Bayesian Mortality Estimation from Disparate Data Sources

Project Summary: The goal of the proposal is to develop a Bayesian statistical framework for mortality estimation from disparate data sources. Using this framework we will produce a suite of principled methods to be used in those situations in which vital registration data are lacking. We will emphasize efficient implementations that can be used by researchers in low- and middle-income countries (LMICs), who may have limited computing resources. In Aim 1, we will develop guidelines on a general statistical framework for mortality estimation. Aim 2 will focus on subnational child mortality with particular emphasis on the under-5 mortality rate (U5MR), which is a key indicator of the health of a population, and the neonatal mortality rate (NMR). Excess mortality estimation during the Covid-19 pandemic, by month, at the country level, will be the subject of Aim 3. We will disseminate results widely and provide software and training in the developed methods. We will produce yearly estimates of U5MR and NMR at the geographical level at which health decisions are made. To achieve this goal, household survey, VR and census data must be combined in a coherent way. Census data on child mortality typically provide summary birth history (SBH) data, which consist of mother's age along with the number of children born and the number who died, but without the times at which those events occurred. We will develop a framework for combining the different data sources, which will entail dealing with the design issues in the household survey, accounting for unknown birth and death times in the SBH data, and estimating the completeness of the VR data (births and deaths). We will also incorporate demographic information via a form of Bayesian benchmarking. Effective and appropriate use of the models will require rigorous model assessment, careful interpretation of results and meaningful and informative graphical summaries. We will develop robust models to evaluate the excess mortality, i.e., the difference between the deaths ob- served in the pandemic and those expected if the pandemic had not occurred. We will model the expected deaths, and incorporate the uncertainty in this endeavor in the excess mortality calculation. Completeness of mortality counts, that is, under-reporting and delays in reporting, will also be considered. For countries who do not report deaths in the pandemic, we must predict the mortality count using available country-level covariate data, and we will adopt flexible yet interpretable regression forms, and acknowledge uncertainty in the covariate data. We will produce user-friendly software for the methods, along with vignettes and training materials, including short courses. The endpoint is to have software that can be used by researchers in LMICs. All aims will be informed by the collaborative team's close links with the United Nations Inter-agency Group for Child Mortality Estimation (for the subnational child mortality aim) and the World Health Organization Division of Data, Analytics and Delivery for Impact (for the excess mortality aim). Together we will develop methods to highlight disparities and inform interventions.