Does a verbal autopsy narrative provide accurate information about treatment default for people who have died from HIV/AIDS?

Maqungo, Monique

Does a verbal autopsy narrative provide accurate information about treatment default for people who have died from HIV/AIDS?

Thesis / Dissertation

2025

Publisher

University of Cape Town

Department

Department of Public Health and Family Medicine

Faculty

Faculty of Health Sciences

Abstract

Background: The South African National Cause-of-Death Validation (2017/18) project utilized the WHO 2016 standard VA questionnaire, which included both structured questions and an open narrative from the next of kin to describe the events surrounding deaths. An unexpected finding was the significant number of HIV/AIDS-related deaths mentioning treatment default in the narrative. Objectives: This study aims to determine the accuracy of VA narratives in identifying ART treatment default by linking data from the national treatment register with NCODV 2017/18 HIV/AIDS related death data. Methods: Secondary analysis of VA data from the NCODV 2017/18 project after linking with TIER.Net data on ART treatment maintained by the Department of Health. Agreement between treatment default identified from VA narratives and from TIER.Net was investigated using Cohen's Kappa (k), and the sensitivity and specificity of the narrative for identifying treatment default was estimated. Results: Data linkage succeeded for 691 (58.9%) deaths, with 62 (5.3%) cases excluded due to inconsistencies. Among the 629 linked cases, 48.3% were identified as treatment defaulters in TIER.Net, compared to 29.4% indicated by VA narratives. The agreement between VA narratives and TIER.Net was 59.1%, with a Cohen's Kappa value of 0.17 (95% CI: 0.10–0.24). Sensitivity and specificity of VA narratives for identifying treatment default were 38.2% (95% CI: 34.4%–42.0%) and 78.8% (95% CI: 75.6%–82.0%) respectively, with positive and negative predictive values of 62.7% (95% CI: 58.9%–66.5%) and 57.7% (95% CI: 53.8%–61.5%) respectively. Conclusions: The VA narratives were not sensitive enough to identify all treatment defaulters, likely related to the lack of structured questions related to defaulting but possibly also related to non-disclosure of HIV status and treatment. Of concern is the high proportion of deceased who had defaulted on treatment.