Studies on the long-term safety and tolerability of HAART are scarce in developing countries. HAART has been universally available in Brazil since 1997, providing a unique opportunity to evaluate the incidence and risk factors for HAART discontinuation or modification. We analyzed retrospective data from 670 treatment-naive patients followed at the HIV cohort of Evandro Chagas Clinical Research Institute, Oswaldo Cruz Foundation, in Rio de Janeiro, Brazil, who first received HAART between January 1996 and December 2006. Our four outcomes of interest were treatment failure (TF-MOD), short-term toxicity (ST-MOD), long-term toxicity (LT-MOD), and overall modification/discontinuation (MOD, composed of TF-MOD, ST-MOD, LT-MOD, and other reasons). Risk factors were assessed using Cox's proportional hazards regression. Incidences of MOD, ST-MOD, LT-MOD, and TF-MOD were 28.3, 24.0, 4.0, and 5.6 per 100 persons-years, respectively. MOD was observed in 69% of the patients; 40% of the MODs were toxicity related. The risk of MOD in the first year of treatment was 32% (95% CI: 28.3–35.5%); the median time from HAART initiation to MOD was 14 months (IQR: 3.0–29.5). The most frequent reasons for ST-MOD were gastrointestinal; women had a higher hazard for ST-MOD. Metabolic toxicity was the most frequent reason for LT- MOD, particularly dislipidemia and lipodystrophy. Increased hazard of TF-MOD was observed among those with lower CD4⁺ lymphocyte counts (<200 cells/mm³). Our results indicate that toxicities can compromise adherence and thus impact future treatment options. This is especially relevant in the context of limited access to second and third line treatment regimens.