OBJECTIVE To compare saquinavir + ritonavir and saquinavir + nelfinavir with nucleoside recycling in patients with multiple failures of highly active antiretroviral therapy (HAART). METHODS This was a prospective, multicenter, randomized open trial. Inclusion criteria were the following: consent, age > 18, previous protease inhibitor (PI) exposure > 6 months, unchanged HAART > 3 months, and viral load > 3 log. The treatments compared were ritonavir 200 mg bid + saquinavir 600 mg bid (Rito-Saq), and nelfinavir 1,000 mg bid + saquinavir 600 mg bid (Nelf-Saq). Nucleoside analogues were recycled, and nonnucleoside inhibitors were not permitted. Trough levels of the three drugs were measured by high-performance liquid chromatography at the month 3 visit. After the study had been completed, genotyping analysis was done on the first serum at entry. RESULTS The study was interrupted due to the availability of new anti-HIV drugs. A random sample of 31 (16 Rito-Saq and 15 Nelf-Saq) patients was divided into two groups, which were comparable in terms of demographic data and previous history of HIV infection. Mean CD4 cell count and plasma viral load (pVL) were 316 +/- 169 and 3.89 +/- 0.87 for Rito-Saq and 448 +/- 238 and 3.85 +/- 0.32 for Nelf-Saq. Previous duration of PI exposure was 31 months for both groups. The mean number of protease gene mutations was 3.8 (range, 2-7) and 4.4 (range, 2-9), respectively. On intention-to-treat (ITT) analysis at month 6, pVL stabilization or decrease >/= 0.5 log was observed in 18 patients (58%): 10 for Rito-Saq and 8 for Nelf-Saq. In a multivariate logistic regression analysis, virological success at month 3 was inversely correlated to baseline viral load (R = 0.14; 95% CI 0.03-2.9; p =.01); and at month 6, virological success was inversely associated to the number of mutations in the protease gene (R = 2.2; 95% CI 0.73-6.53; p =.06). CONCLUSIONS Nelf-Saq and Rito-Saq combinations can be proposed in case of multiple HAART failures. The fact that the virological response was inversely correlated to baseline viral load makes the case for an early switch after a HAART failure.