Dual vs Single Protease Inhibitor Therapy Following Antiretroviral Treatment Failure: Title and subTitle BreakA Randomized Trial FREE

Use of potent combination antiretroviral therapy has been associated with marked reductions in morbidity and mortality related to human immunodeficiency virus type 1 (HIV-1) infection in the developed world.^{1 - 3} Potent regimens are usually defined as those that contain at least 3 antiretroviral agents, one of which is either a protease inhibitor (PI) or, more recently, a nonnucleoside reverse transcriptase inhibitor (NNRTI)^{4 - 6} or, in treatment-naive subjects, a triple-nucleoside combination that includes abacavir.⁷ The PI-containing combination therapy quickly became a standard of care after widespread PI availability in 1996.^{4 ,8 - 9} Despite the improved outcomes, problems associated with combination antiretroviral therapy such as failure and toxicity are apparent.^{10 - 13} Although virologic failure can be caused by suboptimal drug exposure due to poor adherence or unfavorable pharmacokinetics, drug resistance is of concern because of resulting limits on therapeutic options.^{14 - 15}

Studies have shown that response to alternative or "salvage" antiretroviral combination regimens in the setting of virologic failure is limited^{16 - 20} and suggest that higher response rates are seen when 2 or more active agents are used in the salvage regimen and when a therapy switch is made at a lower viral load.^{17 ,19 - 20} Findings from short-term studies suggest that response rates may be improved if choice of agent is guided by drug resistance testing.^{15 ,21 - 25} A currently accepted approach to management of virologic failure in patients already receiving a PI-containing regimen is to include 2 new PIs as part of the salvage regimen.^{4 ,6} However, efficacy of this strategy has not been rigorously tested in a prospective clinical trial. There are 2 rationales for addition of a second PI: to provide pharmacokinetic enhancement of the first drug as with use of low-dose ritonavir to increase drug exposure to other agents^{26 - 32} ; and to provide additional drug exposure with full dosing of both PIs³³ (rationale for the present study). Our goal was to assess whether addition of a second PI to a new 4-drug class regimen including amprenavir would improve virologic response in patients failing a PI-containing regimen. Amprenavir^{34 - 35} has a resistance profile different from the 4 previously approved PIs; viral strains resistant to the latter agents may stay susceptible to amprenavir.³⁶

The AIDS (acquired immunodeficiency syndrome) Clinical Trials Group (ACTG) 398 study was a randomized, double-blind, placebo-controlled study of saquinavir, indinavir, or nelfinavir added as a second PI to the 4-drug class regimen of amprenavir, abacavir (a nucleoside reverse transcriptase inhibitor [NRTI]), efavirenz (an NNRTI), and adefovir dipivoxil (a nucleotide reverse transcriptase inhibitor [NtRTI]) in patients with virologic failure defined by a viral load above 1000 copies/mL while receiving saquinavir, nelfinavir, indinavir, or ritonavir. Randomization was stratified by prior PI use. The study was designed to enroll 460 patients with no more than 33% having had prior NNRTI use. During the trial, the latter was modified to permit a maximum enrollment of 50% NNRTI-experienced patients to better reflect the antiretroviral experience profile of patients failing PI therapy in the United States. The primary objectives were to (1) compare the proportion of patients having a viral load below 200 copies/mL at week 24 across study arms as determined by an ultrasensitive reverse transcriptase-polymerase chain reaction HIV-1 RNA assay (Roche Molecular Systems, Branchburg, NJ); and (2) compare safety and tolerance of the regimens across study arms using the ACTG adverse event grading scheme.³⁷ The stringent primary virologic end point of a viral load of 200 copies/mL was chosen because virus suppression to this level is associated with prevention of emergence of drug resistance, virologic response durability, and improved clinical outcome.^{38 - 41}

Secondary objectives were to (1) compare proportion of patients with virologic failure across study arms (defined as a confirmed increase in viral load above baseline occurring before week 24, indicative of lack of virologic response,⁶ failure to achieve a minimum 0.5 log₁₀ copies/mL decrease in viral load by week 8,^{6 ,40 - 42} a confirmed 1.0 log₁₀ copies/mL increase in viral load above nadir value before week 24, a confirmed increase in viral load at or above 200 copies/mL after a confirmed reduction to below 200 copies/mL before week 24, or viral load at or above 200 copies/mL at week 24)^{40 ,42} ; (2) examine influence of prior NNRTI and PI use on virologic outcome at week 24; (3) compare changes from baseline to weeks 24 and 48 in viral load and CD4 cell count; (4) compare drug concentration areas under the curve (AUCs) of amprenavir, efavirenz, adefovir, saquinavir, indinavir, and nelfinavir (selected for study because of potential for drug interactions^{43 - 44} in a subset of patients; and (5) explore the relationship between baseline HIV-1 drug susceptibility and virologic response, and changes in drug susceptibility at time of virologic failure.

Patients were recruited from 31 AIDS Clinical Trials Units. Inclusion criteria were: 13 years of age or older, laboratory documentation of HIV-1 infection, prior exposure to a maximum of 3 PIs from among saquinavir, ritonavir, indinavir, and/or nelfinavir for a cumulative period of PI therapy of at least 16 weeks, receiving the failing PI-containing regimen at time of screening¹⁵ (defined as a viral load >1000 copies/mL; a higher level than that reported for transient virologic "blips"⁴⁵ ), having a Karnofsky performance status of 70 or more, and having certain laboratory parameter levels (hemoglobin ≥9.1 g/dL for men and ≥8.9 g/dL for women, absolute neutrophil count ≥850/µL, platelet count ≥65 000/µL, aspartate and alanine aminotransferase levels ≤5 times upper limit of normal, serum amylase ≤1.5 times upper limit of normal, serum creatinine ≤1.5 times upper limit of normal, normal serum phosphate and bicarbonate, no glycosuria, and ≤grade 1 proteinuria). Patients were required to be treatment-naive to amprenavir, abacavir, and adefovir dipivoxil. Institutional review boards of the participating institutions approved the study and each patient gave written informed consent.

Patients received open-label amprenavir 1200 mg twice daily, abacavir 300 mg twice daily, efavirenz 600 mg once daily, and adefovir dipivoxil 60 mg once daily (with L-carnitine supplementation 500 mg once daily) and were randomized to receive saquinavir (soft-gel capsule formulation) 1600 mg twice daily (saquinavir arm), indinavir 1200 mg twice daily (indinavir arm), nelfinavir 1250 mg twice daily (nelfinavir arm), or placebo matched for saquinavir, indinavir, or nelfinavir (placebo arm). Selective randomization was carried out to prevent, as possible, patients from receiving PIs to which they had been exposed (ie, randomized to a PI arm containing PI[s] to which they were naive) (Figure 1). Patients with prior use of 3 PIs were randomized to any of the 4 arms because no arm contained a second PI to which they had not been exposed. Prior ritonavir and indinavir use was considered use of a single PI because of overlapping resistance profiles of these agents.^{46 - 48}

Patients were scheduled for follow-up visits at weeks 2, 4, 8, 16, and every 8 weeks until the last patient completed 48 weeks on study. Follow-up consisted of clinical assessments, routine laboratory monitoring, and viral load measurement, the latter being measured twice at baseline. The CD4 cell counts were measured twice at baseline and at weeks 4, 8, 16, and every 8 weeks until the study end. An interim review committee examined study results on April 26, 1999, after 40 virologic failure end points had occurred and again on August 24, 1999, at study team request. The recommendation after each review was to continue the study unchanged. The primary analysis of the study was prespecified in the protocol document and conducted when the last patient reached 24 weeks on study.

A pharmacokinetics substudy was completed at day 14 in 46 patients accrued when the first 187 patients were enrolled into the main study from 14 sites. The 46 patients differed from the overall population (n = 481) with a higher median CD4 cell count (303/µL vs 202/µL) and a higher proportion of white, non-Hispanic enrollees (76% vs 58%). Patients were admitted to the general clinical research centers of the participating institutions. Regular scheduled doses of study medications were given under direct observation and blood samples for assay of PIs, efavirenz, and adefovir dipivoxil levels were drawn prior to dosing and at 1, 2, 4, 6, 8, and 12 hours after dosing. Amprenavir levels were assayed at GlaxoSmithKline; saquinavir, indinavir, and nelfinavir levels at Stanford University School of Medicine; efavirenz levels at State University of New York at Buffalo; and adefovir dipivoxil levels at Harris Laboratories (Lincoln, Neb). Levels of the nelfinavir M8 metabolite, which contributes to its overall activity, were not assayed, but omission of these data does not affect interpretation of the pharmacokinetic or main studies. The AUCs were calculated using the trapezoidal rule assuming steady-state and without infinity extrapolation.

To determine phenotypic HIV-1 drug susceptibilities retrospectively using baseline plasma samples, 200 patients were randomly selected from the entire study population using a random number generator. The 200 patients did not differ from the overall study population in baseline demographic criteria. Baseline phenotypic drug susceptibilities were assessed for all Food and Drug Administration (FDA)–approved antiretroviral agents but only data from drugs received in the study are presented herein. Of the 200 patients, 139 were chosen for study because of receiving study treatment for at least 8 weeks and having an available sample. The 139 patients did not differ from the overall study group demographically. Of the 139, 59 had virologic failure and phenotypic susceptibility testing was done on plasma samples at time of failure. Plasma samples from baseline and at time of virologic failure were tested using a recombinant virus assay (PhenoSense, ViroLogic Inc, South San Francisco, Calif). For this study, virus having a 50% inhibitory concentration (IC₅₀) for an individual drug that was more than 2.5-fold higher than the IC₅₀ for the reference virus (HIV-1_NL4-3) was considered to have reduced drug susceptibility; virus that showed an IC₅₀ that was 2.5-fold or less higher than the reference strain was considered sensitive.⁴⁹ Virus having an IC₅₀ 0.4-fold or less than that of the reference virus for a drug was considered to be hypersusceptible to that drug, based on assay ability to differentiate this level of susceptibility as different from that of the reference virus and because it has been used in prior studies.^{50 - 51} In an additional analysis, an IC₅₀ that was more than 10-fold higher than that of the reference virus was used to define drug resistance. Use of the 10-fold cutoff is arbitrary; it accounts for the known variability (as much as 10-fold in clinical isolates⁵² )of wild-type virus in susceptibility to NNRTIs due to natural polymorphisms in reverse transcriptase. The 10-fold cutoff was used to explore the utility of the phenotypic susceptibility score in predicting virologic outcome with various cutoffs. Ten-fold is a level for which it is known that reduced susceptibility exists.⁴⁹

The intent-to-treat subject allocation was used for the primary analysis. If patients discontinued a component of the original study treatment, they were allowed to continue the rest of the regimen (in combination with FDA-approved agents), with study team approval. Patients stopping any component of the study regimen were characterized as discontinuing study treatment for purposes of the primary analysis but study follow-up continued. The nonstudy, FDA-approved drugs used by patients within 40 days of discontinuing study treatment included zidovudine, stavudine, lamivudine, didanosine, nevirapine, delavirdine, ritonavir, and hydroxyurea. Data from patients who dropped out of the study were censored at the last recorded clinic visit. Data from the first 24 weeks of follow-up were used in the primary analysis. The 48-week analysis also included adefovir-related nephrotoxicity and AIDS-defining events or deaths. All viral load assays were done at The Johns Hopkins University Laboratory (Baltimore, Md). Lower and upper limits of quantification were 200 and 75 000 copies/mL, respectively; samples with more than 75 000 copies/mL were retested after dilution. Viral load values were log₁₀ transformed for analysis. In analyzing the proportion below 200 copies/mL at week 24, missing viral load data at week 24 were considered to be 200 copies/mL or more. Missing data due to death or study discontinuation were counted as virologic failures.

Baseline characteristics were compared for differences among treatment arms using the Mantel-Haenszel exact test for categorical data and F test of analysis of variance for continuous scale data. Primary analysis of proportion below 200 copies/mL at week 24 included pairwise comparisons of each saquinavir, indinavir, and nelfinavir arm with placebo arm and comparison of dual PI treatment (combined saquinavir, indinavir, and nelfinavir arms) with single PI treatment (placebo arm), using the Mantel-Haenszel exact test, stratified as protocol-prespecified for prior PI and prior NNRTI use, without adjustment for multiple comparisons. The study was designed to detect a 40% vs 20% success rate difference in each PI arm vs the placebo arm, with more than 80% power. With study sample sizes, there was 89%, 70%, and 93% power to detect a difference of 40% vs 20% in the saquinavir, indinavir, and nelfinavir vs placebo comparisons, respectively. The study was not designed to compare dual PI arms with each other.

Analysis of time to viral load below 200 copies/mL used a Cox proportional hazards model with NNRTI use, baseline viral load, and treatment arm as covariates, stratified by prior PI use. Analysis of proportion of virologic failures occurring through week 24 and of proportion of patients with viral load below 200 copies/mL at week 24 among those continuing assigned treatment included the 3 pairwise comparisons described above plus a comparison of combined dual PI arms vs the single PI arm (placebo), with stratification, as described for the primary analysis. Assumptions for Cox models used in analyses presented herein were tested and met with the exception of time to virologic suppression for the nelfinavir arm vs the placebo arm comparison, in which nelfinavir effect on suppression was time-dependent. Preliminary comparisons of virologic failure at week 48 (viral load of ≥200 copies/mL) used the stratified Mantel-Haenszel exact test. Analysis of change in log₁₀ viral load from baseline to week 24 used the Buckley-James distribution-free model, which accounts for censored data.⁵³ Change in mean CD4 cell count from baseline to week 24 was analyzed using the Wilcoxon rank test, stratified by prior PI and NNRTI use. Safety and tolerability of regimens (time to first grade 3 or 4 sign or symptom, or first grade 3 or 4 laboratory abnormality) were analyzed using the log-rank test, stratified by prior PI and NNRTI use. Analysis of variance was used to assess effect of treatment arm on drug level AUCs for amprenavir, efavirenz, adefovir, saquinavir, indinavir, and nelfinavir.

Influence of baseline viral load on time to virologic response (viral load <200 copies/mL) and virologic failure used a Cox model and binary regression with log-log link, respectively. Influence of prior NNRTI and PI use on virologic failure was assessed using the likelihood ratio test for logistic regression to calculate risk ratios, adjusted for baseline covariates.

For patients having virologic failure and with phenotypic results available, change in susceptibility to study medications was calculated as ratio of IC₅₀ (fold IC₅₀) at failure divided by IC₅₀ (fold IC₅₀) at baseline. Geometric means of the fold-resistance ratios were tested to assess if these significantly differed from 1 using the 1-sample t test. For drug susceptibility analyses, baseline phenotypic sensitivity scores⁵⁴ were calculated for drug combinations (for 2.5-fold and 10-fold cutoff analyses, a score of 1 was given for each drug in the regimen to which the virus showed an IC₅₀ ≤2.5-fold or ≤10-fold above the reference virus IC₅₀; a score of 0 was given for a drug with an IC₅₀ >2.5-fold or >10-fold above). Relationship between baseline resistance and virologic failure was examined using odds ratios (ORs) generated by logistic regression. The ORs should not be interpreted as risk ratios when analyzed events involve common outcomes.⁵⁵ SAS version 6.12 (SAS Institute Inc, Cary, NC) and Splus version 3.4 (Splus, Insightful Corporation, Seattle, Wash) were used and level of significance was P≤.05.

The 481 patients were randomized between October 13, 1998, and April 14, 1999, with the study ending in April 2000. Randomization by prior PI use is given in Table 1 and baseline characteristics (well-balanced across study arms) of the study group in Table 2. Patient median age was 40 years; 87% were men; and 58% were white, 23%, black, and 15%, Hispanic. The proportions of patients exposed to 1, 2, or 3 prior PIs were 21%, 53%, and 26%, respectively; and 44% were NNRTI-experienced. Median baseline CD4 cell count and viral load were 202/µL and 51 601 copies/mL, respectively.

For the primary analysis, follow-up for virologic outcome and routine toxicities was 24 weeks (Figure 1). Secondary analyses for efficacy and toxicity extended through 48 weeks. At 24 weeks, 29 (6%) of 481 patients were off-study and 201 (42%) had discontinued study treatment (discontinuation of any drug in assigned regimen). Of the 201, 58 (29%) discontinued treatment for virologic failure and 143 (71%) for other reasons. Most of the latter were for protocol-defined or nonprotocol-defined toxicities (47 [33%] and 63 [44%], respectively). Most common reasons recorded by study sites for discontinuing treatment included gastrointestinal symptoms, drug reaction, and patient preference (eg, for low-grade toxicities). Abacavir was discontinued in 55 patients, amprenavir in 40, efavirenz in 39, adefovir dipivoxil in 25, and the second PI in 31 (saquinavir in 13, indinavir in 7, and nelfinavir in 11). Of the 201 patients, 131 (65%) continued at least 1 original drug in the assigned regimen after permanently discontinuing 1 or more regimen drugs. The combined proportion of patients off study and discontinuing treatment was lower in the nelfinavir (but not saquinavir or indinavir) vs placebo arm (P = .02). Pill burdens in the placebo arm were comparable with those in the dual PI arms.

At 24 weeks, 148 (31%) of the 481 patients had a viral load below 200 copies/mL. Proportions below 200 copies/mL were 34%, 36%, 34%, and 23% in the saquinavir, indinavir, nelfinavir, and placebo arms, respectively (Table 3). In the Cox model of time to virologic suppression, the time-dependent effect of nelfinavir vs placebo was significant (P = .003). At week 12, the risk ratio of suppression was 2.37 (95% confidence interval [CI], 1.40-4.02; P = .003) in favor of nelfinavir. As suppression is an early event, week 12 was a relevant time point to assess. The week 24 risk ratio was 3.58 (95% CI, 1.60-8.01). The proportion below 200 copies/mL at 24 weeks in dual PI arms combined (35%, Table 3) was higher than with placebo (23%; P = .002). In 2-way comparisons of proportions below 200 copies/mL between individual dual PI arms and placebo, only the nelfinavir vs placebo comparison was significant (P = .004). The significance of these comparisons is influenced by variable patient numbers in the strata of prior PI use within each arm (Table 1). The performance of the dual PI arms was similar as shown by the proportions with less than 200 copies/mL at week 24. The study was not designed to compare dual PI arms with each other; thus, these comparisons are not reported.

In pairwise comparisons of the time to viral load below 200 copies/mL between each dual PI arm and placebo, the nelfinavir arm had a shorter time to suppression than placebo (this was time-dependent [see above]), whereas the saquinavir and indinavir arms did not (risk ratio, 1.35; 95% CI, 0.91-2.01; P = .13; and 1.13; 95% CI, 0.68-1.87; P = .64, respectively).

The study design had a primary end point of 24 weeks with an extension to 48 weeks. In preliminary analyses, results at 48 weeks were concordant with the week 24 responses. Overall, 28% (133/481) of patients had a viral load below 200 copies/mL at 48 weeks. The proportions below 200 copies/mL in the saquinavir, indinavir, nelfinavir, and placebo arms were 34%, 33%, 26%, and 22%, respectively. A higher proportion in the combined dual PI arms had a viral load below 200 copies/mL than with placebo (30% vs 22%; P = .04).

Mean (SD) changes from baseline in viral load are shown in Figure 2A. At week 2, mean reduction in viral load was comparable across treatment arms (−1.37 to −1.43 log₁₀ copies/mL). At week 24, changes from baseline in viral load in saquinavir, indinavir, nelfinavir, and placebo arms were −1.88, −1.83, −1.64, and −1.02 log₁₀ copies/mL, respectively. The combined dual PI arms had greater viral load reduction vs placebo at week 24, controlling for prior NNRTI use (−2.29 vs −1.59 log₁₀ copies/mL for naive group, respectively, and −1.14 vs −0.44 log₁₀ copies/mL for experienced group, respectively; P<.001). At 48 weeks, mean viral load reduction from baseline in combined dual PI arms was also higher than with placebo, controlling for prior NNRTI use (−1.99 vs −1.51 log₁₀ copies/mL for naive group, respectively, and −1.31 vs −0.82 log₁₀ copies/mL for experienced group, respectively; P<.001).

Figure 2B shows mean changes from baseline in CD4 cell count. At week 24, changes from baseline in saquinavir, indinavir, nelfinavir, and placebo arms were + 53, + 9, + 33, and + 13 cells/µL, respectively. Combined dual PI arms had a greater increase in CD4 cells (+34/µL) vs placebo (+13/µL) at week 24 (P = .048). At 48 weeks, the increase in CD4 cells (+38/µL) in combined dual PI arms was not significantly higher than with placebo (+25/µL; P = .63).

An on-treatment analysis was performed because of the substantial proportion of patients discontinuing treatment and to assist with interpretation of intent-to-treat results. Through week 24, 249 patients (52%) stayed on study treatment. Of these, 129 (52%) had a viral load below 200 copies/mL at week 24. Proportions having a viral load below 200 copies/mL in the saquinavir, indinavir, nelfinavir, and placebo arms were 63%, 68%, 50%, and 40%, respectively (Table 3). Comparison of combined dual PI arms vs placebo showed that 57% vs 40% had a viral load below 200 copies/mL at 24 weeks. A higher proportion of the NNRTI-naive patients had virologic suppression at week 24 vs NNRTI-experienced patients (65% vs 31%, respectively). These on-treatment analyses are concordant with the primary, intent-to-treat analysis but should be interpreted cautiously because randomized groups are not being compared.

At the 48-week analysis timepoint, there were 11 patients with a total of 13 AIDS-defining illnesses, and 9 patients had died. The AIDS-defining illnesses were Pneumocystis carinii pneumonia (3), Kaposi sarcoma (3), cytomegalovirus retinitis (2) and esophagitis (1), Mycobacterium avium intracellulare bacteremia (1), cryptococcal meningitis (1), esophageal candidiasis (1), and progressive multifocal leukoencephalopathy (1). Of the 9 deaths, 6 were HIV-1–related. Of the 16 total patients with HIV-1–related clinical events, numbers and proportions occurring in the saquinavir, indinavir, nelfinavir, and placebo arms were 5 of 116 (4%), 2 of 69 (3%), 5 of 139 (4%), and 4 of 157 (3%), respectively (no significant difference).

Through 24 weeks, 19% (89/481) of patients had grade 3 or 4 signs or symptoms with gastrointestinal symptoms predominating. Proportion with grade 3 or 4 laboratory abnormalities was 28%, with hypertriglyceridemia and hypophosphatemia predominating. In saquinavir, indinavir, nelfinavir, and placebo arms, grade 3 or 4 signs/symptoms occurred in 23 (20%), 15 (22%), 27 (19%), and 24 (15%), respectively; for grade 3 or 4 laboratory abnormalities, 29 (25%), 16 (23%), 41 (29%), and 47 (30%), respectively. There were no significant differences across study arms regarding grade 3 or 4 adverse events. Analysis of specific toxicities at lower grades showed a 17% incidence of grade 2 or more rash, a 17% incidence of grade 2 or more central nervous system abnormalities, and a 5% incidence of abacavir-related hypersensitivity. No significant differences across study arms were seen for these toxicities. Through 48 weeks of follow-up, overall rates of grade 3 or 4 or signs/symptoms and laboratory abnormalities were 21% (102/481) and 35% (166/481), respectively, with no significant differences across study arms and with a profile similar to that of week 24. Adefovir-related nephrotoxicity occurred in 146 (30%) patients.

Mean AUCs for study drugs seen in the 46 patients at day 14 are shown in Table 4. The 12-hour AUCs for amprenavir were 8913, 19 055, 18 621, and 10 000 ng·h per milliliter in saquinavir (n = 12), indinavir (n = 10), nelfinavir (n = 12), and placebo (n = 12) arms, respectively. The difference across study arms was significant (P = .009) with AUCs for amprenavir in saquinavir and placebo arms being about 50% lower than those in indinavir and nelfinavir arms. In 2-way comparisons, amprenavir AUCs in saquinavir and indinavir arms were not different from that in placebo arm (P = .92 and P = .33, respectively), whereas in the nelfinavir arm it was higher than in placebo arm (P = .01). The AUCs for efavirenz and adefovir dipivoxil were comparable across study arms and consistent with prior data.^{56 - 57} The AUCs for saquinavir, indinavir, and nelfinavir were within the ranges reported when given alone or with nucleoside analogs only.^{58 - 60}

In all analyses, baseline viral load was significantly related to virologic outcome. A 10-fold higher baseline viral load and a baseline level above the median (vs below) were each associated with virologic failure (risk ratio, 1.45; 95% CI, 1.24-1.69; P<.001; risk ratio, 1.64; 95% CI, 1.30-2.05; P<.001, respectively). Prior NNRTI use was also correlated with virologic outcome. Of 270 NNRTI-naive patients, 115 (43%) had a viral load below 200 copies/mL at 24 weeks vs 33 (16%) of 211 NNRTI-experienced patients (P<.001 [stratified by prior PI use and treatment arm, and adjusted for baseline log₁₀ viral load]). In contrast, proportions of patients with single vs 2 or more prior PI use with viral load below 200 copies/mL at 24 weeks were not significantly different (37% vs 29%, respectively; P = .13 [adjusted for prior NNRTI use, treatment arm, and baseline viral load]).

Baseline phenotypic drug susceptibilities were determined on samples from 139 randomly selected patients at baseline (Figure 3). Of these 139 patients, 59 had virologic failure and susceptibilities were also assessed at time of virologic failure for them. At baseline, 50 of 139 (36%), 91 of 139 (66%), 37 of 139 (27%), and 26 of 139 (19%) patients had virus with IC₅₀ values of more than 2.5-fold higher than reference virus for amprenavir, abacavir, efavirenz, and adefovir dipivoxil, respectively (drugs common to all study arms). Thus, a substantial proportion had reduced susceptibility to 1 or more study regimen components at baseline. For efavirenz and other NNRTIs, the level of reduced drug susceptibility that may affect treatment response is likely more than 10-fold; lower levels of reduced susceptibility (2.5-fold to 10-fold) are caused by natural polymorphisms in reverse transcriptase that may not affect treatment response.^{52 ,61} The level of reduced susceptibility was more than 10-fold for amprenavir, abacavir, efavirenz, and adefovir dipivoxil in 3 (2%), 5 (4%), 23 (17%), and 1 (1%) of 139 patients studied, respectively.

To assess overall treatment susceptibility, a baseline phenotypic sensitivity score was calculated.⁵⁴ Using a 2.5-fold or less cutoff for sensitivity, the median score for the 139 patients was 3.0, meaning they received 3 drugs on average potentially active against baseline virus. The median score was 4.0 when a 10-fold or less cutoff was used. Only in the latter instance (cutoff 10-fold or less) was the score associated with outcome (ie, a higher phenotypic sensitivity was associated with a higher likelihood of virologic suppression) at 24 weeks (OR, 2.03; 95% CI, 1.17-3.54; P = .01) but was not significantly associated with virologic suppression at 48 weeks (OR, 1.34; 95% CI, 0.78-2.30; P = .29).

Baseline efavirenz susceptibility was also examined regarding virologic outcome because of the importance of prior NNRTI use on virologic response. Reduced baseline efavirenz susceptibility (>10-fold) was associated with lesser likelihood of virologic suppression at 24 weeks (OR, 0.28; 95% CI, 0.09-0.87; P = .03) and 48 weeks (OR, 0.16; 95% CI, 0.04-0.72; P = .02). In contrast, baseline efavirenz hypersusceptibility (an IC₅₀ of 0.4-fold or less compared with reference virus) was detected in 42 of 139 patients and associated with greater likelihood of virologic suppression at 24 weeks (OR, 3.49; 95% CI, 1.62-7.33; P = .001) and 48 weeks (OR, 5.66; 95% CI, 2.57-12.46; P<.001). This association was significant at 48 weeks even after controlling for prior NNRTI use (OR, 3.21; 95% CI, 1.36-7.59; P = .008).

Susceptibility change from baseline to time of virologic failure was also assessed for 59 (of the 139) patients with virologic failure by week 24; this showed that virologic failure by week 24 was mostly associated with decreasing susceptibility to efavirenz but not other study drugs (Table 5 and Figure 3). Overall, there was a 35.4-fold median decrease in susceptibility to efavirenz comparing baseline with time-of-failure IC₅₀ values (P<.001). For NNRTI-naive patients, median fold decrease in susceptibility was 65.1; for NNRTI-experienced patients, it was 26.5. Subset numbers are small and should be interpreted cautiously.

Effective management of patients with virologic failure receiving PI therapy is a major challenge today.^{4 ,6 ,14} Drug options are limited for patients failing their second or third regimen because of intraclass drug cross-resistance.¹⁵ Four drug classes are represented herein. Amprenavir may retain activity against clinical isolates from patients failing therapy with other approved PIs.¹⁵ Abacavir is the most potent NRTI approved to date.⁶² The HIV-1 isolates with limited genotypic evidence of resistance to other nucleoside analogs can retain abacavir susceptibility, making the drug useful for salvage purposes.⁶³ However, multiple nucleoside analog–associated resistance mutations (≥4) limit abacavir activity in patients with extensive prior nucleoside analog use.⁶⁴ Some HIV-1 isolates with reduced susceptibility to nevirapine and delavirdine may retain some in vitro susceptibility to efavirenz, although its in vivo activity in this case is limited.⁶⁵ Use of adefovir dipivoxil (the first NtRTI anti-HIV compound⁶⁶ ) in HIV disease has been halted because of limited antiretroviral activity and nephrotoxicity,⁶⁷ but the closely related tenofovir disoproxil fumarate has a more favorable safety and activity profile^{68 - 69} and is now FDA-approved.

The results of ACTG 398, the largest prospective, randomized trial of salvage therapy for PI failure reported to date, show that only 31% of patients had virologic suppression as defined by a viral load below 200 copies/mL at 24 weeks. This proportion is less than would be desired but this suppression level is a stringent test of a salvage regimen. The study group had advanced disease with a median baseline viral load of 51 601 copies/mL and a median baseline CD4 cell count of 202/µL, 79% had prior use of 2 or more PIs, and 44% were NNRTI-experienced. The study group as a whole averaged a −1.44 log₁₀ decline in viral load at 24 weeks, a change likely to confer clinical benefit.⁴²

A key finding is that combined dual PI-containing arms were superior to the single PI-containing arm (placebo) regarding the proportion of patients with virologic suppression (viral load <200 copies/mL at 24 weeks [35% vs 23%, respectively; P = .002]). This is the first study to our knowledge that shows superiority of dual vs single PI therapy in a prospective, placebo-controlled trial in treatment-experienced patients.

A second key finding is that the NNRTI-naive subgroup had a higher rate of virologic suppression (viral load <200 copies/mL) compared with the NNRTI-experienced subgroup (43% vs 16%; P<.001). This emphasizes the importance of having at least one (preferably more) potent agent, against which little or no viral cross-resistance is likely to exist, to use as the cornerstone of a salvage regimen.¹⁴

Surprisingly, the number of prior PIs used (1 vs ≥2) was not associated with a significant difference in virologic outcome at 24 weeks. Mean (SD) durations of prior PI use in the single and 2 or more PI-exposed groups were 82 (42) and 115 (49) weeks, respectively. One would expect that greater duration of prior PI use would influence virologic outcome but both groups were extensively exposed, possibly preventing detection of a difference.

Data showing that efavirenz lowers amprenavir levels by 30% or more became available after the ACTG 398 study was under way.¹⁷ Concern about this possibility and lack of data on possible 3-way drug interactions between amprenavir, a second PI, and efavirenz led to inclusion of a pharmacokinetic substudy in this trial. This substudy showed that 12-hour AUC for amprenavir in the placebo arm was about 50% of that in indinavir and nelfinavir arms, and at the lower end of the range described for prior pharmacokinetic studies of amprenavir in the absence of efavirenz.^{17 ,34 - 35} Interestingly, amprenavir AUC in the saquinavir arm was also about 50% of that in indinavir and nelfinavir arms and comparable to that in the placebo arm. Efavirenz lowers amprenavir AUC, presumably by inducing cytochrome P-450 (CYP) 3A4.^{43 ,70} Saquinavir, indinavir, and nelfinavir all inhibit CYP3A4 and amprenavir metabolism in vitro,^{43 ,70 - 71} and would therefore oppose the efavirenz effect. Saquinavir, however, is the weakest inhibitor of CYP3A4; thus, it may not have opposed the efavirenz effect on amprenavir levels to the same extent as indinavir and nelfinavir in vivo.^{43 ,72} Although lower amprenavir levels in the placebo arm may have contributed to virologic failure in that arm, lower amprenavir levels do not fully explain the overall study results as illustrated by the virologic failure rates in the 4 arms of the pharmacokinetics substudy (Table 4).

The extent of resistance to the assigned treatment regimen, as defined by the baseline phenotypic sensitivity score using a 10-fold cutoff, was predictive of suppression at 24 weeks. Hypersusceptibility to NNRTIs has been reported in HIV-1 isolates from patients with nucleoside analog–associated resistance mutations and limited data sets have suggested it is associated with better response to NNRTI-containing regimens.^{50 - 51} Data herein, however, are the first to firmly establish the importance of efavirenz hypersusceptibility in long-term virologic response to salvage regimens.

The rates of virologic success seen in this study are comparable to or better than those seen with other approaches to salvage therapy, including multidrug rescue therapy (a regimen containing at least 6 agents⁷³ ), structured treatment interruptions, and the use of nucleoside-adjunctive agents such as hydroxyurea or mycophenylate mofetil.^{74 - 77} Salvage therapy results from NNRTI-naive patients receiving lopinavir, a PI with high serum levels in the presence of low-dose ritonavir, suggest successful viral suppression is possible with prior PI use if drug levels exceeding the IC₅₀ of resistant strains can be achieved.^{32 ,78} The findings herein have shown progress and limitations in the challenging therapeutic arena of providing effective salvage therapy in the setting of virologic failure with PIs.