Clinical trials have shown that highly active antiretroviral treatment (HAART) is able to reduce HIV plasma viral loads to undetectable in 70% to 90% of patients and to increase CD4 cell counts. HAART in community settings (i.e., nonclinical trial situations) is reported to be much less effective. STUDY DESIGN: Observational study. PURPOSE: The aim of our study was to evaluate the effectiveness of protease inhibitor (PI)-based HAART in the Luxembourg HIV cohort after 36 months of treatment in previously treated and untreated patients. The secondary aim was to identify surrogate markers associated with long-term virologic and immunologic outcomes. PATIENTS AND METHOD: Seventy-three PI-naive patients, who started on HAART, combining one PI and two nucleoside reverse transcriptase inhibitors (NRTIs),with a follow-up of 3 years, were evaluated with plasma viral load and CD4 cell counts every 3 months and were analyzed retrospectively. Patients who had been treated previously with NRTI (n = 48) were at a more advanced stage of disease. RESULTS: Overall, there was a mean decrease in viral load compared to baseline of -1.89 log RNA copies/mL (SD = 1.40) that persisted at month 36. Sixty-two percent (62%) of patients reached an undetectable viral load (i.e., below 500 copies/mL): 82% and 53% of NRTI-naive and NRTI-experienced patients, respectively (p =.013). CD4 cell counts increased progressively in both groups with a sustained effect (mean increase of 146 cells/mL +/- 241) at month 36. NRTI-naive patients had a mean increase of 257 cells/mL (SD = 305), in contrast to experienced patients who had an increase of 108 cells/mL (SD = 206) at 3 years. Proportions of patients with a CD4 count under 200 cells/mL fell after 3 years for NRTI-naive (from 66% to 43%) and for experienced patients (from 32% to 13%).Predictors of short duration of viral load response were in decreasing order of importance: clinical AIDS, the use of saquinavir hard gel formulation as initial PI, and the number of NRTIs previously used. Viral load response was the only significant predictor of CD4 changes. CONCLUSION: In a community setting, effectiveness of PI-based HAART at 3 years is still achieved for most patients. NRTI-experienced patients have a good long-term response rate even if it is lower than NRTI-naive patients. A poor treatment response is associated with a more advanced stage of disease before HAART is introduced.