Are the poorest poor being left behind? Estimating global inequalities in reproductive, maternal, newborn and child health

Introduction Wealth-related inequalities in reproductive, maternal, neonatal and child health have been widely studied by dividing the population into quintiles. We present a comprehensive analysis of wealth inequalities for the composite coverage index (CCI) using national health surveys carried out since 2010, using wealth deciles and absolute income estimates as stratification variables, and show how these new approaches expand on traditional equity analyses. Methods 83 low-income and middle-income countries were studied. The CCI is a combined measure of coverage with eight key reproductive, maternal, newborn and child health interventions. It was disaggregated by wealth deciles for visual inspection of inequalities, and the slope index of inequality (SII) was estimated. The correlation between coverage in the extreme deciles and SII was assessed. Finally, we used multilevel models to examine how the CCI varies according to the estimated absolute income for each wealth quintile in the surveys. Results The analyses of coverage by wealth deciles and by absolute income show that inequality is mostly driven by coverage among the poor, which is much more variable than coverage among the rich across countries. Regardless of national coverage, in 61 of the countries, the wealthiest decile achieved 70% or higher CCI coverage. Well-performing countries were particularly effective in achieving high coverage among the poor. In contrast, underperforming countries failed to reach the poorest, despite reaching the better-off. Conclusion There are huge inequalities between the richest and the poorest women and children in most countries. These inequalities are strongly driven by low coverage among the poorest given the wealthiest groups achieve high coverage irrespective of where they live, overcoming any barriers that are an impediment to others. Countries that ‘punched above their weight’ in coverage, given their level of absolute wealth, were those that best managed to reach their poorest women and children.

The standard error for the CCI was estimated through the bootstrapping resampling technique. A total of 50 samples with replacement were drawn from the whole sample and for each quantile and the standard error is based on the distribution of these samples' average.
The full definition of each indicator used in the CCI calculation, including numerator and denominator, is presented in the Table S1. Children with suspected pneumonia who were taken to a health service from an appropriate health institution or provider

Wealth index and absolute income
The wealth classification used in the analysis is based on an asset index created through principal components analysis and adjusted for urban and rural area of residence. The variables used to calculate the score include household assets (e.g. cookstove, bicycle, car), building materials of the house (e.g. wood floor, brick walls, corrugated roof) and access to utilities (e.g. sanitation, electricity). The continuous score is provided with the original survey datasets and calculated according to a standard methodology. [5,6] It was then split into five equally sized groups (quintiles) and into ten equally sized groups (deciles) at the household level.
The attribution of absolute income values to households was done according to the methods described in Fink et al. [7] In summary, an income distribution for each country is generated by using the consumption share in the country's GDP and the Gini coefficient to generate the parameters for a log-normal distribution. [8] This distribution is then used to simulate the income distribution of a large number of observations, which, in turn, are divided into quintiles, deciles or centiles. The average calculated for each quantile from this simulation exercise is then used as the income value for the households, in constant 2011 international dollars adjusted by purchasing power parity (PPP).

Expected coverage for a given income level
The estimation of the average CCI coverage for a given level of absolute income in international dollars was done through a linear multilevel model where the outcome was the CCI and the predictor was the log-transformed income. Quintiles were level 1 units and countries level 2 units in the multilevel model. The adjusted CCI coverage and its standard error was then estimated from the resulting model.

Inequality measures
Absolute inequality across the wealth distribution was measured using the slope index of inequality (SII). The SII was estimated through individual-level logistic regression with the intervention coverage (yes or no) as the outcome and the wealth quintiles as the predictor. The difference between the extremes of the wealth distribution is then estimated using the resulting model. When used for coverage indicators, the SII can vary from -100 to +100, where zero means absence of inequality. A positive value indicates that the richer groups present higher coverage than the poorer groups, while negative values mean the opposite. [8] Statistics package All analyses were carried out with Stata (StataCorp. 2017. Stata Statistical Software: Release 15. College Station, TX: StataCorp LLC). The individual level analyses (such as estimation of the CCI and components) accounted for the survey design by adjusting for the clustered nature of the sample, stratification and sample weights. Country-level (or quintile-level) analyses were not weighted by population size so that each country (or quintile) has the same weight in the analyses (e.g. correlation estimates).

Supplementary results
Supplementary results are presented in a supplementary Excel spreadsheet containing Table  S1, Table S2, Table S3 and Figure S1.