Assessing levels and trends of child health inequality in 88 developing countries: from 2000 to 2014

ABSTRACT Background: Reducing child mortality was one of the Millennium Development Goals. In the current Sustainable Development Goals era, achieving equity is prioritized as a major aim. Objective: This study aims to provide a comprehensive and updated picture of inequalities in child health intervention coverage and child health outcomes by wealth status, as well as their trends between 2000 and 2014. Methods: Using data from Demographic Health Surveys and Multiple Indicator Cluster Surveys, we adopted three measures of inequality, including one absolute inequality indicator and two relative inequality indicators, to estimate the level and trends of inequalities in three child health outcome variables and 17 intervention coverages in 88 developing countries. Results: While improvements in child health outcomes and coverage of interventions have been observed between 2000 and 2014, large inequalities remain. There was a high level of variation between countries’ progress toward reducing child health inequalities, with some countries significantly improving, some deteriorating, and some remaining statistically unchanged. Among child health interventions, the least equitable one was access to improved sanitation (The absolute difference in coverages between the richest quintile and the poorest quintile reached 49.5% [42.7, 56.2]), followed by access to improved water (34.1% [29.5, 38.6]), and skilled birth attendant (SBA) (34.1% [28.8, 39.4]). The most equitable intervention coverage was insecticide-treated bed net for children (1.0% [−3.9, 5.9]), followed by oral rehydration therapy for diarrhea ((8.0% [5.2, 10.8]), and vitamin A supplement (8.4% [5.1, 11.7]). These findings were robust to various inequality measurements. Conclusions: Although child health outcomes and coverage of interventions have improved largely over the study period for almost all wealth quintiles, insufficient progress was made in reducing child health inequalities between the poorest and richest wealth quintiles. Future efforts should focus on reaching the poorest children by increasing investments toward expanding the coverage of interventions in resource-limited settings.


Background
Reducing child mortality was one of the Millennium Development Goals (MDGs). Since 1990, worldwide under-five mortality has dropped from 91 deaths per 1,000 live births in 1990 to 43 in 2015 [1]. Despite this apparent reduction in the under-five mortality rate, the national-level aggregate data often obscure the fact that the poorest global populations have not experienced nearly such improvements [2]. It is estimated that children in the poorest households were three times more likely to die before the age of five than were those in the richest quintile. This trend remained constant in some countries, but worsened in the majority of them [3,4]. The Sustainable Development Goals (SDGs) prioritize improving equity as a major goal for the period 2015-2030 and emphasize reducing child health inequalities both within and between countries [5].
A large amount of the previous literature on child health equity in developing countries has focused on coverage of interventions in the health sector of selected countries such as Brazil, Vietnam, China, Thailand, and Tanzania [6-10]. These studies provided firm evidence that socioeconomic status is an essential factor associated with child health. Studies by Barros AJ et al., Victora CG et al., and the Countdown 2008 equity analysis group have investigated the levels and trends of socioeconomic inequality in maternal and child health interventions in the Countdown countries (a maximum of 54 Countdown countries) (Countdown countries-a group of highburden priority countries, which together represent 95% of maternal and child mortality [11]) between 1990 and 2008 [12][13][14]. These studies showed that, although there was noticeable improvement in child health interventions, inequalities persisted in many countries. Since 2010, the Countdown group has reported detailed equity profiles for each Countdown country [15].
Our study aims to assess the level and trends of child health inequalities by wealth status in 88 developing countries with available data from 2000 to 2014. Our study will expand upon previous analyses by widening the scope of the countries from 54 to 88 developing countries, including both Countdown and non-Countdown countries. The time frame is also extended from 2000-2014, with the most updated data. In addition, we include more indicators to measure child health inequalities than have previous studies. For example, we tracked inequalities in three child health outcome indicators (infant mortality, under-five mortality, and stunting). Previous studies analyzed inequality in coverage for 12 child health interventions [12,16-18]; we added an additional five interventions in our study, including access to improved water, access to improved sanitation, Bacillus Calmette-Guérin (BCG) immunization, polio immunization, and care seeking for diarrhea. To our knowledge, this is the first study that provides a comprehensive picture of the progress made in reducing inequalities in child health outcomes and across 17 child health interventions for 88 countries between 2000 and 2014.

Data sources and procedures
Our primary data sources are the Demographic Health Surveys (DHS) and Multiple Indicator Cluster Surveys (MICS) conducted between 2000 and 2014. The DHS and MICS data are highly comparable, as their technical teams have collaborated closely and work together through interagency processes to ensure that survey tools are harmonized [19]. We excluded interim DHS surveys due to their relatively small sample size and limited available indicators [20]. We identified 88 developing countries with available DHS or MICS data subsequent to 2000. Following previous studies [13,14], we divided the surveys into three rounds according to the date of Among the 88 countries, 41 had undertaken all three rounds of surveys, enabling us to analyze the trends of inequality in child health outcomes and interventions with available indicators. For countries with multiple years of data available within one survey round, we used data from the most recent year available, so as to avoid overweighting such countries in our analysis.
To avoid repeating previous work on measuring indicators by wealth quintiles, we utilized the WHO Global Health Observatory (GHO) database [21]. Using the DHS and MICS surveys, the GHO provides calculated data by wealth quintile for a group of indicators of child health outcomes and child health interventions, such as infant mortality, under-five mortality, and SBA. One issue in using the GHO database is that estimates are only available until 2012, and data from earlier years (2000)(2001)(2002)(2003)(2004) or after 2012 in some countries are unrecorded. Therefore, we manually estimated values of health indicators by wealth quintiles for these years following the method adopted by the WHO GHO [22]. Table A2 lists the data from GHO and estimates from our own calculations.

Selection of child health indicators
Indicators on child health outcomes We selected three child health outcome indicators, infant mortality, under-five mortality, and stunting prevalence. We chose these indicators for the following reasons: reducing under-five mortality (including infant mortality) is one of the health-related MDGs that have been the focus of global health agendas [23]. Stunting is a cause of public concern in many developing countries, and reducing stunting prevalence is one of the key components of improving child health. These three indicators have been commonly used in previous literatures for assessing the health status of children (Table A3) [7,9,24,25].

Indicators on child health interventions
We included 17 interventions that collectively account for all stages of the continuum of care for child health. Twelve of these were included in previous studies (Table A3) for their well-documented effects on child mortality and health [12,[16][17][18]. We added five additional interventions that have demonstrated a significant impact on child health and have available data in DHS and MICS: access to improved water, access to improved sanitation, BCG immunization, polio immunization, and care seeking for diarrhea [9,[26][27][28][29][30][31].
Notably, the wealth index variable in DHS and MICS, which categorizes households into five wealth quintiles, is calculated using ownership of selected assets, such as televisions and bicycles; materials used for housing construction, and types of water access and sanitation facilities. When analyzing the inequality of access to improved water or sanitation, we followed the DHS guideline on constructing wealth indexes and produced a new set of wealth indexes that do not include water sources and sanitation facilities [32,33].

Measurements of inequality
Following previous studies [12,13], we adopted the three most commonly used inequality measurements, including one indicator for absolute inequality (the difference of the indicators between the richest wealth quintile, 'Q5', and poorest wealth quintile, 'Q1') and two indicators for relative inequality (the ratio of Q5 to Q1 and concentration index).
For each indicator at the country level, we generated values for each wealth quintile (Q1 to Q5). We calculated the absolute differences between Q1 and Q5 to assess the indicator's absolute inequality status, and the ratio of Q5 to Q1 to assess the indicator's relative inequality status.
We generated concentration indexes from the concentration curves that plot the cumulative proportion of one variable against the cumulative proportion of the population ranked by wealth. Concentration indexes capture the extent to which health outcomes/ health interventions differ across individuals' ranks by wealth. A detailed calculation method for concentration indexes is referred to in the World Bank's instructions [34]. The concentration index is expressed on a scale ranging from −100 to 100, with zero representing perfect equality. If the health indicator is undesirable, such as infant mortality, a negative value of the concentration index means that ill health is more prevalent among the poor. The more negative a concentration index, the more inequality it suggests. If the health outcome is desirable, a positive value of concentration index indicates the health of the rich is better than the health of the poor. The more positive a concentration index, the more inequality it suggests. Calculations, features, and limitations of the three measurements are presented in Table A4.
To provide a more comprehensive picture of inequality, we also included the interquartile range (IQR), bottom decile, and top decile for the three inequality measures. We calculated the 95% confidence intervals (CIs) for Q1, Q5, and three inequality measures. The calculation methods of the 95% CIs are presented in the Appendix Method section.
This study focused on the absolute inequality between Q1 and Q5, as it represents the most intuitive among all inequality measurements. By showing the scope of the gaps, policymakers and the general public can easily understand by how much the poorest population lags behind the richest [35].

Inequality analysis
When conducting trend analysis of an indicator, we consider it essential to compare the same group of countries over time. In order to achieve it, we only included the countries with valid data on an indicator in all three survey rounds. For example, 34 countries had a variable indicating child stunting prevalence in all three rounds. When analyzing the inequality of child stunting prevalence over time, we only included these 34 countries in the analysis. Table A5 contains data availability for each indicator.
To determine whether the inequality status of an indicator at either aggregate or country level changed significantly over time, we compared the 95% CI of the indicator in each available round. If the 95% CIs of the indicator in the two rounds did not overlap, we concluded that there was a significant change over time, or vice versa. We are aware that there are cases when the 95% CIs do not overlap, yet the differences are still significant, so we also conducted a t-test for the nonoverlapping values. STATA 14 was used in analysis.

Results
Mean level of inequality status in child health outcomes and interventions for 88 countries, most recent years Table 1 presents the mean level of absolute inequality in three child health outcomes and nine selected child health interventions, using data from the most recent survey of each country. Full tables with 17 interventions are presented in Tables A6 and  A7. Except for the insecticide-treated bed net (ITN) for children, the difference between Q5 and Q1 for all indicators differed significantly from zero: a result smaller than zero for outcome indicators and a result larger than zero for intervention indicators. This suggests that children in the richest quintile had better health outcomes and higher coverage rates of interventions than did children in the poorest quintile across countries.  32.6]). The most equitable coverage of intervention was ITN for children (1.0% [−3.9, 5.9]), followed by oral rehydration therapy (ORT) for diarrhea (8.0% [5.2, 10.8]) and vitamin A supplementation (8.4% [5.1, 11.7]). These results remain consistent over all three inequality measurements. (Table A6) Trends of mean inequality status in child health outcomes and child health interventions: round 1 vs. round 3 Figure 1(a and b) shows the mean prevalence of the three child health outcomes and the coverage rates of nine selected child health interventions by wealth quintile in round 1 and round 3. A complete figure, with eight additional interventions, is presented in Appendix Figure 1. Figure 1(a) shows in each wealth quintile that the mean infant mortality rate and under-five mortality rate reduced substantially over time. For example, the under-five mortality of the poorest quintile decreased from 139.9 per 1,000 live births in round 1 to 97.4 per 1,000 live births in round 3. The mean difference between Q5 and Q1 also largely shrank; the difference in infant mortality decreased from 34.4 per 1,000 live births in round 1 to 22.4 in round 3; the difference in under-five morality decreased from 62.6 per 1,000 live births in round 1 to 44.4 in round 3. The mean stunting prevalence, however, experienced a much smaller reduction between round 1 and round 3. For example, the stunting prevalence for the poorest quintile was 42.8% in round 1 and was only slightly reduced to 41.4% in round 3. The mean difference between Q5 and Q1 also remained large and did not improve over time, moving from 20.4% in round 1 to 20.2% in round 3.
We observed that the coverage rates of all nine interventions increased between round 1 and round 3 in each wealth quintile for countries with available data in round 1 and round 3 (Figure 1(b)). Consistent with results presented in Table 1, we found that access to improved sanitation had the largest difference between Q5 and Q1, 54.7% in survey round 1 and 59.1% in survey round 3. Other indicators that rank highly for mean inequality include supply of improved water, SBA, and four or more antenatal care visits: the differences between Q5 and Q1 for these three indicators were 46.0%, 44.0%, and 31.9% respectively in round 1, and decreased to 34.9%, 38.1%, and 26.5% in round 3. Except for access to improved sanitation and care seeking for suspected pneumonia, seven out of nine selected indicators showed reductions in inequalities between round 1 and round 3.
Country-level change in inequality status in child health outcomes and child health interventions: round 1 vs. round 3 Figure 2 shows the changes in absolute inequalities in three indicators that have data available in all three rounds in more than 30 countries: stunting prevalence, access to improved sanitation, and SBA. Figure A2 further shows the change in three more indicators with data available in more than 30 countries, including access to improved water, coverage of full immunization, and ORT for diarrhea. Tables A8 and A9 present the inequality status of the indicators in each round for each country.
In Figure 2(a), we observed that among the 34 countries with valid data on child stunting status in all three rounds, five of them experienced a significant reduction in the difference between Q5 and Q1 during survey rounds 1 and 3, including Dominican Republic, Egypt, Ghana, Mongolia, and Peru (marked in orange in Figure 2(a)); three of them were Countdown countries. Two countries (Congo, Dem. Rep., and Lao PDR, marked in green in Figure 2(a)) experienced significantly enlarged inequality in stunting prevalence between rounds 1 and 3. Strikingly, the inequality in stunting prevalence between Q5 and Q1 in Laos increased from −12. Sixteen of the 18 countries were Countdown countries. Nine of them reduced the difference between Q5 and Q1 by more than 20 percentage points, including Dominican Republic, Ethiopia, Jordan, Moldova, Mongolia, Nepal, Peru, Suriname, and Vietnam. However, 14 countries experienced significant increases in the absolute difference, (marked in green in Figure 2(b)) with 10 of them representing Sub-Saharan African countries. Thirteen of them were Countdown countries. Figure 2(c) shows that, among the 39 countries with available data on delivery with SBA in all three rounds, 14 experienced a significant reduction in the difference between Q5 and Q1 between round 1 and round 3, including seven countries in Sub-Saharan African (Benin, Congo  difference between Q5 and Q1 by more than 20 percentage points, including Cambodia, Egypt, Ghana, Malawi, Peru, Rwanda, and Zambia. On the other hand, seven countries showed a significantly enlarged difference between Q5 and Q1 in SBA (marked in green in Figure 2(c)): four countries located in Sub-Saharan Africa (Cameroon, Ethiopia, Niger, and Uganda); two located in South Asia (Bangladesh and Nepal); and one in East Asia & Pacific (Lao PDR). All of them were Countdown countries. The estimates of relative inequality (Tables A10-A13) were basically aligned with the estimates of absolute inequality.

Discussion and conclusion
This study has two salient findings. Firstly, remarkable improvements in child health and coverage of interventions have been observed between 2000 and 2014 in both Q1 and Q5, yet large inequalities remain  in these indicators. Except for ITN, the mean differences between Q5 and Q1 for all other indicators were significantly different from zero in the most recent survey round.
Though analyzing determinants of inequality is beyond the scope of this study, we speculate that measuring the progress of reaching MDGs by focusing on national means could lead to less policy attention paid to inequalities [36]. A 2010 UNICEF report found that progress measured by national aggregates often conceals large and widening disparities in child health-despite apparent statistical success in reducing under-five mortality, inequalities between the poorest and the richest households grew by more than 10% [37]. As the Countdown equity analysis indicates, child health interventions tend to reach the wealthiest children first in the absence of policy instruments for addressing inequality [14]. The second salient finding is that the progress made in reducing child health inequalities differs greatly by country, with some countries making significant improvements between 2000 and 2014, some significantly deteriorating, and others remaining statistically unchanged. For example, we found that 18 countries significantly reduced the difference between Q5 and Q1 in access to improved sanitation over time, with nine of them reducing the difference by more than 20 percentage points, indicating a remarkable improvement in equality of access to improved sanitation. Meanwhile, we observed a significant growth in absolute difference in 14 countries, implying a deteriorating level of equality in access to improved sanitation.
Although significant reductions or increases in child health inequalities have occurred in both Countdown and non-Countdown countries, we found that the deterioration in inequality was heavily concentrated in Countdown countries. All countries experiencing deteriorating inequality in child health outcomes were Countdown countries. Among those countries with deteriorating inequality in the coverage of child health interventions, more than 90% were Countdown countries. The following may be plausible explanations for this situation: firstly, out-ofpocket medical payments for receiving care could be too high for the poorest households in Countdown countries, thus deterring their access to the related care. For example, in Ethiopia, delivery at health institutions costs $17-22, while home delivery with traditional birth attendants costs only $0.8. Consequently, poor families may choose not to use SBA during deliveries [38]. Secondly, travel costs and foregone earnings are important opportunity costs in consuming health services for the poorest households, especially those living in remote rural areas. In Uganda, for example, 55.4% of the poorest population (the majority of which reside in rural areas) considered the long distance to a health facility a barrier to accessing health services, compared with 17.2% in the richest population [39]. Thirdly, poor quality of care could prevent poor rural patients in countries such as Rwanda from using the services [40]. It is estimated that, in rural Bangladesh, the absentee rate for physicians is 40% at larger clinics and 74% at smaller subcenters with a single physician. The number of patients was found to be negatively correlated with absentee rate [41]. Other factors, such as lack of proper knowledge/education on illness, lack of awareness of exemption of or subsidy to medical care spending for the poor, and cultural and social norms could also contribute to these inequalities [42][43][44].
This paper provides two distinct contributions to the study of child health inequalities. Firstly, taking advantage of available data, we extended the previous analyses by expanding the scope of considered countries (88 countries with 57 Countdown and 31 non-Countdown countries) and time (between 2000 and 2014). Secondly, we expanded upon the inequalities studied by including three child health outcomes and five interventions, such as access to improved water and sanitation.
There are several limitations to this study. Firstly, the classification of wealth quintiles is country-specific and time-sensitive. The poorest quintile in an upper-middle-income country could be better off than the richer quintiles in a less developed country. As a country's economy grows, the definition of the poorest quintile may also change [12,32]. Secondly, due to the varying availability of variables in the DHS and MICS by country and year, we were only able to conduct a trend analysis for a subset of countries, which limited our knowledge on the progress of reducing inequality in all 88 countries.
Moving forward, updating assessments on child health inequality, both within and across countries, is essential as more data become available. More research needs to be conducted to identify the facilitators or barriers to reducing inequality in child health. For example, similar to Barros et al.'s findings [12], our study shows that interventions that had to be delivered at health institutions, particularly those that need access to secondary-level or tertiary-level care, such as SBA, were among the least equitable. Interventions that could be delivered at the community level, such as vitamin A supplements, tend toward greater equitability [45]. This may suggest that expanding community health programs could be positively linked to the reduction of inequality in intervention coverage. In addition, improving equality should be prioritized at the national and global health agendas. In the SDG era, it is critical to monitor progress in child health, to focus not only on population means but also on inequalities.

Disclosure statement
No potential conflict of interest was reported by the authors.

Funding information
This study is funded by NIH K01 HD071929.

Paper context
While previous studies on child health inequalities exclusively focused on interventions in the health sector, we yielded a comprehensive evaluation of child health, tracking inequalities in three outcome indicators (mortality and stunting) and 17 interventions including access to improved water and sanitation. We also extended the previous analyses by expanding the scope of the countries (88 countries), and time length (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014). We call for greater investments toward the poorest to reduce child health inequalities.  Appendix for paper 'Assessing levels and trends of inequality in child health outcomes and the coverage of child health interventions in developing countries: from 2000 to 2014'  Table A2. Wealth quintile data from WHO Global Health Observatory (GHO) and estimates from our own calculations.  However, in fact, the Q5 group improves at a faster rate than the Q1 group, which appears to suggest the Q5 group did better than the Q1 group, and these two groups should become less equal. This is counterintuitive.

References
• It will not reflect the overall status of a health indicator in a country. For example, the absolute inequality of a country with 90% the children in Q1 and 100% of the children in Q5 covered by polio vaccine is the same as that of a country with 5% of the children in Q1 and 15% of the children in Q5 covered. Yet intuitively, the latter one should be of higher concern.
(Continued ) Concentration index quantified the degree of socioeconomic-related inequality in a health variable, which incorporates information from all income groups instead of simply the poorest and the richest • It has higher data requirements than the other equality measurements. There are usually two ways to obtain concentration index. One approach is using grouped data, which requires data on the health indicator and the number of individuals for each income group. The second approach uses micro-data, which requires health status data and wealth score for each individual. This study adopted the first approach.
• It could be sensitive to the living standards measure, such as consumption, expenditure, and wealth index. We follow Wagstaff's study and use wealth index to measure living standard.

Ratio of Q5 to Q1
Ratio of Q5 to Q1 measures the rate of the health indicator between the most advantaged group and the least advantaged group Improvement in the ratio implies a faster relative rate of health improvement among disadvantaged groups        Countries with significantly enlarged absolute difference between Q1 and Q5 in round 3 comparing to round 1 are shown in bold. Countries with significantly enlarged absolute difference between Q1 and Q5 in round 3 comparing to round 1 are shown in bold. Countries whose ratio of Q5 to Q1 became significantly farther away from one between round 1 and round 3 are shown in bold. Analyzing health equity using household survey data: a guide to techniques and their implementation. World Bank Publications, 2007). Second, we treated each country as a subject to calculate the mean values. Then we generated standard errors for the point estimates and calculated 95% CI based on the point estimate and the standard errors. Table A7 is generated in the same way as Table A6, yet instead of treating each country as an identical subject, we weighted the countries with their population sizes.
Tables A8 and A9 present absolute differences between Q5 and Q1 with the 95% CIs at the country level. To generate the 95% CI of the absolute difference, we used the formula ðx 1 À x 2 Þ AE Z ð1Àμ=2Þ Ã ffiffiffiffiffiffiffiffiffiffiffiffi ffi . In this study, x 1 and x 2 represent the point estimates for Q5 and Q1 groups, σ 1 and σ 2 are the standard errors of the Q5 and Q1 groups, n 1 and n 2 are the observation numbers of the Q5 and Q1 groups.
Tables A10 and A11 present concentration indexes with the 95% CIs at the country level. For each country, we generated the 95% CI following exactly the World Bank instruction. Detailed information can be found below: http://siteresources.worldbank.org/INTPAH/Resources/ Publications/459843-1195594469249/HealthEquityCh8.pdf (Accessed on March 2nd, 2017) Tables A12 and A13 present ratios of Q5 to Q1 (Q5/Q1) with the 95% CIs at the country level. To generate the 95% CI, we used Taylor Expansion, using the formula x 1 =x 2 AE z ð1Àμ=2Þ Ã ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi . In this study, x 1 and x 2 represent the point estimates for Q5 and Q1 groups, σ 1 and σ 2 are the standard deviation of the Q5 and Q1 groups, n 1 and n 2 are the observation numbers of the Q5 and Q1 groups.