Currently, definitive efficacy trials (such as the "confirmatory" trials to prove clinical benefit of drugs which have been given accelerated approval) are laboring under ground rules which make these studies extremely difficult to accomplish in today's treatment environment. The critical problem is the definition of clinical proof or confirmation as requiring an endpoint of death or progression to an AIDS-defining infection. This definition -- explicitly referenced in the FDA's announcement of its workshop on design issues in confirmatory AIDS trials [see note below] -- assures that these trials will take years, require hundreds of patients, be tedious and inordinately expensive to set up, and be unfeasible and probably unethical to run.
Requiring AIDS-defining illness or death in order to measure drug efficacy locks trials into the time frame required for volunteers to become seriously ill or to die -- a time getting longer and longer as both conventional and experimental treatments improve. Not only are vital results delayed, but compliance becomes difficult or impossible as the standard of care changes, requiring mid-stream amendments to protocols -- changes which damage the interpretation of results. Because of the long time required for volunteers to "achieve" an endpoint, trials must be very large -- making efficacy trials impossible for more than a handful of the many important drugs and combinations becoming available.
These problems are well known. The question is where to go from here. We propose a trial design which uses symptom reduction, instead of death or AIDS-defining infection, as a primary indicator for determining drug efficacy.
By symptom-reduction trial, we mean a clinical trial which recruits volunteers who have one or more designated indicator symptoms, in order to see if those symptoms improve as a result of the treatment being tested. These symptoms must be HIV-related, easily measurable, not intolerable, and unlikely to improve spontaneously or with standard treatment. Examples might include certain skin conditions, recurrent fungal or thrush infections requiring reliance on antifungals, fever of unknown origin, weight loss which cannot be diagnosed, or sleep disturbances.
Participants would be randomized into two groups -- antiviral treatment (for example, Merck protease plus 3TC plus AZT), vs. placebo or comparison treatment, for a few weeks. Both groups would continue their existing treatments (for example, antifungals) as needed throughout the trial. The primary outcome measurement would be reduction in the designated symptom(s) -- rated, for example, by blinded evaluation by physicians, or by reduced usage of specific medication such as antifungals. All outcome measurements would be defined prospectively.
Within weeks, with probably no more than a few dozen volunteers, these trials could produce statistical proof of clinical benefit from the antiviral treatment (compared to a placebo, or to a different antiviral regimen). This benefit, beyond its intrinsic value, would also serve as a marker for a more profound systemic benefit of the treatment -- namely, immune recovery resulting from suppression of HIV. At the end of the controlled-trial period, all volunteers would be offered the most desirable treatment, and followed indefinitely, to see how long the effect lasted.
Note that in this followup period, there will be two groups matched and randomized to be equivalent in every way, except that one started the antiviral treatment (say) eight weeks before the other. This eight-week difference will help researchers distinguish drug effects (both safety and efficacy related) from effects of disease progression, even after the controlled-trial period ends. Is there an eight-week difference in onset of the effect in question, or not?
Also note that the controlled-trial results could be published as soon as the controlled period ended, without much affecting the followup, as everybody would already have access to the most desirable treatment, so they would be unlikely to drop out or change treatments as a result of publication. This means that results important for clinical practice and for the development of new trials would no longer need to be kept secret for months or more by the DSMB (data safety monitoring board, which secretly unblinds the results of an ongoing trial to see if treatment differences are so great that it would be unnecessary and unethical to continue the study), or by a few researchers, but could immediately be disseminated and widely used.
One or two such trials (plus other straightforward research such as an adequate safety database, pharmacology studies of combination use of the new therapy with widely used drugs, etc.) should be enough to fully meet the requirement of clinical confirmation after accelerated approval. In fact, one or two of these trials should meet the efficacy requirement for approval of the drug in the first place -- bypassing accelerated approval by making it unnecessary.
Symptom-reduction trials can be as rapid as the surrogate- marker trials now used for accelerated approval. And they use the symptom improvement -- not just a blood-test value, but a real clinical benefit to the patient -- as an efficacy marker for a deeper benefit of the antiviral.
Will it be possible to recruit enough volunteers with an appropriate indicator symptom for these trials?
Yes, for several reasons. Not many volunteers are needed for each trial. Many people have ongoing HIV-related symptoms which have not responded to any available therapy. Not all patients in a trial necessarily need to have the same symptom. And many kinds of symptoms do respond to powerful antivirals; there are frequent case reports of many different symptoms -- often all of them -- being quickly and completely reversed by a powerful antiviral therapy (such as a protease inhibitor, or AZT plus 3TC plus ddC in a treatment-naive patient).
Symptom Reduction and Survival
What does symptom reduction add to the markers of HIV disease progression which are already in use?
The best marker available now is generally agreed to be viral load; a reduction in viral load shows that the treatment being tested is active as an antiviral in humans. Symptom reduction -- using a marker symptom which is not directly affected by the antiviral -- shows that this antiviral activity has resulted in immune improvement, which in turn has resulted in a clinical benefit. And if the marker symptom is correctly chosen to have no direct response to the treatment being tested (for example, reduction of fungal infections could be used as the marker symptom only if the antiviral regimen is not also antifungal), then it is likely that the clinical benefit seen is the tip of the iceberg, that the immune improvement which produced the benefit which is measured will also be beneficial in other ways which are not immediately apparent.
Will this immune benefit lead to longer survival?
Very likely yes, if the improvement can be sustained, but it will take years to know for sure -- just as it will take years to determine survival benefit in any HIV treatment trial which is started today (except for patients at very late stage disease, when it is usually hard to see the benefit of a new treatment). Meanwhile, we should go ahead with the best trials we can run now. There is no excuse to delay for years until survival benefit is proved.
Until now, only death or AIDS-defining opportunistic infection reduction has been generally accepted as an endpoint for a definitive efficacy trial. AIDS-defining opportunistic infection reduction is in fact a form of symptom reduction -- but with a very impractical choice of symptom to use as the marker, because it takes so long to measure the difference between the trial groups, making it almost impossible to design trials which are workable, reliable, and ethical. Trial designers need permission to select more workable symptom-reduction markers if they choose. Looking for improvement in existing chronic symptoms ties a trial to the time frame of drug effects -- days or weeks. But waiting for new symptoms to develop ties the trial to the time frame of HIV disease progression -- months or years. Confining clinical confirmation solely to changes in death or progression to AIDS might have made sense in the time of BW002 (the early AIDS trial which led to the approval of AZT), but it does not make
Today, the choice of AIDS-defining opportunistic infection as the definitive endpoint for efficacy trials has no scientific basis, benefit, or rationale. It rests on the historical inertia of an AIDS definition created years ago for other purposes. It is used because it is believed likely to be a good surrogate for survival, although this has not been proven. But survival itself -- despite its transcendent value -- becomes less and less feasible as the gold standard for measuring trial results, as treatments improve and survival lengthens. Demanding survival (or a known surrogate) as the endpoint in efficacy trials effectively prevents these trials from being designed and run successfully. The fact that survival is the ultimate goal does not automatically make it a useful guideline for the practical work of designing feasible and productive trials.
Suppose that a treatment causes a patient's viral load to decrease by a hundred times or more, and CD4 count to go up by two to three hundred or more, and all HIV-related symptoms to disappear -- and that this effect can be sustained, through use of various treatments -- can we prove that the patient will live longer as a result? No, we cannot prove it, short of a clinical trial which will take years. So we have a choice to make. We can insist on certainty and demand the survival data -- at the cost of making confirmatory or other efficacy trials prohibitively difficult to run, and guaranteeing that AIDS drug testing will not meet the needs of the epidemic. Or we can take a small chance of being wrong (the chance that the patient, even with all these improvements, is not really better off in the long run). By accepting this small chance of error, we open many opportunities for rapid, feasible, and useful testing of attractive new drugs and combinations. The choice is ours.
History and References
AIDS trials do of course report the development of symptoms less serious than AIDS-defining opportunistic infections. But we do not know of any trial which has used an existing HIV-related symptom as an indicator of effectiveness of an antiviral or other treatment for HIV disease -- requiring that patients have the symptom when they enter the trial, and using improvement or disappearance of the symptom as the primary measure of drug efficacy.
A recent article, "Selection of Endpoints for Assessment of Treatment Efficacy in an AIDS Trial" (by David A. Amato, in AIDS Clinical Trials, edited by Dianne M. Finkelstein and David A. Schoenfeld, Wiley-Liss, 1995), explores the critical difficulties in the selection of workable endpoints today. This article includes two sentences on time to development of non-AIDS-defining symptoms as a possible endpoint -- but no hint of reduction of existing symptoms as a measure of drug efficacy. Outside of AIDS, however, the idea has been discussed, with different terminology.
Researchers have noted that a therapy:
"'tested in a trial can have a remedial or prophylactic target... Remedial targets have been the ones for which randomized controlled trials have been particularly successful. In a study of remedial therapy, the patients under investigation are relatively homogeneous because they all possess the same target to be altered; the target can be directly "measured" and observed for change; the change usually requires only a relatively short time to occur; and the compared differences in the measured changes can be statistically significant without requiring massive numbers of patients.'"
"prophylactic trials, designed to assess relative efficacy in preventing some future adverse occurrence, by contrast, 'are the ones that have been notoriously expensive and that have often created, rather than clarified, controversy.'"
(Robert J. Levine, Ethics and Regulation of Clinical Research, Yale University Press, 1986, Chapter 8, "Randomized Clinical Trials," section on "Excessive Reliance on Randomized Clinical Trials," quoting from A. R. Feinstein, "An Additional Basic Science for Clinical Medicine: II. The Limitations of Randomized Trials." Annals of Internal Medicine, 1983, volume 99, pages 544-550.)
Note that in AIDS, all "efficacy" trials have been the "prophylactic" kind, comparing how well the different treatments prevent an endpoint which the patients do not have when the trial begins. Symptom reduction would allow the "remedial" kind of endpoint to be used in testing efficacy of AIDS treatments.
Arguments Against Symptom Reduction As Outcome Measure
Objection: Symptom-reduction trials do not tell us what we need to know. The real goal of confirmatory trials is to tell us how the long-term benefit of the treatment, especially survival benefit, compares with its long-term harm. Short- term symptom relief does not give us this long-term information.
Reply: There is no way around the fact that long-term information takes a long time to obtain. The practical result is that long-term trials with fixed protocols cannot be run successfully, because treatments are changing rapidly, protocols become undesirable before a long-term trial can finish, and there is no way to keep people on inadequate treatment long enough for the body counts to be obtained. And besides the human cost of having to wait years for the information -- often a separate wait for each drug or combination regimen tested -- there is also the scientific cost of having to wait years for each "turnaround" (for what is learned from the results of one trial to be available for use in the design of another). There is a saying that the best can be the enemy of the good. We must use practical judgment to move ahead where we can, instead of being paralyzed by the lack of a path to perfect knowledge.
Objection: Symptoms in HIV disease are highly individual. Exactly which symptoms would one use as indicators of drug effects?
Reply: This question needs research -- through study of case histories and of clinical-trial data, and perhaps through pilot studies. But there have been major, dramatic symptom improvements, when viral load and CD4 counts have shown that an antiviral treatment has been particularly successful. Symptoms often respond strongly to effective treatment. It would be surprising if no useful symptom-based indicator of treatment success could be defined.
Objection: Symptom information is likely to be burdensome to collect in clinical trials, and is often biased.
Reply: Symptom information is burdensome and biased when trials collect too much data, on the chance that something might be useful later. But a symptom-reduction trial can take a more focused approached. Only one or at most a few symptoms need be recorded. These symptoms can be selected to reduce bias, recording instruments can be fine-tuned, and observers can be trained. And everyone has the symptom at entry into the trial, providing a baseline for comparison.
Objection: Early in the epidemic, some symptoms were found not to predict progression to AIDS.
Reply: All cases of failure of symptoms to predict disease progression need to be reviewed in detail. What happened then, and what can we learn from it today? But early experience must be applied cautiously in the present. Several years ago, it was widely believed that due to immune damage, people with CD4 counts under 200 -- certainly under 100 -- could not improve much by any antiviral treatment alone, no matter how effective it was against the virus. But now there are cases of CD4 count going from under 100 to several hundred with only antiviral treatment -- and simultaneously most or all HIV-related symptoms going away. The difference reflects better antivirals, and more choices of treatment for patient management. The only antiviral available years ago -- AZT monotherapy -- has very limited effectiveness. What happened in the past should not automatically be applied today, especially if it stops us from developing opportunities available now.
Also, note that the development of a symptom during disease progression is different from the rapid resolution of that symptom due to antiviral treatment. Even if symptom development fails to predict later disease outcome, symptom resolution after therapy must still be investigated separately.
Objection: If the symptoms take a long time to resolve, you could miss an important drug.
Reply: Case reports suggest that many symptoms improve rapidly and dramatically when viral load falls greatly due to successful antiviral treatment. Trial designers can avoid choosing indicator symptoms which do not change rapidly.
Objection: These trials cannot test drug benefit on people who are completely asymptomatic. Will they be excluded from the drug indications?
Reply: No. These trials use indicator symptoms to demonstrate an underlying benefit -- which, if it exists, is unlikely to go away just because there are no symptoms to show it immediately. Drug indications must be based on all available knowledge -- including viral load and T-helper count changes, which usually can be observed in asymptomatics.
Objection: Symptom reduction may be a useful outcome measurement for immune-based, anti-pathogenesis, or certain miscellaneous therapies, where no validated endpoints are available. But for antivirals, it has no advantage over viral load, which directly measures the effect being sought.
Reply: Some people still have doubts that viral load reduction caused by a drug will translate into concrete clinical benefit for patients. The symptom reduction outcome addresses this specific concern. In other respects, for testing antivirals, viral load may be superior, because it is further along on the development curve. And of course viral load can be used as an additional outcome measurement in a symptom-reduction trial.
Objection: Symptom-reduction trials are likely to be rejected on various grounds. And by proposing them, one will be seen as buying into the position that clinical confirmation is needed for each treatment or combination which has received accelerated approval. But AIDS is different from other diseases. There is tremendous individual variation -- and a virus which may be able to mutate against any treatment. In this situation, we need to build an armamentarium of active treatment options, so that when one treatment does not work for a patient, others will be available. The idea that every treatment must have clinical confirmation, even after good safety and viral load response have been shown, is entirely inappropriate to the AIDS situation, and is a major obstruction. We should be telling those who insist on such confirmation that the emperor has no clothes, that they are wrong.
Reply: We proposed symptom reduction for two reasons. First, it has certain intrinsic advantages as an outcome measure, especially for testing experimental treatments other than antivirals. But also, the demand for death or opportunistic infection endpoints is politically so strong that it may be impossible to move forward unless a compromise can be found. Symptom reduction goes part way. It is a clinical measurement of real benefit, not a blood test. But it is largely a short-term measurement, not a long-term one.
There are many reasons why the death and opportunistic infection endpoints have strong support. First, drugs have been approved in the past without enough guidance on how to use them; often doctors and patients do not even know if an approved treatment will extend life or shorten it. Also, many people believe that it is morally wrong to market a drug which has not been proven to have benefit; and they see any shortcut to such proof as weakening standards of science and public protection, and letting big corporations rake in profits without doing their job. And many believe that taking the time now to get the science right will serve the greatest good of the greatest number eventually. Some want their expected death to be meaningful in this way, as a service to others in the future.
Clearly some of the issues in the clinical-trials debate are pragmatic -- which course of action will actually lead to faster progress and better treatments, both in the near future and the far future? Other parts of this dispute are moral in nature, and therefore less subject to verbal resolution. The best outcome may be a compromise both sides can accept. This dispute is not new, but has roots in long-standing conflicts in clinical-trial design. Feinstein and Levine (references above) sketched differences between "pragmatic" trial designers (usually clinicians), and "fastidious" ones (usually biostatisticians). We suspect that this conflict may be partly rooted in personality differences, which may be with us indefinitely -- not just technical beliefs, which would tend to resolve as technology develops.
And we cannot ignore institutional incentives. Large, long-lasting trials provide predictable income streams and staffing demands, suitable for long-range planning by large medical institutions; small, rapid trials are harder to plan for. And very expensive drug development favors large companies, which can meet any costs demanded; in return they receive a monopoly -- in effect, an important form of capital for the large companies -- as smaller and potentially more efficient companies are excluded from drug development unless they sell out to large companies on their terms. In these ways, marginally unworkable standards for drug development are potentially worth billions of dollars to key institutions -- a massive institutional incentive which cannot fail to affect what is or is not funded, published, approved, and eventually accepted into mainstream medical care.
Symptom Reduction As Outcome Measure:
Summary of Advantages Over Disease Progression, in HIV Efficacy Trials
- Symptom reduction measures a clinical benefit to patients, not just a blood-test value. Since the trial designers choose an indicator symptom not directly affected by the treatment being tested, this particular symptom serves as a marker for a broader systemic improvement.
- Trials can be run in a few weeks, because many HIV-related symptoms do respond to a powerful antiviral within that time. There is no need to wait for disease progression in the control group.
- Trials can be small, because every participant produces a measurement -- not just the fraction who reach an endpoint, as in disease-progression trials.
- Because these trials last only for weeks, and participants are not gravely ill, compliance will be better than with trials which last so long that the standard of care overtakes them. The dropout rate should be much lower than for disease- progression efficacy trials.
- Volunteers need not have very advanced illness; therefore, it will be feasible to recruit therapy-naive patients when necessary for scientific reasons.
- The fact that these trials are small and very rapid, and then release results for everybody, will reduce pressure to include additional patients in order to provide treatment access for them.
- These trials can easily be randomized and double blind. Therefore, any subjectivity in the rating of symptoms should not cause a consistent bias.
- Placebos can be used when needed, because trials are short, and symptoms can be selected to be not immediately dangerous or severe.
- Symptom-reduction trials can measure the efficacy of antiviral, immune-based, pathogenesis-oriented, supportive, or other kinds of therapy -- all with an identical trial design if desired. This facilitates testing combinations of these different kinds of potential treatments.
How can we use the upcoming FDA workshop on confirmatory trials to open a national discussion on allowing improvement of existing symptoms -- not just prevention of death or AIDS- defining opportunistic infections -- as a measure of drug efficacy? We want the professional community to fairly consider this kind of design, and accept or reject it on its merits. If it is accepted in the professional consensus, then researchers could get funding for such trials, and the FDA would let sponsors know that it could accept the results as definitive proof of efficacy.
Note: Our last issue announced the FDA workshop on the design of HIV clinical trials, September 6-7, with an advisory committee meeting on September 8; both meetings will be held at the Natcher Auditorium of the National Institutes of Health in Bethesda, Maryland. The workshop will focus on "confirmatory" efficacy trials (for drugs which have already received accelerated approval). These meetings are open to the public, but advance registration is recommended, and to be sure of seating, persons should arrive early even if registered; 300 people have registered as of August 16, and he auditorium holds 500.
To register, send a fax to Heidi Marchand or Kimberly Miles, Office of AIDS and Special Health Issues, 301/443-9216; include your name, organization if any, address, and phone number. For more information, call Heidi Marchand or Kimberly Miles, 301/443-0104.
[AIDS Treatment News will attend, and is seeking professional and community support for getting the trial-design suggestions outlined above "onto the table" for consideration, at these meetings and otherwise.]