GBD Demographic Estimation Methodology

The GBD estimation strategy for fertility, mortality, and population is designed to work with the diversity of data sources and potential biases in data available for each of these demographic components and to use replicable statistical code for data synthesis. The analysis can be divided into seven main steps: age-specific fertility estimation, under-5 mortality estimation, adult mortality estimation, age-specific mortality estimation using a relational model life table system, HIV adjustments, accounting for fatal discontinuities such as wars or natural disasters, and population estimation. For each component, it is useful to think of the data available, the data processing steps required to account for known biases, and the data synthesis stage, which deals with the challenges of both missing measurements in given location-years and the common problem of different measurements disagreeing with each other.
For GBD 2019, we instituted the GBD standard location list, which consists of all national-level locations as well as subnational locations in the UK, India, China, and the USA. In each modelling step, effects of the covariates were derived from empirical data observed from standard locations. This ensured that our estimates were derived from robust relationships extrapolated from locations with more robust empirical data, thus ensuring long-term stability in our estimates.
Below, we provide a high-level description of each analytical component, with an emphasis on new steps and other updates for GBD 2019. Methods used in the GBD demographic estimation process have been described extensively in previous publications,14 (link), 15 (link), 16 , 17 (link), 18 (link) and additional detail on estimation for the 2019 cycle is available in appendix 1.
This study complies with GATHER;¹⁹ (link) a completed GATHER checklist is available in appendix 1. Analyses used Python version 3.6.2 and 3.6.8, Stata versions 13 and 15, and R versions 3.4.2 and 3.5.0.