Between the poorest and richest households from 1990 and 2016, the absolute gap in under-5 mortality rate has narrowed significantly and the difference in aggregated under-5 mortality rate halved for all LMICs (excluding China). The absolute declines in under-5 mortality rate in the poorest households in all regions were more than a third higher than those in the richest households. The relative difference between the poorest and richest under-5 mortality rates, however, remained similar between 1990 and 2016, with children in the poorest quintile being twice as likely to die before their 5th birthday compared with those in the richest quintile. Similarly, the disparity in under-5 mortality rate across all quintiles decreased significantly on the absolute scale but remained approximately constant on the relative scale during 1990 and 2016.
We provided estimates and UIs for quintile-specific under-5 mortality rate in 137 LMICs based on a statistical model. Our model results confirmed the empirical patterns from previous studies12x12Minujin, A and Delamonica, E. Mind the gap! Widening child mortality disparities. J Hum Dev. 2003;
Crossref | Google ScholarSee all References,29x29Houweling, TA, Kunst, AE, Huisman, M, and Mackenbach, JP. Using relative and absolute measures for monitoring health inequalities: experiences from cross-national analyses on maternal and child health. Int J Equity Health. 2007;
Crossref | PubMed | Scopus (76) | Google ScholarSee all References that at high national-level under-5 mortality rate, the expected ratio of poor to rich under-5 mortality rate tends to be low. As the under-5 mortality rate at the national level decreases, the expected ratio tends to increase. The association confirms the inverse equity hypothesis32x32Victora, CG, Vaughan, JP, Barros, FC, Silva, AC, and Tomasi, E. Explaining trends in inequities: evidence from Brazilian child health studies. Lancet. 2000;
Summary | Full Text | Full Text PDF | PubMed | Scopus (362) | Google ScholarSee all References that small disparities are expected at high mortality because most of the population, including the richest households, have inadequate access to basic health care and services. The initial decrease in the national-level under-5 mortality rate is likely to be driven by a decrease in under-5 mortality rate among the richest households, who selectively benefit from improved access to resources.33x33Cutler, D, Deaton, A, and Lleras-Muney, A. The determinants of mortality. J Econ Perspect. 2006;
Crossref | Scopus (386) | Google ScholarSee all References Eventually, the poorer subpopulations catch up and when they do, experience faster reductions than the wealthiest subpopulations.
At the regional level, west and central Africa continued to have the highest quintile-specific under-5 mortality rate and one of the lowest ratios of under-5 mortality rate in the first quintile to under-5 mortality rate in the fifth quintile during 1990–2016. However, increasing relative disparities have been observed in the region since 1990, as indicated by a significant increase in the ratio of under-5 mortality rate in the poorest quintile to richest quintile. As the aggregated under-5 mortality rate for all quintiles combined in this region decreased from 198·7 (90% UI 192·7–205·2) deaths per 1000 livebirths in 1990 to 94·7 (83·4–110·3) deaths per 1000 livebirths in 2016,1x1United Nations Inter-agency Group for Child Mortality Estimation. Levels and trends in child mortality: Report 2017. UNICEF,
New York; 2017
Google ScholarSee all References this ratio increased significantly. Because under-5 mortality rate is still high in many countries in this region, our model findings on the association between national-level under-5 mortality rate and the ratio of under-5 mortality rate in the poorest quintile to richest quintile suggest that relative disparities will possibly continue to increase after 2016, as national-level under-5 mortality rates further decrease. Policy interventions with an equity focus, which reach the most disadvantaged and vulnerable children, might help to change these trends. Efforts are needed to reduce high mortality across quintiles as well as address the increasing relative disparities in west and central Africa.
In south Asia, the large disparities in absolute and relative scales were mainly driven by results from India because its population size is the largest among all countries in the region. India’s national-level under-5 mortality rate decreased from 125·8 (121·8–130·2) deaths per 1000 livebirths to 47·4 (38·8–47·3) deaths per 1000 livebirths between 1990 and 2016.1x1United Nations Inter-agency Group for Child Mortality Estimation. Levels and trends in child mortality: Report 2017. UNICEF,
New York; 2017
Google ScholarSee all References In our study, India was identified as a high disparity country on absolute and relative scales. A further breakdown by smaller age groups can help to better understand the persisting high disparity in under-5 mortality rate in India. A previous study34x34Jain, N, Singh, A, and Pathak, P. Infant and child mortality in India: trends in inequalities across economic groups. J Popul Res. 2013;
Crossref | Scopus (2) | Google ScholarSee all References showed that in India during 1992–2006, relative disparities in mortality between the first and third years of life increased, whereas the inequality of mortality in the first year of life decreased.
Several major improvements and advantages in the data processing and modelling approach were used in this study. We calculated under-5 mortality rate by wealth quintile with an equal number of births in each quintile. This procedure has the benefit of providing a stable estimate of under-5 mortality rate for the richest quintile, since more births fall inside this quintile than when the standard method is used. The approach differs from the conventional way of deriving quintiles using data from Demographic and Health Surveys23x23Rutstein, SO and Johnson, K. The DHS wealth index. ORC Macro,
Calverton, MD; 2004
Google ScholarSee all References,35x35Rutstein, SO. The DHS wealth index: approaches for rural and urban areas. Macro International Inc,
Calverton, MD; 2008
Google ScholarSee all References and Multiple Indicator Cluster Surveys, in which the number of household members are the same in each quintile. The statistical model incorporates the association between national-level under-5 mortality rate and expected third quintile-disparity ratios. The model did reasonably well in validation exercises in which data were left out at random and at the end of the observation period (appendix pp 9–10appendix pp 9–10). The results suggested that the model-based estimates were unbiased and UIs were conservative (containing the left-out observations more often than expected), hence suggesting that the approach worked well to construct estimates for country-years with missing information.
One of the main limitations of our study is due to the nature of the data used: we used household assets at the time of the survey as a proxy for household economic status. The household characteristics recorded in surveys only reflect the condition at the time of the interview, whereas the mortality data recorded a period before the survey was done. Additionally, although the set of assets and amenities were tailored in each survey to represent conditions in each country at a specific time, variation within each country might in some cases not be covered adequately. The principal component approach used to construct the wealth indices is also not guaranteed to accurately assign low scores to a country’s poorest households. This limitation might explain our finding that mortality rate in the poorest quintile is lower than mortality rate in richer quintiles for a subset of country-years, hence reflecting problems in the index rather than reflecting lower mortality rate among the country’s poorest populations. Finally, the fact that the wealth index is country specific implies that absolute country-period specific differences in economic status between the poorest and wealthiest quintiles vary.36x36Fink, G, Victora, CG, Harttgen, K, Vollmer, S, Vidaletti, LP, and Barros, AJ. Measuring socioeconomic inequalities with predicted absolute incomes rather than wealth quintiles: a comparative assessment using child stunting data from national surveys. Am J Public Health. 2017;
Crossref | PubMed | Scopus (0) | Google ScholarSee all References This limitation is not restricted to the analysis of disparities based on wealth indices—an income-based or consumption-based relative index would face similar problems because of different consumption patterns and prices within and between countries as well as over time. If interest lies in the estimation of cross-country differences in mortality rate associated with absolute differences in wealth, measures such as a proposed predicted absolute income measure based on households’ asset rank, national consumption, and inequality levels36x36Fink, G, Victora, CG, Harttgen, K, Vollmer, S, Vidaletti, LP, and Barros, AJ. Measuring socioeconomic inequalities with predicted absolute incomes rather than wealth quintiles: a comparative assessment using child stunting data from national surveys. Am J Public Health. 2017;
Crossref | PubMed | Scopus (0) | Google ScholarSee all References can be used. However, any analysis based on absolute differences would not provide a standardised assessment of relative within-country disparities as done in this study.
The second main limitation in our study is data availability: we did not have data for 38 of 137 countries, and data at lower under-5 mortality rates and for more recent years are scarce. The absence of data for 38 countries results in model-driven estimates for those countries. The disparity pattern in these countries might differ from what the model suggested. For this reason, we did not present the country-specific estimates of under-5 mortality rate by wealth quintile for the 38 countries without any empirical data. We presented aggregated results based on all 137 countries, as opposed to results based on the 99 counties with data, to communicate our best estimates and related uncertainty on all LMICs (excluding China). The aggregated results are mainly driven by the 99 countries with available data as they accounted for 97% of all under-5 deaths in the 137 LMICs during 1990–2016. A comparison of the aggregated results based on the 99 countries with empirical data and the results based on the 137 LMICs is given in the appendix (pp 57–65)appendix (pp 57–65). This comparison shows that the overall and regional ratios of quintile-specific to national-level under-5 mortality rate based on the 137 and 99 countries are approximately the same across quintiles over time. The aggregated quintile-specific under-5 mortality rates based on the 137 countries are slightly lower than those based on the 99 countries with empirical data, since countries without data tend to be countries with lower national-level under-5 mortality rates than those countries with data.
Because data are scarce on low levels of national under-5 mortality rate (<20 deaths per 1000 livebirths), estimates for the country-years corresponding to those levels were more uncertain and largely based on model extrapolation. Data for countries without information on disparities at low mortality are needed to assess the country-specific situations. Finally, most of the countries with data only have a small number of datapoints. Data are also limited in the most recent period; this study only contains 41 datapoints from 38 countries with reference year from 2010 onward. Extrapolations using past trends were used to derive trends in the most recent years. Efforts are needed to collect reliable, disaggregated, and timely data to better understand trends in mortality disparities.
Despite data limitations, validation exercises (appendix pp 9–10appendix pp 9–10) suggest that our model-based estimates provide valuable information past the most recent datapoint. Point estimates are expected to be unbiased (we expect that the median difference between future observations and current estimates is equal to zero) and UIs are generally expected to be wide enough to convey the uncertainty of the estimates. In those countries where the model projections differ most from the truth (which will become clear with future data collection), we expect that future observations are likely to be less than the UIs. In those countries, observations will indicate less disparity than suggested by the model projections. This validation result suggests that use of the model-based projections presented in this study will not result in the undesirable situation whereby an underestimation of disparities results in lack of action to try to improve disparities.
In our study, we did not incorporate quintile-specific adjustments to reduce the bias associated with retrospective data in countries with high prevalence of HIV. Instead, we assumed that the observed ratios of quintile-specific to national-level under-5 mortality rates provide unbiased information of the true ratios. This assumption might result in the underestimation of the relative burden of HIV/AIDS-related deaths in children in the poorest quintiles. Additionally, we were not able to consider potential variation of reporting errors across quintiles because of the scarcity of information on the quintile-specific occurrence of such errors.
Our study provides a systematic assessment of under-5 mortality rate by wealth quintile for all LMICs (excluding China) and highlighted that the relative gap in child survival between the poorest and richest populations has remained constant during 1990 and 2016. Policy makers should not only acknowledge the progress made in child survival for the poorest subnational population across LMICs (excluding China), but also address the continued existence of within-country disparities and call for greater action to truly close the gap. Identification of patterns of inequity in under-5 mortality rate in countries is crucial for programming and planning.