Indirect estimation of mortality assessment and improvement of methods

This study investigates the extent of bias in the estimates of infant and under-five mortality derived from the Brass children ever born children surviving (CEB/CS) method as a result of HIV/AIDS. The bias is estimated by comparing the infant and under-five mortality derived from the CEB/CS method with direct estimates from the full birth history data from recent DHS data. The estimates from the full birth history data have been corrected for bias due to HIV/AIDS using the method used by IGME.

IMRs and U5MRs derived from data from women aged 25-39 were underestimated by up to 15% in the six countries studied. Estimates of bias in data derived from women aged 20-24 differed between countries. The results from these younger women could be affected by differences between the indirect and direct methods of estimation. In two of the countries, estimates of overall bias of more than 30% were observed. The bulk of the overall bias is due to the effect of HIV on the survival of mothers and their children. The choice of model life table does not introduce much bias, especially in estimates of under-five mortality where the absolute bias in most countries was less than 3%.

Standard techniques of indirect estimation of child mortality use data from summary birth histories consisting of only two questions - number of children ever born alive and the number of children dead. However the estimation is based on several assumptions about fertility and mortality patterns, and rates computed for recent periods are biased. We propose and apply an innovative approach based on imputation of full birth histories onto summary birth histories. The resulting imputed full birth history is used to calculate child mortality rates using standard life table procedures. We apply the approach to data from the Malawi 2008 Population Census and the 2004 and 2010 Demographic and Health Survey datasets. Preliminary results are promising, with most of the imputed child mortality rates falling within the 95% confidence intervals of the rates directly computed from the 2010 DHS survey. In addition, choice of the existing full birth history data for the imputation did not appear to affect the resulting mortality rates computed from the imputed full birth history data.

The under-5 mortality rate (U5MR) is an important measure of the well-being of a country’s children, and its estimation is particularly critical as we approach the deadline of Millennium Development Goal 4 to reduce U5MR by two-thirds between 1990 and 2015. However, for the great majority of developing countries without well-functioning vital registration systems, estimating levels and trends in under-5 mortality is challenging, not only because of limited data availability but also because of issues with data quality. Global estimates of child mortality are often constructed without accounting for potential biases in data series which may lead to inaccurate point estimates and/or uncertainty intervals. We propose a Bayesian spline regression model for assessing levels and trends in the U5MR, whereby biases in data series are estimated for each source type through the inclusion of a multilevel model to improve upon the drawbacks of current methods. Preliminary results show that the proposed model is able to flexibly capture changes in U5MR over time and gives point estimates and uncertainty intervals that reflect potential biases in data series.

In the absence of a well-functioning vital registration system to track mortality in a population, health planners often rely on routine health surveys to provide this most basic health information. Sibling survival histories, where a survey respondent is asked about each of his or her siblings’ births and, if applicable, deaths, provide a direct way to estimate adult mortality by survey. The purpose of this paper is to refine the methods which account for the selection bias, zero-surviving reporters and recall bias inherent in these surveys to generate plausible estimates of adult mortality rates even in the presence of a relationship between family size and adult mortality.

We have implemented changes to the previous method, referred to as the Corrected Sibling Survival (CSS) method, such that it (1) uses appropriate survival weights that account for the study design, and (2) recovers the mortality experience of the families that are not represented because none of the siblings is alive and eligible to respond to the survey. We validate these methodological developments in a range of simulation environments. We also present new ways of adjusting for recall bias and handling sparse data in survey designs where the age range of the respondents is narrower than the age range desired for estimation.