbed with birth order in urban regions. In rural area, it increased until the 5th birth order but decreased thereafter. The finding, however, still embraces the idea that some intersections, such as living in rural area and having high birth order, require separate and additional commitments to improve equity in access to and use of maternity care.
The scarcity of studies on disparities in maternal health care drawing on the intersectionality theory limited the findings of this study to be situated within the existing literature.59 Wuneh et al investigated the influence on birth care of the interaction between wealth and maternal education in selected regions in Ethiopia, controlling for some other covariates.23 The present findings are consistent with those of Wuneh et al who showed that women at the intersection of being wealthier and having higher education were more likely to use SBA. Studies done elsewhere, such as in India60 and Nepal61 similarly found that intersections formed by multiple forms of disadvantage and privilege had, respectively, worse and better chances of using maternal health care services. The study from Nepal, however, omitted the main effects from the regression model,61 and the interaction effects may thus be biased.42 While low sample size of the intersections of wealth and residence prevented me from studying them, prior studies showed that disparities occur in the use of maternal health care services among intersections created by these variables56 62 though these findings could suffer omitted variable bias.
Aggregate coverage of SBA and PNC increased in Ethiopia over the last two decades. Nevertheless, disparities remain by such equity stratifiers as age, wealth, region and place of residence.54 This is contrary to the country’s pledge to make reproductive health care services accessible.63 The country did not meet63 the 2020 SDG’s calls for breakdown of the national-level indicators by pertinent equity dimensions.64 It could be because national strategic plans do not include separate strategies for marginalised groups of women and lack disaggregated indicators required to track progress in these women. Instead, the plans are mostly dominated by aggregate indicators.63 65 Furthermore, development of the country’s national strategic materials has not been underpinned by intersectionality.66 The absence of intersectionality-informed strategies may be attributable in part to a lack of local studies applying intersectionality. However, it is necessary to break down national indicators by the intersections of multiple dimensions of inequality to perform more insightful and detailed assessments of health equity.56 Identifying and reaching the most disadvantaged subpopulations of women through the intersectionality approach would contribute to the promotion of equity and help protect against the claim of ‘false universalism’ (Carey, 2014, as cited in Neal et al).62
Women at the intersection of multiple axes of disadvantage may benefit from alternative policy options such as ‘targeted universalism’,67 where the target groups, underserved women, can be easily identified through intersectionality-driven studies. However, targeted approaches could run the risk of ‘stigmatisation’ of women with marginalised identities68 unless care is taken.
The study has a few strengths. First, it would motivate creation of intersectionality informed strategies and policies in Ethiopia. Second, the study contributes to the better understanding of disparities in the receipt of health care as it helps researchers move beyond the traditional singular approach to intersectioal social identities. It uncovers disparities that would not have been visible should the singular approach be employed. Third, because findings from a single study are often not conclusive, a meta-analysis is required to produce more precise estimates,51 and this study would contribute to future meta-analyses. Finally, the computation of predictive probabilities can potentially circumvent problems inherent to interactions in a multiplicative scale that have a limited practical application.13 69
However, it is worth mentioning that the study has limitations. Although the EDHSs have the luxury of offering large sample sizes, quantitative intersectional study demands far more observations than are available in the surveys to obtain precise estimates. In this regard, while point estimates varied between intersections, 95% CI of several intersections overlap, making it difficult to determine whether these estimates differ. Nonetheless, many other intersections had more precise estimates that would help inform policies and strategies. The surveys collected nationally representative samples. However, lower number of observations than are in the surveys were used in the analysis due to, for example, a case-wise deletion, which would have affected the generalisability. Still, the number of observations used for the analyses is not too small to seriously affect validity of the results.
A few variables like ethnicity and some intersections were not included in the study due to inadequate observations, which could have impacted the findings. Furthermore, in this study, an attempt is made to make associations, and findings should not be interpreted as causations. In the EDHSs, some determinants may occur well after utilisation of the services. For example, wealth index was constructed using, among other things, household possessions such as sanitation facilities, but these possessions might have occurred 2 years after a woman had used the PNC, further complicating the claim of causation.
Conclusion
The study divulged that women’s experiences of health care utilisation, or lack thereof, are shaped by the intersection of various social locations. Methodologically, it showed how quantitative studies underpinned by the intersectionality would bring invisible subgroup of women to the forefront. For example, it led to the identification of some women at the intersection of multiple axes of disadvantages that had the lowest predicted coverage of maternal health care services.
Drawing on its findings, the study suggests that health care policies and or strategies be developed taking account of the interlocking nature of social identities in causing disparities in the utilisation of maternal health care. Such policies and strategies would create environment that facilitates the development of targeted interventions for women at the intersection of multiple axes of marginalisation, and the policies and strategies would be built on the principle of multisectoral collaboration. However, the study also calls for more intersectionality studies, underpinned by social theories, to produce strong evidence for eliminating disparities. To this end, the study recommends that future studies, preferably using big data, incorporate structural drivers to gain a more deeper understanding of causes underpinning disparities. Such studies help inform health care policies and or strategies that are geared towards upstream forces rather than blaming entirely women for disparities.
Data availability statement
Data are available in a public, open access repository. The datasets generated and/or analysed during the current study are available on the DHS website at https://dhsprogram.com/data/
.
Ethics statements
Patient consent for publication
Ethics approval
The ICF Institutional Review Board (IRB) approved the data collection instruments. Both the ICF IRB and Ethiopia IRB reviewed DHS survey tools specific to the context in Ethiopia. The ICF IRB verifies the studies conform with rules of the U.S Department of Health and Human Services to safeguard human participants. Requirements for the privacy and confidentiality is upheld for everyone who took part in the surveys. The study subjects were briefed about its purposes and participated after consenting verbally.
Acknowledgments
The researcher acknowledges the DHS program for the permission to utilise the datasets.
This post was originally published on https://bmjopen.bmj.com