Using verbal autopsy to assess the prevalence of HIV infection among deaths in the ART period in rural Uganda: a prospective cohort study, 2006-2008

BACKGROUND: Verbal autopsy is important for detecting causes of death including HIV in areas with inadequate vital registration systems. Before antiretroviral therapy (ART) introduction, a verbal autopsy study in rural Uganda found that half of adult deaths assessed were in HIV-positive individuals. We used verbal autopsy to compare the proportion of HIV-positive adult deaths in the periods before and after ART introduction. METHODS: Between 2006 and 2008, all adult (≥ 13 years) deaths in a prospective population-based cohort study were identified by monthly death registration, and HIV serostatus was determined through annual serosurveys. A clinical officer interviewed a relative of the deceased using a verbal autopsy questionnaire. Two clinicians independently reviewed the questionnaires and classified the deaths as HIV-positive or not. A third clinician was the tie-breaker in case of nonagreement. The performance of the verbal autopsy tool was assessed using HIV serostatus as the gold standard of comparison. We compared the proportions of HIV-positive deaths as assessed by verbal autopsy in the early 1990s and the 2006-2008 periods. RESULTS: Of 333 deaths among 12,641 adults of known HIV serostatus, 264 (79.3%) were assessed by verbal autopsy, of whom 59 (22.3%) were HIV-seropositive and 68 (25.8%) were classified as HIV-positive by verbal autopsy. Verbal autopsy had a specificity of 90.2% and positive predictive value of 70.6% for identifying deaths among HIV-infected individuals, with substantial interobserver agreement (80.3%; kappa statistic = 0.69). The HIV-attributable mortality fraction estimated by verbal autopsy decreased from 47.0% (pre-ART period) to 25.8% (ART period), p < 0.001. CONCLUSIONS: In resource-limited settings, verbal autopsy can provide a good estimate of the prevalence of HIV infection among adult deaths. In this rural population, the proportion of deaths identified by verbal autopsy as HIV-positive declined between the early 1990s and the 2006-2008 period. Verbal autopsy findings can inform policy on HIV health care needs.