South Africa has the largest number of people living with HIV and the largest HIV treatment program in the world, supplying antiretroviral therapy (ART) to 66% of the 7.6 million people living with HIV in the country in 2019. To reach the remaining 34%, the already overburdened health system needs to find ways of attracting and retaining groups at higher risk of attrition and optimizing convenience for providers and patients. We identified three examples of “differentiated service delivery”, an approach that adapts HIV services to patient and health system needs: (1) male clinics, attended and staffed exclusively by men, (2) youth clinics, exclusively for youth aged 12–25, offering flexible hours and youth-targeted services and (3) a pharmacy-led fast-track ART refill program where stable ART patients can pick up medication without seeing a clinician. We explore attrition (defined as death or loss to follow-up at end of follow-up time) in these services using data from a large, established HIV cohort in Khayelitsha, a high HIV-prevalence, low-income area in South Africa. The first study examines whether males attending two male clinics (Male Clinic 1 and Male Clinic 2) show lower attrition compared to those attending general primary healthcare clinics. Using exposure propensity scores, we matched male clinic patients 1:1 to males at other clinics and used Cox proportional hazards regression to estimate the association between attrition and attending a male clinic. In the unmatched cohort, patients from male clinics (n=784) were younger than males from general clinics (n=2726), median age: 31.2 vs 35.5 years. Those initiating at male clinics had higher median CD4 counts at ART initiation (Male Clinic 1: 329 [210–431], Male Clinic 2: 364 [IQR: 260–536] vs. general clinics 258 [IQR: 145–398] cells/mm3). The matched analysis included 1563 person-years among 1568 patients. Patients initiating ART at male clinics had lower attrition (HR: 0.83; 95% CI: 0.69–1.00). When matching and modelling was conducted for Male Clinic 1 and 2 separately, only the more established Male Clinic 1 showed a protective effect (HR 0.83; 95% CI: 0.65–1.07). The second study investigates whether attrition from care among youth (aged 12–25) on ART is lower among youth attending two youth clinics (Youth Clinic A and Youth Clinic B) compared to those attending general primary healthcare clinics. We also conducted a sub-analysis of patients attending adherence clubs (a model of ART delivery led by a lay facilitator, including a peer support group). We hypothesized that the effect of peer support in adherence clubs might be enhanced by the age-specific clubs at the youth clinics. It may also be further enhanced by additional elements of the adherence club model offered only in Youth Clinic A, including integration of family planning. Youth at the youth clinics were more likely than those at general clinics to have initiated ART before August 2011, particularly those at Youth Clinic B (23% compared to 11% at general clinics). The distribution of age, sex, and CD4 count at ART initiation was similar across youth and general clinics. We observed a protective effect of youth clinics against attrition: HR 0.81 (95% CI: 0.70–0.93) for Youth Clinic A and 0.85 (0.74–0.98) for Youth Clinic B, compared to youth at general clinics. Youth Clinic A club patients had lower attrition after joining an adherence club compared to general clinic patients in adherence clubs (crude HR: 0.56, 95% CI: 0.32–0.96; adjusted HR: 0.48, 95% CI: 0.28–0.85), while Youth Clinic B showed a smaller difference (crude HR: 0.83, 95% CI: 0.48–1.45; adjusted HR: 1.07, 95% CI: 0.60–1.90). The third study assesses attrition among patients in a pharmacy-led fast-track ART refill program compared to matched stable, otherwise healthy, patients who were eligible for the fast-track program at the same point in time and at the same facility but did not join. Matched pairs were followed from the date of the fast-track patient’s first fast-track ART pick-up, and attrition was compared using Cox proportional hazards regression. Fast-track patients and matched controls had similar characteristics at ART initiation and at fast-track enrolment. Fast-track patients were less likely to have previously experienced tuberculosis (23% vs 28%), diabetes (1% vs 7%) and hypertension (12% vs 27%), compared to those not in fast-track. Fast-track enrolment was highly protective against attrition (HR: 0.40; 95% CI: 0.31–0.51). We hypothesized that some of the association could be explained by confounding, arising from clinicians differentially referring patients to fast-track, possibly based on social, health, or mental health characteristics not reflected in the data. In a bias analysis using a plausible range of effects of such unmeasured confounding, the hazard ratio accounting for random and systematic error was 0.60 (95% interval: 0.42–0.89). All three studies show some protective effects of these differentiated models of service delivery against attrition. While stand-alone youth and male clinics are not feasible in all settings, and fast-track models may not be suited to all patients, these results suggest that multiple approaches tailored to specific populations’ needs can contribute to improving retention.