OBJECTIVE: Community ART groups (CAG), peer support groups involved in community ART distribution and mutual psychosocial support, were piloted to respond to staggering antiretroviral treatment (ART) attrition in Mozambique. To understand the impact of CAG on long-term retention, we estimated mortality and lost-to-follow-up (LTFU) rates and assessed predictors for attrition. METHODS: Retrospective cohort study. Kaplan-Meier techniques were used to estimate mortality and LTFU in CAG. Individual- and CAG-level predictors of attrition were assessed using a multivariable Cox proportional hazards model, adjusted for site-level clustering. RESULTS: Mortality and LTFU rates among 5729 CAG members were, respectively, 2.1 and 0.1 per 100 person-years. Retention was 97.7% at 12 months, 96.0% at 24 months, 93.4% at 36 months and 91.8% at 48 months. At individual level, attrition in CAG was significantly associated with immunosuppression when joining a CAG, and being male. At CAG level, attrition was associated with lack of rotational representation at the clinic, lack of a regular CD4 count among fellow members and linkage to a rural or district clinic compared with linkage to a peri-urban clinic. CONCLUSIONS: Long-term retention in this community-based ART model compares favourably with published data on stable ART patients. Nevertheless, to reduce attrition, further efforts need to be made to enroll patients earlier on ART, promote health-seeking behaviour, especially for men, promote a strong peer dynamic to assure rotational representation at the clinic and regular CD4 follow-up and reinforce referral of sick patients.