Article Text
Abstract
Objective To measure inequalities in the distributions of selected healthcare resources and outcomes in Ethiopia from 2000 to 2015.
Design A panel data analysis was performed to measure inequalities in distribution of healthcare workforce, infrastructure, outcomes and finance, using secondary data.
Setting The study was conducted across 11 regions in Ethiopia.
Participants Regional population and selected healthcare workforce.
Outcomes measured Aggregate Theil and Gini indices, changes in inequalities and elasticity of healthcare resources.
Results Despite marked inequality reductions over a 16 year period, the Theil and Gini indices for the healthcare resources distributions remained high. Among the healthcare workforce distributions, the Gini index (GI) was lowest for nurses plus midwives (GI=0.428, 95% CI 0.393 to 0.463) and highest for specialist doctors (SPDs) (GI=0.704, 95% CI 0.652 to 0.756). Interregion inequality was the highest for SPDs (95.0%) and the lowest for health officers (53.8%). The GIs for hospital beds, hospitals and health centres (HCs) were 0.592(95% CI 0.563 to 0.621), 0.460(95% CI 0.404 to 0.517) and 0.409(95% CI 0.380 to 0.439), respectively. The interaction term was highest for HC distributions (47.7%). Outpatient department visit per capita (GI=0.349, 95% CI 0.321 to 0.377) and fully immunised children (GI=0.307, 95% CI 0.269 to 0.345) showed inequalities; inequality in the under 5 years of age mortality rate increased overtime (P=0.048). Overall, GI for government health expenditure (GHE) was 0.596(95% CI 0.544 to 0.648), and the estimated relative GHE share of the healthcare workforce and infrastructure distributions were 46.5% and 53.5%, respectively. The marginal changes in the healthcare resources distributions were towards the advantaged populations.
Conclusion This study revealed high inequalities in healthcare resources in favour of the advantaged populations which can hinder equal access to healthcare and the achievements of healthcare outcomes. The government should strengthen monitoring mechanisms to address inequalities based on the national healthcare standards.
 ethiopia
 gini index
 healthcare outcomes
 healthcare resources
 inequality
 inequality decomposition
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BYNC 4.0) license, which permits others to distribute, remix, adapt, build upon this work noncommercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is noncommercial. See: http://creativecommons.org/licenses/bync/4.0/.
Statistics from Altmetric.com
Strengths and limitations of this study
This study attempted to provide a comprehensive picture of the extent of healthcare resources and outcomes inequalities in relation to government health expenditure (GHE) within an under resourced country.
The application of different inequality measures and Gini inequality decomposition models helped characterise the inequalities in healthcare resources and outcomes in a decentralised health system.
The estimated Shapley value of the total GHE share of the selected explanatory variables indicated the priority resources, while the multidimensional Gini inequality decomposition provided the relative inequalities, the marginal changes and the elasticity values of the distributions.
The regional overall Gini values for the distributions analysed are biased downwards around 10%.
The analysis emphasised only the supply perspective of the health system.
Introduction
The concept of health has both moral and right elements. The central objective of many health systems is to ensure health equity among populations. Health equity may be viewed as the absence of systematic differences in health among populations regardless of their social, economic, geographical, power and prestige status.1–3 The guiding principle within this concept is health equality4 that may be achieved by making healthcare accessible and by addressing any socially unacceptable inequalities within healthcare, which are amenable to policy decisions.4–6 Thus the principle of health equality begins with creating equal opportunities for people to access needed healthcare resources,2 7 irrespective of their personal characteristics and ability to pay.8 9
The distinction between inequity and inequality can be blurred10 11 because both concepts refer to unjust and socially unacceptable differences. The unfair inequities due to avoidable causes are specific forms of inequalities.12 Nevertheless, the two concepts are not synonymous. Inequality is viewed as the quantitative description of avoidable unfair differences without value judgements that do not belong to the legitimate occurrences from individual responsibility.13 14 In the human rights field, the concept is used in a much broader sense to describe differences among individuals, of which some could be unavoidable, at least with current knowledge and approaches.15 Generally, the concept of inequality is dynamic, open to different interpretations and is highly linked to the socioeconomic structure of people.16 Like inequity, any measure of inequality involves normative judgements.17 We applied Braveman’s definition of inequality which refers to "differences in the distribution of resources or outcomes among people due to conditions that can be minimised or modified by policies".5
The success of equality policies in healthcare is subject to the influences of the political context,18 quality of information concerning the inequality17 19–21 and the appropriateness of the actions targeting identified unjust inequalities. However, government policies may favour the poor, especially when the share of the private sector is minimal,22 and the economic, political, moral or practical aspects may be used as criteria for the allocation of resources.13 Despite the unclear link between decentralisation and inequality, decentralised policies have been common practice to ensure social justice and address inequalities.23 24 The local governments in decentralised systems are likely to vary in power, boundaries, capacity, socioeconomic and demographic factors, living conditions and healthcare needs of their constituencies.20 25 26 These conditions highlight the complexity and the likely occurrence of inequalities, and the coexistence of inequalities and judgements.
Ensuring fair allocation of human and material healthcare resources to people across regions contributes to better healthcare outcomes. Healthcare inequality is one of the conditions that hinders the success of healthcare systems and has been a concern to policymakers and planners. However, little evidence is available regarding the extent of healthcare inequality in the decentralised system of Ethiopia. Our study, therefore, aims to measure inequalities in selected healthcare resources and outcomes from the year 2000 until 2015. The findings are anticipated to contribute to a better understanding of the effects of the existing health policies, provide information for action towards minimising unfair inequalities, contribute to policy decisions for strengthening the universal health coverage and eventually contribute to achieve the sustainable development goals (SDGs) for health in Ethiopia and perhaps beyond.
Methods
Setting
Ethiopia is a federal democratic country that consists of nine national regional states and two chartered cities (hereafter regions). According to the Central Statistical Agency (CSA) of Ethiopia, 80.6% of the total population for the year 2015 resided in rural areas. The geographic location of the regions, population proportions and other indicators for the year 2015 are presented in figure 1.
Study design and data source
We performed panel data analysis to measure inequalities in the distribution of selected healthcare resources and outcomes in Ethiopia. Panel data analysis allows a better understanding of the trends and extent of inequalities in healthcare systems.27 The panel consisted of annual data for 11 regions (n=11) from the year 2000 to 2015. The data were related to government health expenditure (GHE), healthcare workforce, healthcare infrastructure, population and healthcare outcomes of each region. The regional data were retrieved from the Health and Health Related Indicators of Ethiopia. This bulletin has been published annually by the Policy Planning Directorate of the Federal Ministry of Health of Ethiopia since 1994. We also used census based annual population estimates for the regions by the CSA of Ethiopia, plus the 5 yearly reports of the Ethiopia Health and Demographic Surveys (EDHS) for 2000 to 2016.28–32
Patient and public involvement
We used data from the public domains and focused on the supply perspective of the healthcare system. There was no direct patient or public involvement in the data collection and analysis. This study intends to answer the following three basic distributional inequality questions in the context of Ethiopia:
Are healthcare resources and outcomes fairly distributed across the regions?
How were the trends and extent of the overall inequalities in the selected distributions, and which inequality component (interregion, within region, interaction term Gini) dominantly explained the overall inequality of each distribution?
Which healthcare resources had the dominant share of the GHE and what were the relative inequality changes with respect to the marginal change in the average GHE?
Variables (indicators)
Total GHE and total number of each selected health professional were considered to analyse the finance and healthcare workforce dimensions of the healthcare system. These dimensions are vital for the proper functioning of the healthcare infrastructure. The total number of each functional healthcare infrastructure (health centres, the different levels of public hospitals together and public hospital beds) in each region were also healthcare resources related variables. The annual hospital outpatient department (OPD) visits per capita, the proportions of fully immunised (FIMM) children, and the Ethiopia Health and Demographic Surveys' 5 yearly reports on under 5 child mortality rates (U5MR) and infant mortality rates (IMR) per 1000 live births for each region were healthcare outcome related indicators.
GHE is a crucial determinant of healthcare,33 34 especially in countries like Ethiopia where the public sector is the main provider of healthcare services. Missing data for GHE for the years 2013 and 2014 and for physicians for the year 2015 for all the regions were estimated using the annual average growth rate of each distribution. The five central hospitals in Addis Ababa (AA), which were financed and managed by the Federal Government of Ethiopia, were included in the analysis for the AA region. The annual total of each distribution (variable) in a region was divided by the annual total population of that region. The ratio of each distribution was again weighted by a fixed number of people to ensure consistency of the indicator, because the regions differ in population size. A summary of the indicators used in the analysis is presented below.
Dimension  Indicator 
Finance: 

Healthcare workforce:  
 
 
 
 
 
 
 
 
 
Healthcare infrastructure:  
 
 
 
Healthcare outcome:  
 
 
 
 
^{*}=a + b  ^{**}=a + b+c + d 
Analysis and interpretation of inequality
We applied various methods to measure and decompose the inequalities. The Theil L (θ_{L}) and Theil T (θ_{T}) indices were calculated to quantify the overall inequality of the distributions over the 16 year time period. These measures were applied to highlight different aspects of the same distribution with respect to the annual population size of the regions. The θ_{L} (mean logarithmic deviation) is more sensitive to changes at the lower tail of a distribution, while θ_{T} is more sensitive to changes at the upper tail.35 36 The θ_{L} and θ_{T} can be calculated using the following mathematical expressions37–39:
Where n represents the number of regions, P_{i} is the population proportion of the i ^{th} region, and Y_{i} is the proportion of a given distribution of the i ^{th} region.
Despite the perfect decomposition of the Theil index into between and within region inequality components, this technique hampers the reranking effect on the overall inequality of a distribution. Also, the assumption of symmetric distribution with equal variance40 was easily violated in our case because the regions are heterogenous and more likely to have differences in the distributions.
Furthermore, we calculated the Gini index (GI) which is one of the most commonly used measures of distributional inequalities in healthcare with respect to populations.41–44 GI is sensitive to differences in distributions about the middle,45 insensitive to outliers and has a neat relationship with the Lorenz curve.46 Thus GI can be algebraically described as twice the area between the Lorenz curve and the 45 degree line of equality,38 and can be calculated using the following mathematical equation47:
μ: where μ is the mean value of the overall distribution, n is the number of regions, is the value of a distribution in the i ^{th} region, is a region’s population share and is the relative rank of the i ^{th} region.
Despite the claim that the GI allows direct comparison of inequalities between units with different population sizes,48 the small number of regions (n=11), the wide difference in population size among the regions and the direct association between population size and GI may lead to biased results—that is, the comparison of Gini inequalities among the regions and the inequality changes over time can lead to bias. Therefore, we considered the simple first order bias correction term due to a small sample proposed by Deltas, which is expressed as follows49:
where represents the adjusted GI for small sample and n is the number of regions. Accordingly, the GIs for the regions in our study would be underestimated by about 10%. The values of the GI range from 0 (absolute equality) to 1 (absolute inequality).50 The extent of the Gini inequality was judged based on the five scale values categorised as absolute equality (GI <0.2), high equality (GI=0.2–0.3), inequality (GI=0.3– 0.4), high inequality (GI=0.4–0.6] and absolute inequality (GI >0.6).51 A distribution with a Gini value above 0.5 can also be considered polarised.39 This scale was used only to create simplicity of interpretation because the extent of inequality is context specific and can be judged differently.
We further applied different Gini decomposition techniques. First, Pyatt’s52 overall Gini decomposition technique was applied to quantify the extent of the net between region Gini ( ), within region Gini ( ) and the interaction pseudoGini ( ) inequality components of each distribution. The sum of these components provides the overall Gini of a distribution. This modelling approach utilises the values of observations >0. The interaction term (transvariation, an overlap or crossover term) is a reranking effect that occurs when the highest distribution in one region overlaps with the lowest distribution of the same variable in another region.52–55 This method avoids the ambiguity that might arise from the interaction term in the Theil’s index of inequality decomposition52 56 and is more appealing to devise appropriate measures for reducing inequalities.54
Second, we calculated the extent of the overall inequality change between the baseline and end of study period using the Jenkins and Van Kerms’57 decomposition of inequality changes at two points in time. Third, the Shapley postestimation statistics was done after running the logarithm of the overall GHE (logGHE) per capita regression model on the explanatory variables (healthcare resources) to estimate their relative share of the overall GHE per capita.58 This method uses the r^{2} value of the regression model to precisely quantify the estimates by handling the problem that could arise from the residual. The estimates additively yield the overall GHE and point those variables that require explanations.59–61 Finally, we applied the multidimensional decomposition of the overall GHE inequality (G_{GHE}) by the explanatory variables using the Lerman and Yitzhaki62 method of decomposition as follows:
where K is a healthcare resource variable, which ranges from k=0, ……K, R_{k} is the Gini correlation of the ranked explanatory variable with the overall GHE inequality, G_{k} is the Gini of the explanatory variable and S_{k} is the GHE share of the explanatory variable.
This technique incorporates the concept of concentration index and was used to quantify the relative marginal change, the relative GHE inequality and the Gini elasticity of the explanatory variables with respect to the marginal change in the mean GHE and populations of the regions over time. The relative GHE inequalities and the elasticity values of the explanatory variables were calculated manually. These measures enabled us to explain the Gini of inequalities.63 64 We used the bootstrap and Jackknife techniques as appropriate to determine 95% CI for the indices.65 66 All analyses were performed using Stata Statistical Software Release 14 (StataCorp LP, College Station, Texas, USA). The interpretations and discussions of the findings were based on the Gini values.
Results
The Theil and Gini indices consistently revealed high overall inequalities in the GHE per capita, healthcare workforce, healthcare infrastructure and in some of the healthcare outcome indicators from 2000 to 2015 in Ethiopia. The aggregate inequality values of at least two of the three indices of all the indicators except for the SPDs and OPD visit per capita had intersection points in common. Despite considerable inequality reductions between the baseline and end of the study period, the GIs of many of the indicators remained strongly correlated with the overall GHE per capita inequality. The marginal increase in the mean GHE per capita resulted in a relative marginal increase in the distributions in favour of the advantaged populations of the regions. The elasticity value of <1 for all the indicators suggests the shortages. The net between region inequality dominantly explained the overall inequality of each distribution, and the interaction term was >0.
Inequality in GHE
The average overall GI for the GHE per capita was 0.596 (95% CI 0.544 to 0.648), and marked reduction in the inequality was observed after 2006 (figure 2A). The overall GI for the regions ranged from 0.317 (95% CI 0.252 to 0.381) for HA to 0.624 (95% CI 0.526 to 0.722) for SNNP. The net interregion inequality and the interaction term accounted for 54.4% and 37.9% of the overall GHE per capita inequality, respectively (table 1).
Inequality in healthcare workforce
The overall GI ranged from 0.428 (95% CI 0.393 to 0.463) for NMWs to 0.682 (95% CI 0.646 to 0.718) for the SPDs (table 1). Including zero values in the analysis of the time series observations, the GI for the SPDs was 0.704 (95% CI 0.652 to 0.756). The net interregion inequality ranged from 53.8% for the HOs to 95.0% for the SPDs, and the interaction term was the lowest for the SPDs (1.6%) and the highest for the HOs (38.7%). Over the 16 year time period, AM (GI=0.387, 95% CI 0.281 to 0.493) and OR (GI=0.319, 95% CI 0.193 to 0.444) among the agrarian regions and BG (GI=0.368, 95% CI 0.248 to 0.487), GA (GI=0.356, 95% CI 0.233 to 0.478) and AF (GI=0.323, 95% CI 0.223 to 0.424) among the pastoral/semipastoral regions showed inequality in the SPDs. The inequality in the distribution of the HOs was common to all regions, while SO (GI=0.638; 95% CI 0.435 to 0.841) and AA (GI=0.633, 95% CI 0.494 to 0.771) were the regions with absolute inequality. There were inequalities in NMWs and SKHPs in SO and in all the agrarian regions, except for TG. PHARP in GA, and MLABP in BG and HA regions were fairly equally distributed. GA and all the urban/urban dominated regions had inequality in the distribution of the EVT’THP. Including zero values in the analysis, the overall GI for PHARP and EVT’THP were 0.541 (95% CI 0.492 to 0.590) and 0.467 (95% CI 0.402 to 0.531), respectively. The magnitude and trend of the overall inequalities in the distributions of the GHE and the healthcare workforce at the national level is indicated in figure 2A,B.
Inequality in healthcare infrastructure
The overall GI for the HC, PHP and HPB distributions, and the net between region inequality of the same distributions accounted for 0.409 (95% CI 0.380 to 0.439), 0.460 (95% CI 0.404 to 0.517) and 0.592 (95% CI 0.563 to 0.621), and 44.7%, 94.0% and 92.1% of the overall inequality of each distribution, respectively (table 2). The interaction term was the highest for the HC distributions (47.7%). The overall inequality trend for the healthcare infrastructure is illustrated in figure 2C. BG (GI=0.223, 95% CI 0.168 to 0.278), HA (GI=0.290, 95% CI 0.244 to 0.336) and AA (GI=0.242, 95% CI 0.150 to 0.335) regions had equally distributed HCs. All the regions had GIs for the PHP and HPB distributions <0.3 except for the GI for HPBs in the SO (GI=0.314, 95% CI −0.120 to 0.748) and SNNP (GI=0.598, 95% CI −0.185 to 1.380) regions, which were insignificantly high. The regional disparities in the distributions of selected healthcare resources is graphically illustrated in figure 3. For example, the bottom 50%, middle 40% and the top 10% of the populations in TG had access to 31.2%, 50.5% and 18.3% of the total NMWs in the region, respectively.
Inequality in healthcare outcomes
The overall GIs for hospital OPD visits per capita (GI=0.349, 95% CI 0.321 to 0.377) and FIMM children (GI=0.307, 95 CI 0.269 to 0.345) revealed inequalities. The net between region inequality of the same indicators accounted for 75.1% and 59.0% of the overall inequality, respectively (table 2). We observed a continuous reduction in the overall inequality in FIMM children throughout the 16 year time period (figure 2C). The GIs for OPD visits per capita in AF (GI=0.341, 95% CI 0.249 to 0.432) and GA (GI=0.427, 95% CI 0.303 to 0.550), and for FIMM children in the regions SO (GI=0.524, 95% CI 0.384 to 0.665), AF (GI=0.393, 95% CI 0.219 to 0.566) and BG (GI=0.311, 95% CI 0.217 to 0.405) were higher than the GIs in the other regions. The net between region inequality of U5MR and IMR accounted for 47.1% and 49.5% of the overall inequality of each indicator, respectively.
The overall Gini inequality changes between baseline and the end of the study period for GHE per capita indicated a 31.2% reduction in inequality (P=0.030). The GIs for GPs, APHYs, HOs, NMWs, SKHPs and MLABP were each reduced by more than a third (P<0.01) (table 3). The reductions in GIs for SPD, PHARP and ENV’THP distributions were insignificant (P>0.05). The GIs for the overall distributions of HCs and the coverage of FIMM children were reduced by 60.3% and 63.8% (P<0.001), respectively. The inequalities in U5MR increased by 64.5% (P=0.048).
The estimated Shapley value indicated that the relative share of the healthcare workforce and healthcare infrastructure distributions were 46.5% and 53.5% of the overall GHE, respectively. HOs (18.09%), NMWs (17.20%), MLABP (10.65%), PHARP (9.58%) and HCs (32.32%) had higher relative shares of GHE (table 4). Column 3 of table 5 shows the strong correlation between these variables and the overall inequality in GHE per capita. Columns 5 and 7 indicate the relative GHE inequality and elasticity values of the explanatory variables relative to a marginal change in the average GHE per capita. The elasticity values of all the variables were <1.
The highest and lowest relative GHE inequality and Gini elasticity were for HOs (0.8525 vs 0.8513) and environmental health personnel (0.3750 vs 0.3612), respectively. The marginal increase in the average GHE resulted in a marginal increase in the distributions of all the explanatory variables (column 6) towards the privileged people among the regions (the negative sign indicated the concentration of the indicator among the advantaged populations). Ceteris paribus, a 1% increase in the GHE per capita, resulted in a 0.0027% increase in the distribution of the GPs towards the most affluent people among the regions.
Discussion
This study analysed the trend and degree of inequalities in the distributions of selected healthcare resources and outcomes from 2000 to 2015 in Ethiopia. Our findings revealed a high degree of overall inequalities in most of the distributions included in our analysis. The net between region inequality in GHE accounted for 54.4% of the average overall GHE inequality in Ethiopia. Although better economic position of a country and sufficient allocation of GHE can positively influence health system outcomes,67 decentralisation in under resourced countries may lead to more interregional inequalities.68 Nevertheless, evidence from high income countries with decentralised governance revealed no increased interregional inequalities in health expenditure per capita.69 The high overall GHE inequality observed in our study could be due to multiple factors, including the regional differences in prioritising health, development priorities or compliance to the national health policies.70 Like the evidence from other studies, the small allocation of GHE, the difference in the roles of the regional governments with respect to spending on health,71 differences in economic performances of the regions and the weak balancing mechanisms34 72 could have contributed to the GHE inequalities shown in our findings.
The significant reduction in the GHE inequality observed in our study could be related to an increased fair allocation of the GHE due to the progressive improvement of the national GDP during the past decade.73 Also, the reduction in GHE inequality might imply the increased regional governments’ commitments to spending more on health,74–76 improvements in governance77 78 or increased commitment of the development partners to health. Nevertheless, the overall GHE inequality remained too high and sufficient enough to contribute to the regional inequalities in healthcare. The strong correlation between the overall GHE inequality and the inequalities in some of the healthcare resources may indicate the prioritisation of healthcare resources by national and regional governments, whose success depends on the rational distribution of healthcare resources.79 Evidence from a study in South Africa showed that regions with a better capacity to use the health budget also had better opportunities to receive, allocate and spend more.80
A region with a high density of fairly distributed healthcare workforce is more likely to serve the healthcare needs of its people than a region with a low workforce density.35 The high net between region inequalities in the healthcare workforce observed in our findings might imply the shortage and maldistribution of the limited available healthcare workforce. This was reflected by the relative marginal increase in the distributions of the healthcare resources towards the advantaged populations among the regions and the elasticity values of <1 in all the healthcare resources analysed. The elasticity values of <1 for all the healthcare resources in our study, indicate that the resources are yet necessary inputs to ensure access to healthcare in the regions.34 One study in China also reported prorich concentration of the health workforce,81 while others from a developed country reported the reductions in inequalities among regions following a decentralised system.82
The overall inequalities in GPs and NMWs in our study were more than threefold (GI=0.591 vs GI=0.191) and about twofold (GI=0.428 vs GI=0.267), respectively, compared with the inequalities reported from China.83 The absolute inequality in the distribution of SPDs across the years observed in our study coupled with the prolonged time required to train and produce the qualified GPs and SPDs, plus the turnover of the SPDs for different reasons in Ethiopia, ranging from 21.4% in DD to 43.3% in the AM region,84 could make the inequality reduction more challenging. The overall inequality in APHY distribution in our study (GI=0.612 vs GI=0.532) was slightly higher than that reported from a study in Fiji48 and extremely higher than the reports from Japan41 85 and Mongolia.42 The relatively homogenous but small Gini values for APHYs across the regions may imply the hidden inequity phenomenon.15
The overall healthcare infrastructure distributions (HCs, PHPs, and HPBs) were also highly unequal compared with the findings reported from China.86 Nevertheless, the continuously marked reduction in the inequality of the HC distributions and the twothirds (60.3%) reduction in inequality indicate the efforts of the central and regional governments in improving access to primary healthcare among rural residents.87 In contrast, the increasing tendency in the overall inequality of the PHPs may also imply an increasing inequality in the healthcare workforce, especially among GPs and SPDs. Similarly, others reported a wide disparity in the geographic distribution of healthcare workforce, health facilities and hospital beds.86 The high inequalities in the distributions of the healthcare infrastructure in our study might inform the regional differences in development, management, institutional capacity, priorities and strategies followed by the regions to meet the healthcare needs of the local people.
The high overall inequalities in the distributions of the healthcare resources in the context of Ethiopia, where most people live in rural areas, can challenge the uniform achievement of the health sector goals in the country. As Atkinson stated, "a smaller cake more fairly distributed may be preferable to a bigger one with high levels of inequality". 16 Thus the high inequality in the GHE across the regions might challenge the reduction of inequalities for other healthcare resources. It is pivotal to minimise the existing inequalities among the regions, yet success may depend on the government’s commitment to improve financing, allocation of healthcare resources in addition to identification of other context specific opportunities and barriers to reduce the inequalities.
The economic, social, political and infrastructural positions of a country or a region are fundamental to determine the success in healthcare outcomes.67 The expansion of the PHPs in Ethiopia are believed to have created an opportunity for people to use hospital services.87 However, the net between region inequality in OPD visits per capita was considerably high (75.1%), and the reduction in inequality over time was insignificant. Over a 16 year period, two of the four pastoral/semipastoral regions (AF and GA) had high inequality in OPD visits per capita. In Ethiopia, this inequality might be explained by low access to the hospital services among the majority of rural residents, the shortage of qualified providers, the weak referral linkage between the primary healthcare units and PHPs and other individual factors, such as low awareness of the benefits of hospital services and financial constraints. A recent study in Ethiopia also reported the practice of daily or weekly rationing of laboratory tests, hospital drugs prescriptions, radiological investigations and the provision of the second best treatment,88 all of which might contribute to the low utilisation of hospital services. People living in remote areas, those who were poor, and of ethnic minorities, were also reported to have a low hospital services use.7 79
Despite the marked reduction in the overall inequalities in FIMM children (63.8%) over the study period, the U5MR presented a significant inequality. Microlevel studies in Ethiopia identified the association between hospital inaccessibility and death from a vaccine preventable disease (measles)89 as well as the association between HC inaccessibility and a high child mortality.90 Others also reported an increasing tendency in socioeconomic inequality in neonatal and under 5 child mortality rates.21 Due to the possible occurrence of unavoidable contextual and individual differences, such as the biological factors, it would be difficult to further reduce the small degrees of overall interregional inequalities in the U5MR and IMR.91
Strengths and limitations of this study
This study provided a comprehensive understanding of the extent and trend of inequalities in selected healthcare resources and outcomes over a 16 year period in Ethiopia. Application of the different econometric models helped us describe the inequalities in the decentralised system. The comprehensive use of data from the annually published Health and Health Related Indicators bulletin of the Federal Ministry of Health of Ethiopia, the CSA of Ethiopia census based population estimates for the regions and data from the 5 yearly EHDS reports could have contributed to the reliability and credibility of the findings. The analysis was based on a small number of regions, which could lead to a downward biased GIs for the regions with small populations. Despite this, we observed relatively higher inequalities for some of the indicators among the regions with smaller population sizes; the Gini values of the regions generated by Pyatt’s Gini decomposition technique might be underestimated. We also acknowledge that the decomposition of the inequality change in two points in time cannot provide information about the whole story of the inequality dynamics over time. Also, the analysis was based on aggregate level data that emphasised the supply perspective of the health system.
Policy implications
The success of the healthcare system as a whole and the SDGs for health in the context of Ethiopia require a continuous and coordinated effort to further reduce the observed inequalities in healthcare access. This study highlighted not only the magnitude and trend of the inequalities in the distributions of the healthcare resources, but also an overall shortage of the healthcare resources. The relatively higher GHE share of the healthcare infrastructure also shows the regional and central governments' greater emphasis on the expansion of healthcare facilities, rather than meeting the healthcare workforce standards of the facilities. This situation calls for a more coordinated effort to meet the healthcare needs of all people across all regions, especially those living in the pastoral and semipastoral areas. The healthcare resource gaps identified in this study imply the need for:
Implementing healthcare standards and certifying the health facilities on the fulfilment of predetermined minimum healthcare workforce and material requirements, and strengthening mechanisms for monitoring inequalities in basic healthcare access in all regions.
Building the institutional capacity of the regional health bureaus to closely track and address inequalities in distributions within each region.
Introducing mechanisms to raise sufficient health budget without increasing the burden of cost on poor citizens.
Conclusion
Despite the progressive reductions in inequalities, there are still significant inequalities in the distributions of some healthcare resources and outcomes in Ethiopia. The small GHE per capita coupled with high inequalities make the situation more challenging. Similar to most member states, Ethiopia has committed to achieving SDGs by 2030. Unless Ethiopia significantly scales up its efforts to increase the GHE per capita and implement inequality reduction mechanisms, the proportional progress towards achieving the health related SDGs across the regions may become less feasible. Further investigation of context specific barriers to more equitable access in healthcare and their root causes is of paramount importance to contribute to the inequality reduction in healthcare in Ethiopia, and perhaps similar low income settings.
References
 1.↵
 2.↵
 3.↵
 4.↵
 5.↵
 6.↵
 7.↵
 8.↵
 9.↵
 10.↵
 11.↵
 12.↵
 13.↵
 14.↵
 15.↵
 16.↵
 17.↵
 18.↵
 19.↵
 20.↵
 21.↵
 22.↵
 23.↵
 24.↵
 25.↵
 26.↵
 27.↵
 28.↵
 29.↵
 30.↵
 31.↵
 32.↵
 33.↵
 34.↵
 35.↵
 36.↵
 37.↵
 38.↵
 39.↵
 40.↵
 41.↵
 42.↵
 43.↵
 44.↵
 45.↵
 46.↵
 47.↵
 48.↵
 49.↵
 50.↵
 51.↵
 52.↵
 53.↵
 54.↵
 55.↵
 56.↵
 57.↵
 58.↵
 59.↵
 60.↵
 61.↵
 62.↵
 63.↵
 64.↵
 65.↵
 66.↵
 67.↵
 68.↵
 69.↵
 70.↵
 71.↵
 72.↵
 73.↵
 74.↵
 75.↵
 76.↵
 77.↵
 78.↵
 79.↵
 80.↵
 81.↵
 82.↵
 83.↵
 84.↵
 85.↵
 86.↵
 87.↵
 88.↵
 89.↵
 90.↵
 91.↵
Footnotes
Patient consent for publication Not required.
Contributors AW contributed in developing the concept, analysing the data and interpreting the findings, and in the write up of the manuscript. AT contributed from the beginning in the development of the concept, in shaping the entire work and in intellectual development and continuous review of the manuscript. He is the guarantor. AAS participated in developing and refining the methodology, and revising the manuscript. AO participated in developing the methodology and interpretation of the findings.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or notforprofit sectors.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement The study was based on publicly available data. The retrieved datasets are available from the corresponding author on reasonable request.