No More Guesstimation: How to Reliably Estimate Trends in Child Mortality

Reliable estimates of child mortality are critical for monitoring progress toward Millennium Development Goal 4 (MDG-4), so I asked Dr. Kenneth Hill, who currently chairs the Technical Advisory Group (TAG) of the United Nations Inter-agency Group on Mortality Estimation (UN IGME), for some insights into the methodological innovations captured in a new sponsored collection, Child Mortality Estimation Methods, published in PLOS Medicine today.

In relation to MDG-4, why is the work the TAG does important?

When the Millennium Development Goals were established in the year 2000, the objective was to focus national and international effort on a small number of critical areas of development with measurable targets.  The reasons for having measurable targets were to be able to identify potential barriers to achieving the targets and to be able to hold countries and international agencies accountable for success or failure.  MDG-4 was focused on the improvement of child health, the primary target being the reduction of under-5 mortality by two-thirds from 1990 to 2015.  However, most high-mortality countries do not have accurate vital statistics, and the measurement of under-5 mortality relies on data from surveys of one sort or another.  The work of the TAG focuses on the development and implementation of methods to derive country-specific estimates of levels and trends of under-5 mortality from disparate sources of variable quality, thus providing the basis for assessing progress towards MDG-4 and identifying barriers to progress.

What are the major limitations of current methodology?

The greatest barriers to producing timely and accurate national estimates of child mortality arise from data limitations.  Only about 60 countries globally have well-functioning civil registration systems that provide accurate and timely estimates of child mortality.  The remaining 130 or so countries rely on surveys to produce estimates, and a number of them – often the poorest ones or those with persistent security issues – have very few nationally-representative surveys on which to base estimates of trends.  Even in better-off or more stable settings, the sorts of surveys – collecting detailed birth histories from a representative sample of women – that underpin the estimates for many low- and middle-income countries have small samples (limiting the time frame for which statistically-stable estimates can be derived, often to intervals as wide as 5 years) and are only conducted every five years or so.  This detracts from the timeliness of estimates, and makes it almost impossible to capture very recent trends. Given the data availability, there are also good surveys and bad ones, and we haven’t found an effective way of distinguishing between them.  Then there are some technical issues, such as selection biases, most serious in countries with HIV epidemics where high-risk children will be less likely to be reported because their mothers have also died.  Current estimates also tend to underestimate statistical uncertainty in the estimates, an issue we are actively working on.

What did you find most surprising about the most recent UN IGME estimates?

Most surprising to me, and probably a result that most people would find counter-intuitive, is the excellent performance in reducing under-5 mortality of the countries of North Africa.  As a region, these countries have already achieved the target reduction in under-5 mortality required by MDG-4.  Also surprising, but profoundly relieving, is the clear evidence of a recent acceleration in progress in sub-Saharan Africa.  This region has for decades been seen as under-performing, but seems to have turned a corner in the first decade of this century.

Why did UN IGME choose to provide estimates of child mortality indicators by sex for the first time in the 2011 estimates?

UN IGME has historically only made national-level estimates of child mortality indicators.  However, there is increasing interest in looking at within-country differentials in mortality as well.  We started by looking at differentials by sex because basic patterns are well-established.  In populations with no differences in treatment of girls and boys, girls always have lower mortality than boys, so an observed equality of the mortality risks of boys and girls actually probably indicates discrimination against girls.  UN IGME wanted to investigate child mortality differences between girls and boys by country and region, so we established an expected boy/girl differential by mortality level, and then compared observed country differences to that pattern.  Some world regions show evidence of discrimination against girls on this basis.

Current trends indicate that we will not meet MDG-4 by 2015, what do you think is most needed to increase the rate of reduction of child mortality?

We need to focus efforts on the world regions that have been performing relatively poorly over the past decade (sub-Saharan Africa, southern Asia, Oceania), and to focus on causes of death that account for a large portion of under-5 mortality and for which cheap and effective interventions exist (notably pneumonia, diarrhoea and malaria, which account for more than one-third of all under-5 deaths globally).  Another priority area is neonatal mortality, which accounts for 40% of global under-5 deaths; however, major reductions in neonatal mortality will probably require a much higher proportion of births to occur in health facilities, but appropriate care of the newborn can make a big difference in any setting.

Where will the TAG focus to improve the accuracy of future estimates?

Ultimately, low- and middle-income countries need to improve their civil registration systems in order to provide timely and accurate estimates of child mortality.  However, such improvements are still a long way in the future, and periodic, well-designed household surveys remain a priority.  In the interim, the TAG will focus its efforts on (i) improving on the current loess fitting procedure; (ii) the estimation of appropriate uncertainty bounds around the estimates; (iii) increasing the flexibility of the method used to estimate bias in direct estimates of U5MR in countries affected by HIV to incorporate the effectiveness of antiretroviral therapy in reducing AIDS mortality; (iv) to improve methods for estimating child mortality for countries affected by civil unrest; and (v) developing better ways of evaluating data quality on which to base inclusion and exclusion criteria in the fitting procedures.

+++++++++++

Kenneth Hill is a demographer who specializes in deriving estimates of mortality from incomplete or deficient data.  He has a Ph.D. from the London School of Hygiene and Tropical Medicine, and has spent most of his career at the Johns Hopkins Bloomberg School of Public Health and the Harvard School of Public Health. Kenneth_hill_1@yahoo.com

The UN Inter-agency Group for Child Mortality Estimation (IGME) annually reports on country, regional and global trends in child mortality. In the Child Mortality Estimation Methods Collection the independent Technical Advisory Group (TAG) to the UN IGME introduces the group’s methodological innovations in estimating child mortality. http://www.ploscollections.org/childmortalityestimation #plosChildMortality

The Collection is produced with support from UNICEF and the TAG of the UN IGME.