Verbal Autopsy:Reimagining data and automated cause assignment (using ALPHA network data)

The overall aim of the grant is to develop affordable, robust, calibrated methods to assign causes to individual deaths and characterize the distribution of deaths by cause from verbal autopsy (VA) data. As part of ALPHA, all partners are being asked to provide their VA data (all ages) alongside any physician review data available, residency data and HIV data for secondary analysis to address the following objectives:

(i) Preparing VA data in a standardised format, and comparing sex- and age- specific cause-specific mortality fractions across different ALPHA partner study areas
(ii) Assessing the performance of automated methods for assigning causes of death based on verbal autopsy data, against physician review or HIV sero-survey data (where each of these data sources are available)

Once these secondary data have been shared with the ALPHA team, we will check the data for consistency and integrity and clean them with the help of the owners of the data and merge the clean data to form a pooled data set. In collaboration with the team at The Ohio State University, we will prepare distributions of deaths by causes using automated methods, and calculate the cause-specific mortality fractions and rates stratified by key socio-demographic characteristics of the population and look at how they change over time. The performance of automated methods will be assessed in comparison to physician review assignment of cause of death or HIV sero-survey data. These results and outputs will be shared with each partner, with opportunity to provide input, and be involved with writing any papers.