Cause-specific mortality (cont.)

Information on causes of death is obtained from vital registration systems, verbal autopsies (see session PAPP104-S01), and – in some countries – other sources such as police statistics and cancer registries.

The causes of disease used by GBD studies are based on the International Classification of Diseases (ICD). However, translating vital statistics for different countries and dates into the set of causes analysed in a GBD study requires careful processing to allow for changes in the coding rules between the editions of the ICD and national modifications to it and to redistribute deaths from junk codes used for unknown or only partially described causes to other causes of death.

Estimates for populations for which no, or only limited, data exist on causes of death have to be imputed on the basis of models fitted to populations that have data. For the 1990 GBD study, deaths in such populations were divided between the causes on the basis of the level of all-cause mortality and the regions of the world into which they fell.

For the 2010 GBD study, a more complicated procedure was adopted based on an analytical tool known as CODEm or the Cause of Death Ensemble model (Foreman, Lozano, Lopez et al. 2012). This tool models trends in an individual cause of death using multiple models and a longer list of covariates and pools the estimates from them into an ensemble selected to have the best out-of-sample predictive performance.