In the HIV care and treatment program of Arua, in rural Uganda, more than 70% of the patients remained in care after 1 or 2 years of treatment, but over 35% of patients remained at risk of developing opportunistic infections (<200 cells/mm2). Estimates of retention in care appear to be more favorable than those reported during the early years of the Drug Access Initiative (DAI) program among individuals treated in urban hospitals of Kampala, where patients were requested to pay the care fees (49–56% at 1 year, and 46% at 2 years of ART) and more than 30% were ARV-experienced [3, 5]. Like in other African settings, most of our patients started treatment at an advanced stage of HIV disease [4, 10, 22–24], which partly explains why a considerable proportion of patients had low CD4 cell counts after 1 or 2 years of therapy. As reported previously [11, 25–27], early mortality in our cohorts was high with 88% of deaths occurring in the first 6 months of follow-up under ART. Delays in access to care (>80% of patients started treatment at an advanced clinical stage of HIV disease), delayed diagnosis and treatment of severe opportunistic infections (e.g. TB), undiagnosed severe immune reconstitution inflammatory syndromes, and nutritional deficiencies might have contributed to this increased early mortality [28–31].
Despite the frequency of clinical intolerances reported, the initial ART regimen was well tolerated, and few cases of severe (grades 3 or 4) laboratory-based toxicity were diagnosed. Self-reported adherence at 12 and 24 months of ART was high and similar to that reported in other studies, both in resource-limited countries  and North America [33, 34], with a higher proportion of patients reporting good adherence in the M24 cohort than the M12 cohort (90% compared with 80%, respectively). Although reported adherence is subject to measurement error and we cannot exclude that some patients over-reported their level of adherence to ART during the survey, these findings were consistent with plasmatic ARV pharmacological measurements obtained in the same patients. However, the reported levels of adherence do not necessarily reflect patient's compliance throughout the duration of treatment.
The observed immune and virological responses at 1 and 2 years of ART were similar to those reported in other African settings [4, 10, 11, 22, 35, 36] and in Western countries . However, more than 35% of patients had a CD4 count <200 cells/ml and were therefore at increased risk of death due to occurrence of opportunistic diseases. In the M12 cohort, undetectable VL was achieved by 75% of the patients, and 89% had a VL <1,000 copies/ml. In Malawi, 87% of the patients receiving a 2 NRTI/1 NNRTI-based regimen had a VL <1,000 copies/ml after a median of 10 months . The DART study conducted in Zimbabwe and Uganda reported that 72% of patients who received AZT/3TC/TDF for a year had undetectable VL . Also, in a teaching hospital of Kampala, undetectable VL was observed in 86% of patients treated with d4T/3TC/NVP for 1 year . Similarly, in a patient cohort in Khayelitsha, South Africa, 70% of patients achieved undetectable VL at 1 year of ART . In our M24 cohort, undetectable VL was achieved by 71% of the patients. Lower virological success rates have been reported in four urban clinics in Senegal, Côte d'Ivoire, Uganda, and Kenya after 2 years of ART (40–69% had a VL <400 copies/ml) .
Overall, 7% of patients with virological failure at M12 and 13% at M24 had resistant virus, which is similar to estimates reported for patients treated with 2 NRTI/1 NNRTI-based therapy in east Africa or South-east Asia after more than 9 months of treatment (9–18%) [10, 35, 38]. This figure might be underestimated if patients with unsuppressed HIV VL (400–999 copies/ml) but not virological failure also had viral mutations conferring resistance to therapy. More conservative estimates of prevalence of resistance would therefore be 18% and 21% of M12 and M24 patients, respectively, on treatment.
Estimates of resistance and virological and immunological failure might also have been underestimated if some of the deaths or patients who were lost to follow-up died as a result of treatment failure and resistance. However, most deaths and losses to follow-up occurred within the first months of therapy when the likelihood of treatment failure was lower. Conversely, because our study is based on outcome data collected at a single time point and no confirmation of laboratory data was obtained in the following weeks, we might have slightly overestimated rates of virological and immunological failure due to measurement errors or to individual temporal variability in biological markers (e.g. HIV viral blips). As in previous studies conducted in patients using primarily d4T/3TC/NVP, the most commonly observed resistance mutations were those associated with the use of NNRTIs (K103N) and 3TC (M184V) [7, 10, 38], and none of the isolated viruses had mutations conferring resistance to the new NNRTIs (etravirine).
Patients on ART for 2 years were twice as likely as those on 1 year treatment to develop virological failure. Patients diagnosed with TB after ART initiation were also more likely to fail therapy, suggesting that patients' virological response could be impaired when rifampicin-based anti-TB treatment is initiated while a patient is under ART. The potential interaction between NVP and rifampicin in HIV-infected patients treated for TB has been reported in previous studies [39–43]. This can result in subtherapeutic NVP plasma concentrations  and favor the development of resistance to this drug. Alternatively, pill burden associated with concomitant treatments could lead to decreased adherence to one or both regimens.
Early identification of patients with virological failure and switch to second-line therapy is important to prevent the accumulation of resistance that might compromise the effectiveness of subsequent lines of treatment. In this program where virological monitoring of patients is not routinely performed, patients presenting general clinical symptoms, those with a weight loss below the value observed at therapy initiation, or those reporting poor adherence were more likely to have virological failure. The association between high VL levels and subtherapeutic NNRTI concentrations has been previously reported in Malawi  and Cambodia , highlighting the importance of maintaining and reinforcing patients' adherence over time. In our study, patients severely immunosuppressed at the time of the survey were not more likely to have virological failure, although this might change if patients remain in a failing regimen for longer periods of time .
The pharmacological assessment showed increased plasmatic levels of NNRTI in 23% of patients on NVP-based therapy and 39% of those on EFV. These figures are higher than what has been described in Western countries, and are likely to be related to the lower weight of our patients at ART initiation  compared with European patients [45, 46]. In contrast, subtherapeutic levels of NVP and EFV were found in 15% and 2% of the patients, respectively. As in previous studies, the risk of having low NNRTI concentrations in plasma was nearly 4 times higher in patients with poor self-reported adherence [44, 47]. We also found a higher risk of subtherapeutic ARV levels in patients who reported digestive disorders at the time of the survey. Vomiting, malabsorption, or misclassification bias (e.g. if patients who reported symptoms were also more likely to report poor adherence) could contribute to explain this finding.