LAUSR

OBJECTIVES Measuring retention is critical for antiretroviral therapy (ART) management and program monitoring, however many definitions and data sources, usually from single health facilities, are used. We used routine electronic data, linked across facilities, to examine the impact of definitions and data sources on retention estimates among women in Cape Town, South Africa. DESIGN Retrospective cohort study METHODS:: We compiled routine electronic laboratory, pharmacy and clinic visit data for 617 women who started ART during pregnancy (2013-2014) and estimated 24-month retention using different definitions and data sources. We used logistic regression to assess consistency of associations between risk factors and retention, and Receiver Operating Characteristics (ROC) analyses to describe how different retention estimates predict viremia at 12 months on ART. RESULTS Using all available data sources, retention ranged from 41% (no gap > 180 days) to 72% (100% 12-month visit constancy). Laboratory data (expected infrequently) underestimated retention compared to clinic visit data that identified > 80% of women considered retained in all definitions. In all estimates, associations with known risk factors for non-retention remained consistent and retention declined over time: 77%, 65% and 58% retained using all data sources in months 6-12, 12-18 and 18-24, respectively (p < 0.001). The 180-day gap definition was most strongly associated with viremia (OR 24.3 95%CI 12.0-48.9, all data sources). CONCLUSION Researchers must carefully consider the most appropriate retention definition and data source depending on available data. Presenting more than one approach may be warranted to obtain estimates that are context-appropriate and comparable across settings.