Evaluating the performance of the InterVA Model for determining AIDS mortality in the Adult Population of Addis Ababa

Background Verbal Autopsy (VA) is a method for determining the cause of death based on an interview with next of a kin or other caregivers. Most commonly, causes of death are assigned by physicians who review completed VA questionnaires. This is time consuming and expensive and led to several initiatives for attributing causes of death using algorithms such as the InterVA model. We evaluate the performance of the InterVA model using gold standard, which is based on hospital diagnosis and HIV serostatus of the deceased.

Methods Surveillance of hospital admissions and outpatient visits was initiated in Zewditu Memorial Hospital which was used later to define the gold standard of TB/AIDS deaths. A VA interview was conducted for those died in and out of the hospital. The InterVA model is used to interpret the VA interviews to arrive at probable causes of death.

Results The model assigned the most likely cause of death for all the records with a certainty value of 80.7%. The proportion of AIDS death determined by the gold standard is 56.0% while that of the InterVA is 54.4%. Classifying all deaths as AIDS and Non-AIDS, the sensitivity Value is 0.86 and specificity is 0.73. The sensitivity and specificity of TB/AIDS death are 0.92 and 0.78, respectively.

Conclusion The analysis of VA based on probabilistic model produces promising results when compared to the gold standards to estimate adult AIDS mortality. The InterVA model can be applied to guide priority health interventions activities in low income countries.