A Bayesian Hierarchical Multivariate Poisson-Lognormal Model to Estimate Age- and Cause-Specific Child Mortality in Data-Scarce Countries

• Activity Number: 256 - Contributed Poster Presentations: Section on Bayesian Statistical Science
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Austin Edward Schumacher* and Jon Wakefield
• Companies: University of Washington and Univ of Washington
• Keywords: bayesian; hierarchical; mortality; child mortality; verbal autopsy

Abstract

As investment increases in implementing age-targeted disease-specific childhood interventions in data-scarce countries, effectiveness requires knowledge of both the age patterns of child deaths and the causes responsible at each age. Current methods either (i) model single causes, (ii) estimate age- or cause-specific child mortality only in broad age groups, (iii) produce estimates in each age group separately and independently, (iv) develop all-cause and cause-specific mortality in two separate estimation frameworks, or (v) utilize a single source of data for each country. We propose a novel Bayesian hierarchical multivariate Poisson-lognormal model to use national registration and survey-based cause of death assessment data together to estimate age- and cause-specific child mortality accounting for correlations in age, cause, and time. We explore the statistical properties of this model via simulation and apply it to data from Bangladesh to estimate cause-specific age and time trends in child mortality.