Estimating the infant mortality rate from DHS birth histories in the presence of age heaping

Abstract

Background The infant mortality rate (IMR) is a critical indicator of population health, but its measurement is subject to response bias in countries without complete vital registration systems who rely instead on birth histories collected via sample surveys. One of the most salient bias is the fact that child deaths in these birth histories tend to be reported with a large amount of heaping at age 12 months. Because of this issue, analysts and international agencies do not directly use IMR estimates based on surveys such as Demographic and Health Surveys (DHS); they rely instead on mortality models such as model life tables. The use of model life tables in this context, however, is arbitrary, and the extent to which this approach appropriately addresses bias in DHS-based IMR estimates remains unclear. This hinders our ability to monitor IMR levels and trends in low-and middle-income countries. The objective of this study is to evaluate age heaping bias in DHS-based IMR estimates and propose an improved method for adjusting this bias. Methods and findings Our method relies on a recently-developed log-quadratic model that can predict age-specific mortality by detailed age between 0 and 5. The model’s coefficients were derived from a newly constituted database, the Under-5 Mortality Database (U5MD), that represents the mortality experience of countries with high-quality vital registration data. We applied this model to 204 DHS surveys, and compared unadjusted IMR values to IMR values adjusted with the log-quadratic model as well as with the classic model life table approach. Results show that contrary to existing knowledge, age heaping at age 12 months rarely generates a large amount of bias in IMR estimates. In most cases, the unadjusted IMR values were not deviating by more than +/- 5% from the adjusted values. The model life table approach, by contrast, introduced an unwarranted, downward bias in adjusted IMR values. We also found that two regions, Sub-Saharan Africa and South Asia, present age patterns of under-5 mortality that strongly depart from the experience represented in the U5MD. For these countries, neither the existing model life tables nor the log-quadratic model can produce empirically-supported IMR adjustments. Conclusions Age heaping at age 12 months produces a smaller amount of bias in DHS-based IMR estimates than previously thought. If a large amount of age heaping is present in a survey, the log-quadratic model allows users to evaluate, and whenever necessary, adjust IMR estimates in a way that is more informed by the local mortality pattern than existing approaches. Future research should be devoted to understanding why Sub-Saharan African and South Asian countries have such distinct age patterns of under-five mortality.

Accumulating Birth Histories Across Surveys for Improved Estimates of Child Mortality

“…They find that while there was overall no significant reduction in age heaping over time, the proportion of age reported which are heaped is relatively low and should not be a major concern. This is in line with Prieto et al (2021) who find that the age heaping at 12 months induced bias in DHS-based IMR estimates is smaller than previously thought. They admit, however that their model cannot be used to produce empirically supported IMR adjustments for Sub-Saharan Africa and South Asia and call for future research that allows understanding the distinct age patterns of under-five mortality in Sub-Sharan African and South Asian countries.…”

Modeling Age Patterns of Under-5 Mortality: Results From a Log-Quadratic Model Applied to High-Quality Vital Registration Data

“…This unusual age pat tern, which has been attrib uted to a com bi na tion of fac tors, includ ing inad e quate weaning foods (Cantrelle and Leridon 1971;Garenne 1982), is absent from the Western expe ri ence, and thus our log-qua dratic model is not able to reproduce it. Outside these extreme cases, many sub-Saharan Afri can pop u la tions tend to dis play an unusu ally late age pat tern of under-5 mor tal ity (Guillot et al 2012), which is not well fit ted by the log-qua dratic model (Romero Prieto et al 2021). As an illustra tion, we show in panel d of Figure 6 observed q(x) val ues from the 2010-2011 DHS for Senegal against log-qua dratic pre dic tions given the same level of q(5y), with k vary ing between −1 and +1.…”

Divergent age patterns of under-5 mortality in south Asia and sub-Saharan Africa: a modelling study

“…3 Greater precision in empirical estimates of age-specific mortality is also necessary for modelling mortality in settings in which empirical data are not available. 4 , 5 …”