Background: Information of numbers and causes of death in the developing world are normally based
on estimates given by population registries or health facilities, which cover only
a small proportion of the target area. However, monitoring and evaluation plays a
major role in formulating good public health policy [1]. Additionally, two of the
Millennium development goals require specific mortality measures, thus highlighting
the need for valid and representative data on mortality [2]. Verbal autopsies (VA),
collecting information on conditions, signs and symptoms experienced before death
by the deceased in a questionnaire a certain time after death by trained fieldworkers
interviewing the family of the deceased address these goals. The most common and generally
accepted method to determine the probable cause of death is the Physician Coded Verbal
autopsy (PCVA), an independent review of the collected questionnaires by physicians
trained in VA coding. Objectives: To compare the results of a Bayesian algorithm to determine probable causes of death
with those from the PCVA approach. Method: A Bayesian model to interpret VA (InterVA) data tries to minimize misclassification
error in cause of death assignment and to alleviate known problems with the PCVA [3,4].
A set of 107 indicators is derived from the VA questionnaire, which influence the
a posteriori probabilities of causes of death [5]. Results: 6,000 records from a rural Health and Demographic Surveillance System in Burkina
Faso are currently entered in a database, covering the years 1999–2008. First results
of the InterVA model compared to the PCVA will be available in the middle of 2010.
As the InterVA model provides up to three possible causes of death (each with a certain
probability) its strength will be providing a standardised method of VA interpretation
and deriving population level cause-specific mortality fractions, taking not only
one possible cause of death into account.
