But I doubted that. While not ready to make such a blanket declaration myself, the case supporting the lecturer's conclusion has been strengthening. Data from three key studies, DMP-006, the Atlantic study and CNA 3005, indicate that the arguments in favor of non-protease inhibitor-based regimens can no longer be lightly dismissed. However, what may be true for one PI-sparing option may not hold true for all of the others, and different regimens need to be evaluated separately.
Shortly after the first protease inhibitors were approved, some clinicians began to see some downsides to their use. Before anyone was worried about their long term metabolic side effects (see last issue), it was apparent that the triple drug regimens were extremely complicated, with a high pill burden. At that time, with the exception of twice daily ritonavir (with a formulation so nauseating that it has never been widely accepted at full dose), the protease inhibitors were administered three times a day, with dietary restrictions. And although background therapy was generally twice a day, some pills (such as ddI and indinavir) could not be taken at the same time or with food, which meant that patients might have to take their medication five times a day. There was a very valid concern that such complicated dosing would lead to poor adherence and treatment failure. Meanwhile, some clinicians had a perception that the unprecedented potency of the protease inhibitors was overkill in those patients with lower viral loads. Wouldn't a simpler, gentler regimen make more sense at least for these patients, who probably were not yet used to taking pills chronically?
With the approval of Boehringer-Ingleheim's non-nucleoside reverse transcriptase inhibitor (NNRTI) nevirapine (Viramune), a simpler option became available, but one that was regarded with suspicion. Nevirapine had been around for a long time, and its development had been plagued with problems. While initially potent, resistance to the drug was quick to develop when used in suboptimal combinations. Many had given up on the drug, but when the principles of combination therapy were better understood and nevirapine was used with two other drugs in patients who had never been on antiretroviral therapy, there were sustained reductions in viral load. Still, nevirapine's reputation was tarnished and it did not have the allure of a "new drug." Few thought that it could possibly be as effective as a protease inhibitor. But some doctors, realizing that nevirapine-based regimens were much more convenient, began using them up-front as sort of a "training wheels" approach to therapy in patients not in immediate danger of clinical progression and not psychologically prepared for the unreasonably complicated protease inhibitor-based regimens. But there was a distinct impression that effectiveness was being exchanged for convenience. Furthermore, there were simply more data available for protease inhibitors with longer follow-up (particularly oft-cited was Merck 035, though it contained less than a few dozen patients after the first couple years). Nevirapine still has only conditional approval status and was never placed on the preferred first line list in the HHS treatment guidelines, but was instead listed only as an alternative option. New data may force both reconsideration and a reclassification of the drug (see Atlantic below), but the biggest boost to this drug's reputation resulted from the development of a competing NNRTI.
The idea that NNRTIs were inferior to protease inhibitors persisted until efavirenz. What changed? Well, for one thing, DuPont never missed a beat in the clinical development of this drug. Informed by Boehringer-Ingleheim's experience with nevirapine, DuPont knew not to repeat the same mistakes. Efavirenz was never tested as a monotherapy for more than two weeks, and thus the rapid rebounds in viral load experience observed with all other NNRTIs were never allowed to occur and besmirch this drug's reputation. Furthermore, DuPont aggressively tested the combination in optimal regimens. For instance, when the company discovered that efavirenz lowered indinavir (Crixivan) concentrations (AUC) by approximately a quarter, they upped the indinavir dose from 800 to 1000 mg TID in the studies combining those two drugs. Many companies and PK experts would not recommend dose adjustment with such an interaction and have noted that it might even backfire by providing a disincentive to adherence. But DuPont has always erred on the side of aggressive dosing, pushing for the maximum drug levels attainable for all drugs combined with efavirenz -- after all, suboptimal blood levels of one of the agents in the regimen might make efavirenz look bad. The goal: no cracks in efavirenz' armor -- and no bad press.
Of course, bad press was just what DuPont got when they priced efavirenz similarly to protease inhibitors, despite considerable pressure from activists, but ironically this press may have played to DuPont's advantage. The company stuck to its guns, arguing that since efavirenz was as effective as a protease inhibitor, it should cost about as much. One can't help but wonder whether this was, in and of itself, a marketing ploy. The pricing debate itself underscored DuPont's message that efavirenz was as "valuable" as a protease inhibitor.
But the foundation for DuPont's claim that efavirenz was equivalent to a protease inhibitor was DMP-006, the pivotal three arm study that secured efavirenz' approval. Treatment Issues has reported on this study before (see November 1998 and February 1999), most recently last February when the study had 48 weeks of follow-up. In short, the study found that efavirenz/AZT/3TC was at least as good as, if not better than, the gold standard indinavir/AZT/3TC. In fact, in the intent-to-treat (ITT) analysis, the efavirenz arm was significantly better. And this was regardless of whether patients entered the study with high viral loads (over 100,000 copies/ml) or not.
In other contexts, similar data might have led to a shift in the standard of care, but initially the 006 data were treated with a certain incredulity. First and foremost, the differences between the triple drug regimens were primarily due to a higher dropout rate on indinavir -- in the early as-treated analysis, the differences did not quite reach statistical significance. And while ITT analyses are generally thought to be more rigorous and to reflect the real-world usage of a drug, this may not be the case when the study was conducted in an open-label and unblinded fashion (see On-Treatment Versus Intent-to-Treat below). Bias can creep in, which, as the FDA put it in the efavirenz package insert, made it "difficult to assess the relative efficacy of the treatment arms given the disproportional discontinuations in an open-label study." For example, a patient who simply did not like indinavir's dietary restrictions could easily dropout and go on some other regimen (especially when nelfinavir became available). Finally, DuPont is conducting the study. There were suspicions among some activists and clinicians that the company, knowing that efavirenz' nervous system side effects could lead to study dropouts, made certain that study participants on efavirenz got more encouragement to stay in the study than would be seen in a real-world clinical setting. Even Dr. Rob Murphy, who presented some of the original results from the study, conceded at a community meeting over a year ago that they "had to do a lot of hand-holding during the first few weeks" to keep patients on study. In and of itself, this was far from unethical, but it could nevertheless have affected the results.
But longer follow-up and a larger sample size could help resolve some of these controversies. This fall Dr. Schlomo Staszewski from Germany's Goethe University presented the week 72 data for the initial study cohort of 450 and, for the first time, data on the durability of viral suppression in an expanded cohort of 1266 patients.
Once again, in the intent-to-treat analysis, the triple therapy efavirenz arm was superior to indinavir/AZT/3TC (60% vs. 40% below 50 copies/ml), while the differences between both indinavir-containing arms (the third arm was efavirenz/indinavir) were insignificant. DuPont prefers to present the ITT analysis because that is where the drug looks best, but even in the observed therapy analysis (shown in Figure 1), efavirenz/AZT/3TC was statistically better at weeks 48 and 60 (p<0.05), though the p-value at week 72 was not quite significant.
Observed Data (72 Weeks)
Data from the extended cohort reinforced the findings on efavirenz' durability. At baseline, over 80% of these patients were treatment naive and mean CD4 cell counts ranged from 332 to 350 (depending upon arm), with viral loads between 57,500 and 63,100. Kaplan-Meier curves on the durability of response were presented (see Figure 2). Loss of response was defined as 1) any dropout, AIDS-related event or virologic failure (two consecutive viral load results over 400 copies/ml) or as 2) virologic breakthrough alone. By either definition, the durability of response on the efavirenz/AZT/3TC arm was better both for the overall cohort and for those patients who entered the study with a viral load of more than 100,000 copies/ml.
This Kaplan-Meier Curve shows the duration of response for each arm. Failure is defined only as a rebound in viral load. Even when failures due to adverse events (and there were more on the IDV-triple arm) are excluded from the analysis, the best response is on efavirenz triple therapy.
Granted, this is still a DuPont run study, but an NIH trial comparing efavirenz and nelfinavir corroborated the findings that this NNRTI may be at least as effective as a protease inhibitor (see Figure 3). ACTG 364 was a double-blind and placebo-controlled study (making it more rigorous than DuPont 006) in 196 nucleoside analog-experienced patients (median experience of approximately six years) rolled over from earlier ACTG studies. However, subjects were naive to protease inhibitors, and NNRTIs. All received a new nucleoside analog background regimen (which was open-label) and were randomized to receive 1) efavirenz, 2) nelfinavir or 3) efavirenz/nelfinavir. At baseline, the mean CD4 cell count was 389 cells and the viral load was 8,130 copies/ml. Not surprisingly, the virologic response was best on the four drug arm (see Table 1), but the efavirenz arm was also significantly better than the nelfinavir arm. And in this study, there was no major difference between the study arms in the rate of premature study discontinuation.
Shows the proportion of patients at each time point who have HIV-RNA <500 copies/ml, are on their original study medication, and who have not experienced an AIDS defining event.
NFV = nelfinavir
Table 1: ACTG 364
|Outcome||EFV + NFV + NRTIs
N = 65
|EFV + NFV + NRTIs
N = 65
|NFV + NRTIs
N = 66
|HIV RNA <500 copies/ml||71%||60%||33%|
|HIV RNA<50 copies/ml||67%||44%||22%|
|HIV RNA >500 copies/ml
CDC Category C event
|Discontinuations for Adverse Events||3%||3%||5%|
|Discontinuations for Other Reasons||8%||0%||0%|
Table 2: The Atlantic Study Viral Load Response at Week 48 (N = 181)
|Intent-to-Treat Analysis||As-Treated Analysis|
Of course, patients in this study had virus that could have evolved cross-resistance across the class of nucleoside analogs, or at least high level resistance to AZT. Thus, it isn't surprising that the patients on three-drug therapy (which may have been tantamount to monotherapy in this case) did not do as well as those on quadruple therapy. What is surprising -- given the poor history of NNRTIs in suboptimal settings -- is that efavirenz was so much better than nelfinavir. It is possible that nelfinavir, which the virus can also evade by just making one mutation to its protease enzyme, is a weaker protease inhibitor than indinavir (they have never been in a randomized comparison -- although data from observational studies suggest they may be similar in the clinic; see below). It is also possible that the nelfinavir arm was more vulnerable to poor adherence than efavirenz. While the study was double blinded (which would normally eliminate an adherence bias), differences in the drug's side effects (diarrhea vs. neurocognitive) could have effectively unblinded the study to its participants. It may have been easier for those randomized to active efavirenz to remember to take their night-time dose (in order to minimize its side effects) than for those on nelfinavir to remember all three daily doses. Moreover, in the nelfinavir BID versus TID studies, the BID arm performed better, presumably because of better adherence. So in a nelfinavir BID vs. efavirenz competition, nelfinavir might have fared better.
Another argument in favor of using protease inhibitors as first-line therapy is that there is little data on the effect of NNRTI-anchored therapy in lymphoid tissue. Last year, a small study (Ruiz et al, AIDS 13(1):F1-8) suggested that patients with undetectable viral loads were much more likely to have normal looking lymph nodes free of viral activity if they were on protease inhibitor regimens than if they were simply on dual nucleoside therapy. The study had obvious flaws. Even if the patients were undetectable, dual nucleoside therapy has repeatedly been shown less effective than HAART. Despite this, some have tried to use it as evidence that a two-pronged antiviral attack (going after the reverse transcriptase and protease enzymes) was better than merely attacking reverse transcriptase. Nevertheless, data on the effect of NNRTIs in the lymphoid compartment were lacking.
Now data from a small but more rigorous NIH-sponsored study demonstrate that patients with an undetectable viral load by the ultrasensitive assay have also cleared most of the virus in the lymphoid tissue, regardless of whether they were on a PI- or an NNRTI-based regimen (ICAAC LB-15). The study was in 13 patients with less than 50 copies/ml. Nine, with baseline viral loads of approximately 384,375, were treatment naive when starting on efavirenz/d4T/3TC. Four others with much lower baseline viral loads (about 49,000) started on a protease inhibitor and two nucleosides. After about eight months of therapy, patients had lymph node biopsies. Seventy-five percent of the patients in each group had undetectable virus in their lymph nodes (NASBA assay <100 copies/ml). In situ hybridization could detect the production of HIV particles by only one cell in the germinal centers of only one patient from each group, and replication competent HIV in latently infected cells could only be assayed in five of eight efavirenz and two of three indinavir patients. Essentially, there was no difference.
All the buzz about efavirenz brought more attention to nevirapine, particularly when efavirenz was priced so much higher than the older non-nucleoside. In fact, even before efavirenz hit the market, nevirapine's sales began to spike up. It was only a matter of time until someone began comparing it directly to a protease inhibitor as well.
At ICAAC and in Lisbon, 48 week data was available for 235 of the 298 patients enrolled in the Atlantic Study, which compared two different classes of PI-sparing regimens (NNRTI-based or triple nucleoside analog) to an indinavir-based regimen. Participants were randomized to receive on an open-label basis either 1) indinavir, 2) nevirapine or 3) 3TC, each in combination with ddI/d4T. Arms were well matched at baseline, with a mean viral load of 4.36 log copies/ml (approximately 24,000) and a mean CD4 count of 447 cells.
Viral load data were available for approximately 181 patients (see Table 1, above, and Figure 4) and responses were similar for each arm. CD4 cell responses were also comparable, and no unexpected adverse events were observed.
The percentage of patients with undetectable viral loads is roughly the same for all three arms.
NVP = nevirapine
At this point, each arm would appear to be roughly equivalent; however, upon closer inspection, some differences emerge (see Figure 5). In a post hoc analysis that stratified patients by their entry viral load (below or above 51,286 copies/ml), a difference was noted in the triple nucleoside arm: fewer of those with the higher baseline viral loads had viral load suppression to below 50 copies/ml. This difference fell just shy of statistical significance (p=0.08).
Shows the proportion of patients in each study arm with low and high baseline viral loads who achieve HIV RNA <50 copies/ml. Note that fewer patients with high baseline viral loads are below the limit of detection on the triple nucleoside analog arm.
Some have interpreted these findings as evidence against using triple nucleoside regimens in patients with high viral loads, although use in patients with lower viral loads might be acceptable. However, these patients have only been on study for 48 weeks, and in the quest for chronic manageable therapy, 48 week data may not be adequate; differences between the arms may simply take longer to detect among those with lower baseline viral loads. Granted, this is only one nucleoside analog regimen, and there may be better performing nucleoside analog combinations (see Abacavir below). However, current studies do not address what sequencing options remain for a patient after failing triple nucleoside analog therapy or whether this is indeed the best way to utilize these agents.
In addition to the Atlantic Study, at least two other studies suggest that nevirapine can be as effective as a protease inhibitor. One was a retrospective analysis of 690 PI- or NNRTI-naive patients (half were completely antiretroviral naive) treated with nevirapine or one of five protease inhibitors at an HIV clinic in London between November 1994 and December 1998 (Lisbon, Abstract 237). Viral load responses, rate of rebound, CD4 cell response and the rate of treatment discontinuation were compared. In multivariate analyses that adjusted for baseline CD4 cell count, clinical stage and treatment history, nevirapine performed similarly at six months to the protease inhibitors (with the exception of the old Invirase formulation of saquinavir, which performed worse than nevirapine and the other protease inhibitors). Preliminary data from the second trial, the Combine study, being conducted in Spain and Argentina, were presented at the Retrovirus Conference as Treatment Issues went to press. The study is comparing nevirapine to nelfinavir (both with AZT/3TC) in 142 treatment-naive patients. At baseline the median viral load was 4.8 log (approximately 63,000) and the CD4 cell count was 355. Less than half of the patients on this study have data available out to six months, so the responses may look much different in the final analysis, but so far the nevirapine arm looks as good if not better than the nelfinavir arm. For example, in an intent-to-treat analysis, the proportion of patients with undetectable viral loads (less than 20 copies/ml) was significantly higher on nevirapine (58% versus 33%), but again, the results are preliminary.
In the third key PI-sparing study, another triple nucleoside analog regimen was compared to an indinavir-based regimen. Dr. Staszewski presented the 48 week results of CNA 3005, the study comparing abacavir (ABC) plus ZDV/3TC (Combivir) to indinavir (IDV) plus Combivir (see Treatment Issues, November 1998 and February 1999). In contrast to the earlier studies, this trial was double-blinded and placebo controlled, which meant that all subjects took medication three times a day. In other words, even if the patients were not on indinavir, they had to take an indinavir placebo for a total of 16 tablets per day and adhere to all of indinavir's dietary and liquid restrictions. Thus, there could be no adherence bias in favor of abacavir, and this should be kept in mind by all of us trying to make cross study comparisons (which are never really valid, although we all do it anyway) with the studies mentioned earlier.
At baseline, the mean viral load was 4.8 log (approximately 63,000 copies) and the CD4 cell count was 360. A higher percentage of patients completed 48 weeks of treatment on the abacavir arm (55% vs. 45%); however there was no difference in premature discontinuations on the basis of virologic failure and discontinuation rates for adverse events were also similar (17% for abacavir vs. 21%). Overall, at week 48 in the intent-to-treat analysis, 51% of both groups had HIV RNA less than 400 copies/ml. In the as-treated analysis, 86% of the triple nucleoside arm versus 94% of the protease inhibitor arm were below 400 copies/ml. Using the ultrasensitive assay, 46% of those in the indinavir arm were below 50 copies/ml versus 40% of those on abacavir.
Again, the differences between these responses do not appear to be great and are not statistically significant, but when looking just at those patients with higher baseline viral loads, the triple nucleoside arm looks inferior when using the ultrasensitive assay. Among patients who entered the study with viral loads above 100,000 copies/ml, significantly more patients randomized to indinavir had less than 50 copies/ml at week 48 (45% vs. 31%). However, we should point out that these were small subgroups and may have been influenced by the dropouts on abacavir in this subset.
But if a difference does exist between these triple nucleoside analog arms and the regimens based upon NNRTIs and protease inhibitors, what could be the cause? After all, in naive patients abacavir monotherapy seems to be as potent as an NNRTI or protease inhibitor (see Treatment Issues, November 1998). The difference could be that the virus needs to be attacked from two different sites or at least more "actively." That is, while some claimed that an NNRTI/two nucleoside regimen was a single pronged approach because both classes of drugs are reverse transcriptase inhibitors, the way they work is significantly different. One nevirapine molecule can bind to and gum up the works of the enzyme, just as one indinavir molecule stymies HIV's protease enzyme. Nucleoside analogs are more passive-aggressive. After intracellular conversion into nucleotide mimics (which happens less efficiently in some cell lines), they sit around and wait in a pool with hundreds of thousands of real nucleotides. Whether the reverse transcriptase enzyme chooses one of the imposters to use in building HIV's DNA chain is more or less luck of the draw. While it may be possible to get enough fake nucleotides into the pool to increase the odds of success, there is always a chance that it may not happen, particularly in those cells which do not activate (phosphorylate) the nucleoside analogs as well.
Now, Glaxo is poised to release a new combined tablet of abacavir/AZT/3TC called Trizivir. While we applaud the simplification of dosing that this represents (one pill twice a day), there is a danger that it enshrines this combination as being equivalent to NNRTI- or protease inhibitor-based combinations. So far, it is unclear whether this is the case for patients with higher viral loads, and it is possible that longer follow-up may reveal it to be less effective in patients with lower baseline viral loads as well. Other long-term studies (possibly in comparison to efavirenz) need to confirm the equivalence of this triple nucleoside regimen as a stand-alone first-line regimen before marketing designates it as such.
While the jury is still out on triple nucleoside regimens, the case for efavirenz (and perhaps nevirapine) versus protease inhibitors seems hard to dispute. In all of these studies, as well as in a number of "switching studies" (see last issue), the NNRTIs appear to be at least as good as, if not better than, protease inhibitors as an anchor for a first-line regimen for the majority of patients. However, we still have little data on the long-term use of NNRTIs, and their long-term adverse advent profile may not turn out to be so dramatically different than what has been observed on protease inhibitors.
The performances of efavirenz and nevirapine could be explained by the relative ease of adherence, their long half-life and high effective drug concentrations at trough. Protease inhibitor trough concentrations, on the other hand, vary widely from patient to patient and are greatly affected by food (with the exception of amprenavir, which can be administered with or without a meal). Even with strict q8 hour dosing, it may be difficult to maintain adequate indinavir blood levels consistently in all patients. However, for those patients who do have consistently adequate protease inhibitor blood concentrations, whether alone or increased by the addition of ritonavir, viral suppression could prove even more durable. For example, in the patients responding to therapy for at least a year in Study 006, indinavir's higher barrier to viral mutation may yet delay viral rebound, once differences in adherence and study discontinuation settle out. The goal here, after all, is durable suppression not for one or maybe two years, but indefinitely (or at least as close as we can get to it). In the end, we may discover that the best way to do that is to combine NNRTIs with protease inhibitors (or dual protease inhibitors to negate any drug interactions), and that the whole either/or debate has been wrong-headed. There has been great reluctance to do this, for fear of "losing" both classes at once, but patients rarely experience virologic break through with virus resistant to more than one drug at once. Resistance usually develops first to the drug with the lowest barrier to resistance (for example, when only one mutation renders the virus resistant, like the 3TC mutation, M184V, or for the NNRTIs, the K103N mutation).
The strategy at least merits more thorough investigation. There may be yet another reason to investigate such combinations. Recent data (to be discussed in our next issue) suggests that much of the long-term toxicity (lipodystrophy) originally attributed to protease inhibitors may be the result of a synergystic toxicity with nucleoside analogs. Studies of protease inhibitors combined with NNRTIs are needed to demonstrate whether they have a different long term side effect profile.
An intent-to-treat analysis counts any randomized subject for whom data is not available as a treatment failure (even if they drop out of a trial for personal reasons), while an on-treatment or as-treated analysis only accounts for the responses of those patients remaining on their randomized treatment. This latter method may accurately depict different regimens' relative potencies, but it fails to take into consideration how differences in tolerability and adherence might affect a regimen's performance. Thus an intent-to-treat analysis is generally perceived to be a more conservative method of gauging a drug's effectiveness. This is generally true when a study is double-blinded and placebo controlled. However, when it is not, a weaker regimen may look as potent as a stronger one simply because patients are better able to adhere to it. Such an adherence bias is worth taking into consideration because it reflects how treatment works for most people. Nevertheless, some patients may prefer to take more pills or try to adhere to a more complex regimen if it appears it is more potent, which an on-treatment analysis may show.
Comparing the overall rate of failure or response in clinical trials for the average patient might be inadequate in choosing the best regimen for the individual patient. It can be helpful to know why patients fail on a particular regimen, whether it is poor tolerance, adherence problems or the more rapid development of resistance.
Back to the GMHC Treatment Issues Winter 1999-2000 contents page.