Abstract
INTRODUCTION: Estimating the incidence of antiretroviral discontinuations due to adverse drug reactions (ADRs) is important to inform antiretroviral treatment (ART) regimen recommendations, and to guide prescribing and monitoring policies. Routinely collected clinical data is a useful source of pharmacovigilance data. We estimated the incidences of first-line antiretroviral discontinuations due to ADRs using routine clinical data, and compared them with incidences estimated using data enhanced by folder review, in two South African cohorts. METHODS: We included patients 16 years and older on first-line ART. We selected a stratified random sample of 25% for checking of ART prescription data and reasons for antiretroviral discontinuations retrospectively, including folders reviews where required (enhanced-data sample). We estimated the incidence of antiretroviral discontinuations using Kaplan-Meier and competing risk analyses. RESULTS: In 15396 patients, 40% had a first-line antiretroviral discontinuation by three years on ART. We could determine the reason for 65% of discontinuations using routine data only, and 84% of discontinuations, in the enhanced-data sample of 3837 patients. ADR was the most common reason for discontinuations. In the enhanced-data sample, the cumulative incidence of discontinuations due to ADRs by three years was 30.4% (95% CI: 24.4-36.6) for stavudine; 2.0% (95% CI: 1.5-2.6) for tenofovir, and 1.3% (95% CI: 0.8-2.1) for efavirenz. Using routine data only, the cumulative incidences of discontinuations due to ADRs by three years for stavudine, tenofovir, and efavirenz respectively were 23.9% (95% CI: 20.3-27.7), 1.2% (95% CI: 0.9-1.4) and 0.5% (95% CI: 0.3-0.7). CONCLUSIONS: Although the relative rankings were similar using routine or enhanced data, lack of checking for missing reasons for discontinuation resulted in underestimates of the incidence of antiretroviral discontinuations due to ADRs. Systems to improve data collection of reasons for regimen changes prospectively would increase the capacity of routine data to answer pharmacovigilance questions.