Child mortality estimation 2013: an overview of updates in estimation methods by the United Nations Inter-agency Group for Child Mortality Estimation

Abstract

Background: In September 2013, the United Nations Inter-agency Group for Child Mortality Estimation (UN IGME) published an update of the estimates of the under-five mortality rate (U5MR) and under-five deaths for all countries. Compared to the UN IGME estimates published in 2012, updated data inputs and a new method for estimating the U5MR were used.

Methods: We summarize the new U5MR estimation method, which is a Bayesian B-spline Bias-reduction model, and highlight differences with the previously used method. Differences in UN IGME U5MR estimates as published in 2012 and those published in 2013 are presented and decomposed into differences due to the updated database and differences due to the new estimation method to explain and motivate changes in estimates.

Findings: Compared to the previously used method, the new UN IGME estimation method is based on a different trend fitting method that can track (recent) changes in U5MR more closely. The new method provides U5MR estimates that account for data quality issues. Resulting differences in U5MR point estimates between the UN IGME 2012 and 2013 publications are small for the majority of countries but greater than 10 deaths per 1,000 live births for 33 countries in 2011 and 19 countries in 1990. These differences can be explained by the updated database used, the curve fitting method as well as accounting for data quality issues. Changes in the number of deaths were less than 10% on the global level and for the majority of MDG regions.

Conclusions: The 2013 UN IGME estimates provide the most recent assessment of levels and trends in U5MR based on all available data and an improved estimation method that allows for closer-to-real-time monitoring of changes in the U5MR and takes account of data quality issues.

Figure 1. Illustration of the B-spline regression model for Norway.

From left to right: B-splines and their corresponding spline coefficients (plotted in the same color), observed log(U5MR) and U5MR (black dots) plotted against time, together with the spline estimates (red line). The spline estimate for log(U5MR) in each year is the sum of the non-zero B-splines in that year weighted by their respective spline coefficients.

Figure 2. UN IGME 2013 and UN IGME 2012 estimates of the U5MR for the years 1990 (left) and 2011 (right).

UN IGME 2013 estimates are plotted against UN IGME 2012 estimates. Gray areas represent relative differences of up to 10%, 20% and 30% respectively. Country-specific U5MR estimates are displayed as green points, or highlighted in red if the estimates differ by more than ten deaths per 1,000 live births. Regions are colored according to the given legend.

Figure 3. UN IGME 2013 and UN IGME 2012 estimates of the annual rate of reduction for 1990–2011.

UN IGME 2013 estimates are plotted against UN IGME 2012 estimates. Gray areas represent absolute differences of up to 1%, 2% and 3% respectively (absolute difference). Country-specific ARR estimates are plotted in green for high mortality countries (with U5MR in 1990 of at least 40 deaths per 1,000 live births), or in red for a subset of these if the difference is at least 2% and the UN IGME 2013 and UN IGME 2012 estimates disagree with respect to whether the country is on track to meet MDG 4 (4.4% annual rate of reduction). Regions are colored according to the given legend.

Figure 4. Decomposition of differences in U5MR for 1990 and 2011 into differences due to estimation method and differences due to data. The gray box represents differences up to 10 deaths per 1,000 live births.

Countries with differences of more than 10 deaths per 1,000 deaths due to either factor are highlighted in red.

Figure 5. Comparison of U5MR estimates for Lao PDR and Burkina Faso where the change in database changed the estimate by more than 10 deaths per 1,000 live births.

Estimates compared are from UN IGME 2013 (solid red line with 90% credible intervals given by the shaded regions), default Loess fit to 2013 database (dashed dark blue line) and UN IGME 2012 (solid dark blue line). Connected dots denote data from the UN IGME 2013 database and gray shaded areas around series of observations represent the sampling variability in the series (quantified by two times of the sampling standard errors). The newly-added/updated series for Lao PDR and Burkina Faso are those shown in dark blue. Excluded data series and detailed information on all data series are displayed in Figure S1.

Figure 6. Decomposition of differences in U5MR for 1990 and 2011 into differences due to data quality model and differences due to splines model.

The gray box represents differences up to 10 deaths per 1,000 live births. Countries with differences of more than 10 deaths per 1,000 deaths due to either factor are highlighted in red.

Figure 7. Comparison of U5MR estimates for Afghanistan, Angola, Botswana, Burundi, Central African Republic and South Sudan, where the inclusion of the data quality model changed the estimate by more than 10 deaths per 1,000 live births.

Estimates compared are from UN IGME 2013 (solid red line with 90% credible intervals given by the shaded regions), B2 fit to 2013 database (solid light green line with 90% credible intervals given by the shaded regions), default Loess fit to 2013 database (dashed dark blue line). Connected dots denote data from the UN IGME 2013 database and gray shaded areas around series of observations represent the sampling variability in the series (quantified by two times of the sampling standard errors). Excluded data series and detailed information on all data series are displayed in Figure S1.

Figure 8. Comparison of U5MR estimates for Algeria, Maldives, Oman and Pakistan, where the inclusion of the data quality model resulted in estimates that are closer to VR data.

Estimates compared are from UN IGME 2013 (solid red line with 90% credible intervals given by the shaded regions), B2 fit to 2013 database (solid light green line with 90% credible intervals given by the shaded regions), default Loess fit to 2013 database (dashed dark blue line). Connected dots denote data from the UN IGME 2013 database and gray shaded areas around series of observations represent the sampling variability in the series (quantified by two times of the sampling standard errors). VR data is denoted by connected black squares. Excluded data series and detailed information on all data series are displayed in Figure S1.

Figure 9. Comparison of U5MR estimates for Burkina Faso, Mali and Sao Tome and Principe where the change in curve fitting method changed the estimate by more than 10 deaths per 1,000 live births.

Estimates compared are from UN IGME 2013 (solid red line with 90% credible intervals given by the shaded regions), B2 fit to 2013 database (solid light green line with 90% credible intervals given by the shaded regions), default Loess fit to 2013 database (dashed dark blue line). Connected dots denote data from the UN IGME 2013 database and gray shaded areas around series of observations represent the sampling variability in the series (quantified by two times of the sampling standard errors). Excluded data series and detailed information on all data series are displayed in Figure S1.

Figure 10. Decomposition of differences in under-five deaths in 1990 and 2011 into differences due to the WPP update and differences due to updates in U5MR estimates.

The gray box represents differences up to 10%. Regions with differences of more than 10% due to either factor are highlighted in red.