Abstract Body

Background: Cause of death data are crucial to guide public health policy. Verbal autopsy (VA) is widely used to collect these data, but has mostly been validated against gold standards of variable quality, generated from hospital notes. Full post-mortem is the best way to determine cause of death, though needle autopsy (NA) is more acceptable to relatives and can provide high quality information. In HIV+ individuals, differentiating between deaths due to tuberculosis (TB) and advanced HIV is difficult, but important to guide interventions to reduce early mortality. This ongoing cross-sectional study, nested within the TB Fast Track trial in South Africa, compared causes of death assigned by VA and NA in adults with advanced HIV.

Methods: Adult trial participants had a CD4 count of ≤150 cells/μL and were not taking TB treatment or antiretroviral therapy (ART). Consent for NA was obtained from participants at enrolment or from families after death (if participant not asked). NAs involved biopsies of liver, spleen, and lungs; aspiration of CSF; and broncheo-alveolar lavage. Laboratory tests included: histology; GeneXpert MTB/RIF; and MGIT, bacterial, and fungal cultures. Preliminary NA causes of death were assigned by investigators using demographic and NA data. VAs were performed within one year of death by lay-workers using the 2012 WHO tool; causes of death assigned by InterVA-4 and SmartVA software (using health care experience and narrative data). For comparison, causes of death were categorised as TB, AIDS/HIV-related (not TB), or ‘other’. Cohen’s kappa coefficient was used to measure inter-rater agreement.

Results: The first 20 decedents with NA and VA data were included in the analysis. 12/20 (60%) were female, with median: age at death 37.8 (interquartile range [IQR] 31-44) years; CD4 count at enrolment 31 (IQR 14-56) cells/mL; time from enrolment to death 45 (IQR 20-103) days; time from death to NA 4.5 (IQR 3-7) days; and time from death to VA 106.5 (IQR 60-178) days. Causes of death assigned by NA and VA are shown in Table 1. Agreement, expected agreement and kappa between: NA and InterVA-4 were 40.0%, 42.0%, and 0.00; and NA and SmartVA were 50.0%, 50.0% and 0.00, respectively.

Conclusions: These preliminary data suggest that current VA tools do not perform well at distinguishing between deaths due to TB vs. other HIV-related causes among individuals with advanced HIV disease. Alternative methods should be explored so that the burden of TB mortality can be more accurately recorded.

viewer?TAG_ACTION=SECURE_FILE_DOWNLOAD&V
Table 1: Causes of death assigned by needle autopsy and by verbal autopsy as interpreted by InterVA-4 and SmartVA