Abstract
INTRODUCTION: Retention on antiretroviral therapy (ART) is a prerequisite for adherence and subsequent treatment success. Measuring retention is also easily implementable at facility and population levels, making it pragmatic to monitor ART programme success. However, despite its ubiquitous global use, there is little consistency in the measurement of retention. METHODS: This study retrospectively applied six measures of retention to one cohort of adults (initiating ART after 01-09-2016, with ≥1 year of observation time to database closure on 30-09-2022), in a low-resource, high HIV-burden setting in South Africa. Using routine healthcare data from the Western Cape's Provincial Health Data Centre, loss to follow-up (LTFU), fixed-point retention, visit constancy, visit gaps, treatment interruptions and medication possession ratio (MPR) were described over 5 years from initiation. Individuals were considered "continuously retained" if they did not experience attrition throughout their observed follow-up. Measures were compared using the proportion misassigned and Cohen's Kappa statistic. RESULTS: The median age of the cohort (n = 68,888) was 31 years (interquartile range [IQR] 26-38) at initiation, with 69% (47,631/68,888) female, and a median observed follow-up of 4 years (IQR 3-5). Across different measures, retention was low, and declined over time. There was variable overlap; the proportion continuously retained throughout their observed follow-up ranged from 60% (41,268/68,888 not LTFU) to 32% (22,381/68,888 MPR ≥80%). Retention by all measures was strongly associated with viral suppression. CONCLUSIONS: By all measures, large proportions of people in this setting were considered out of ART care during 5 years of observed follow-up time from initiation. This makes retention a critical target for intervention to improve population-level viral suppression and achieve epidemic control. Measuring longitudinal retention revealed that most people disengaged from ART care at some point after initiation. Certain measures of retention (e.g. treatment interruptions) identified people in and out of care with more granularity, whereas blunter measures (e.g. LTFU) misassigned individuals' retention status and missed patterns of retention over time as people cycled in and out of care between points of measurement. Ultimately, the choice of measure depends on the purpose of the evaluation and on the data available, but, where possible, more granular measures are recommended.