back to top back to top back to top

Which questions best identify adult AIDS deaths using verbal autopsy tools?

Date: Tuesday 30 October 10.45–12.15
Source: Forum 11
Authors: Ben Lopman, Research Associate, Department of Infectious Disease Epidemiology, Imperial College London, United Kingdom
with Mark Urassa, Raphael Isingo, Basia Zaba, Simon Gregson, Adrian Cook, Godwin Chawira, Jennifer Smith and Ties Boerma


Verbal autopsy (VA), in which next of kin to the deceased report information on signs and symptoms preceding death, is currently the only option to obtain cause of death information in populations lacking death registration with medical certification. The World Health Organization (WHO) is currently coordinating an effort to develop a common tool for verbal autopsy. The research meets two urgent needs: to validate VA questions designed to identify AIDS-related deaths in different populations and settings; and to demonstrate the performance of a standardized algorithm developed to monitor AIDS mortality in generalized epidemics.

A statistical algorithm based on verbal autopsy data from Zimbabwe has been developed for estimating AIDS mortality in the community. This research: 1) evaluates the accuracy of the algorithm (in place: Zimbabwe and Tanzania) in identifying AIDS deaths and allow a proper validation of the statistical and computer methods; 2) evaluates the accuracy of the algorithm (in time) by testing it against data from a subsequent period following the period when the algorithm was developed; 3) investigates the feasibility of developing a similar algorithm based on questions to be included in the WHO/Health Metrics Network standardized VA tool.

The research will enable us to: 1) determine if data-driven algorithms produce a consistent set of criteria to categorize AIDS deaths using VA; 2) report whether data-driven algorithms perform consistently in terms of sensitivity and specificity in different times and places; 3) recommend whether such algorithms should be used for monitoring AIDS mortality using VA and, if so, whether a single algorithm is reliable for all settings, or a specifically adapted algorithm is required for each application; 4) suggest what modifications are needed to the general purpose INDEPTH VA tool to allow AIDS deaths to be identified with greater accuracy in high HIV prevalence countries.