South Africa has undergone multiple expansions in antiretroviral therapy (ART) eligibility from an initial CD4+ threshold of ≤200 cells/μl to providing ART for all people living with HIV (PLWH) as of September 2016. We evaluated the association of programmatic changes in ART eligibility with loss from care, both prior to ART initiation and within the first 16 weeks of starting treatment, during a period of programmatic expansion to ART treatment at CD4+ ≤ 350 cells/μl.We performed a retrospective cohort study of 4,025 treatment-eligible, non-pregnant PLWH accessing care in a community health center in Gugulethu Township affiliated with the Desmond Tutu HIV Centre in Cape Town. The median age of participants was 34 years (IQR 28-41 years), almost 62% were female, and the median CD4+ count was 173 cells/μl (IQR 92-254 cells/μl). Participants were stratified into 2 cohorts: an early cohort, enrolled into care at the health center from 1 January 2009 to 31 August 2011, when guidelines mandated that ART initiation required CD4+ ≤ 200 cells/μl, pregnancy, advanced clinical symptoms (World Health Organization [WHO] stage 4), or comorbidity (active tuberculosis); and a later cohort, enrolled into care from 1 September 2011 to 31 December 2013, when the treatment threshold had been expanded to CD4+ ≤ 350 cells/μl. Demographic and clinical factors were compared before and after the policy change using chi-squared tests to identify potentially confounding covariates, and logistic regression models were used to estimate the risk of pre-treatment (pre-ART) loss from care and early loss within the first 16 weeks on treatment, adjusting for age, baseline CD4+, and WHO stage. Compared with participants in the later cohort, participants in the earlier cohort had significantly more advanced disease: median CD4+ 146 cells/μl versus 214 cells/μl (p < 0.001), 61.1% WHO stage 3/4 disease versus 42.8% (p < 0.001), and pre-ART mortality of 34.2% versus 16.7% (p < 0.001). In total, 385 ART-eligible PLWH (9.6%) failed to initiate ART, of whom 25.7% died before ever starting treatment. Of the 3,640 people who started treatment, 58 (1.6%) died within the first 16 weeks in care, and an additional 644 (17.7%) were lost from care within 16 weeks of starting ART. PLWH who did start treatment in the later cohort were significantly more likely to discontinue care in <16 weeks (19.8% versus 15.8%, p = 0.002). After controlling for baseline CD4+, WHO stage, and age, this effect remained significant (adjusted odds ratio [aOR] = 1.30, 95% CI 1.09-1.55). As such, it remains unclear if early attrition from care was due to a "healthy cohort" effect or to overcrowding as programs expanded to accommodate the broader guidelines for treatment. Our findings were limited by a lack of generalizability (given that these data were from a single high-volume site where testing and treatment were available) and an inability to formally investigate the effect of crowding on the main outcome.Over one-quarter of this ART-eligible cohort did not achieve the long-term benefits of treatment due to early mortality, ART non-initiation, or early ART discontinuation. Those who started treatment in the later cohort appeared to be more likely to discontinue care early, and this outcome appeared to be independent of CD4+ count or WHO stage. Future interventions should focus on those most at risk for early loss from care as programs continue to expand in South Africa.