StATS: Ethics of a placebo group (created 2001-08-02)
Dear Professor Mean, Some of my colleagues want to use placebos in their research, but I have warned them about the ethical issues surrounding the use of a placebo group. When (if ever) is it ethical to use a placebo group? --Kibitzing Kathy
Dear Kibitzing, You're asking a statistician about ethics? There's a cute story about statisticians and ethics.
Three candidates were being interviewed for a job as a spokesperson for a new biotech company. The first candidate was a lawyer, and she was asked: "What is one plus one?" Her response was "In matters of commercial law, the operating assumption as determined by multiple legal precedents is that one plus one shall be considered as prima facie evidence equaling two."
The second candidate was a physicist. Her response was "Recent developments in quantum physics have allowed for unusual mathematical representations of physical realities. Yet in all of these representations, the quantities of one and one always add up to two."
The third candidate, a statistician, was also asked "What is one plus one". The statistician got up and closed the door and whispered quietly "What do you want it to be?"
There are several dimensions to the placebo issue. The placebo group provides information to a specific research question that may or may not be relevant. Also, the use of a placebo is intended to withhold treatment information from the patient and usually from the medical personnel evaluating that patient. Finally, the use of a placebo is usually intended to deny a treatment to some of your patients. So, if you want to use a placebo, you have to ask three questions.
- Will the placebo group help answer a relevant scientific question?
- Is it ethical to hide information from your patients?
- Is it ethical to deny a treatment to some of your patients?
The answers to these three questions are difficult, in part because they depend on the medical context, and in part because there is no consensus in the medical community about these issues.
Here is just a sampling of some of the opinions on the subject.
The benefits, risks, burdens and effectiveness of a new method should be tested against those of the best current prophylactic, diagnostic, and therapeutic methods. This does not exclude the use of placebo, or no treatment, in studies where no proven prophylactic, diagnostic, or therapeutic method exists. -- Declaration of Helsinki.
Placebo controlled trials may be ethically conducted even when effective therapy exists, as long as patients will not be harmed by participation and are fully informed about their alternatives. -- Temple and Ellenberg (2000a).
As medical knowledge accumulates, the number of placebo trials should fall. -- Rothman (1996).
Metaphorically, placebos have been predestined to be roadkill on the highway of medical progress. -- Urquhart (2001)
Some guidelines are totally silent on the use of placebos (e.g., Title 45 Code of Federal Regulations Part 46. Protection of Human Subjects).
What is a placebo?
A placebo can take many forms. Sometimes it represents the use of a sugar pill or other inert substance, when testing a medication. It can also involve some sham procedure, when testing a medical procedure. For example, in a study of acupuncture, the placebo group might receive acupuncture at sites 5mm away from the specified acupuncture points.
The placebo has several benefits.
- A placebo provides a greater degree of objectivity in measurements that otherwise might be perceived as subjective.
- A placebo provides protection against possible subconscious desires of the patient to please the researcher.
- A placebo controls for attitudes of the patient and/or doctor that might influence the outcome of the study.
- A placebo provides some level of protection against fraud when a patient and/or doctor has a financial or non-financial incentive to demonstrate a particular finding.
There is considerable controversy over whether a placebo has any effect on medical outcomes. A recent meta-analysis (Hrobjartsson and Gotzsche 2001) suggested that the placebo effect has been overstated. Some have commented that this article undercuts the need for a placebo group in research.
A careful reading of this study reveals that sometimes the placebo effect is confused with other effects, such as regression to the mean or the tendency of some diseases to be cyclical or to resolve spontaneously. The authors argue that an attempt to try to harness the power of the placebo effect is ill-advised, but they still believe there is a place for the placebo in medical research.
This is a good study, because there has been a misguided effort to argue that placebo effects imply that patient attitudes have a strong impact on outcomes, with the unfortunate corollary that if you didn't do well, it must have been because you had a bad attitude. Also a lot of fringe and alternative medicine proponents have argued that their approaches rely on the same mechanisms that produce the placebo effect (!) and that they are tapping into something that traditional medicine cannot explain.
How is a placebo used in research?
In most of the discussion on this page, I will assume that the research involves the investigation of an experimental treatment labeled X. Some information about X may be available, and X may already be in use to some extent. But the presumption is that all of the effects, both positive and negative, for X are not yet completely known. Let's also assume that there is a standard treatment labeled S which represents the current best standard of care.
So the simplest example of the use of placebo is the the classic placebo-control trial. This trial compares:
- treatment X, and
- a placebo for X.
An important alternative is the active-control trial. This trial compares:
- treatment X, and
- treatment S.
When you need to use blinding in an active-control trial, you have to find a placebo for A and a placebo for B. This creates a double-dummy trial, which compares:
- treatment X combined with a placebo for S, and
- treatment S combined with a placebo for X.
Another variation is the add-on trial, which compares:
- treatment S combined with treatment X, and
- treatment S combined with a placebo for X.
Still another variation is the three group comparison:
- treatment X combined with a placebo for S,
- treatment S combined with a placebo for X, and
- placebos for both X and S.
You should also be aware of the difference between equivalence trials and superiority trials. An equivalence trial is intended to demonstrate that treatments S and X are equivalent (the difference in efficacy lies inside the range of clinical indifference). A superiority trial is intended to demonstrate that treatment X is superior to treatment S (the difference in efficacy lies outside the range of clinical indifference).
Any placebo-control trial is a superiority trial (it is hard to imagine a drug company proclaiming that their new drug is as good as placebo). An active-control trial can be either an equivalence or a superiority trial.
Scientific merit of placebos
Research on human subjects (or for that matter on animals) is unethical if the results of the experiment will not provide any useful knowledge. The problem is that what constitutes useful knowledge is often in the eye of the beholder.
The U.S. Food and Drug Administration has come under sharp criticisms in allowing placebo controlled trials.
When a placebo control is used instead of an effective treatment, the effect of a new drug appears large and may be statistically significant even in a small study. The scientific benefit, however, is illusory. Because the study is small, the measurement of the effect is subject to considerable statistical error. Thus, the actual size of the effect, even when a new drug is compared with placebo, remains obscure, and the study does not address the question of the effectiveness of the new treatment as compared with currently accepted treatments.
The small placebo-controlled studies fostered by the FDA benefit drug companies, which can more easily obtain approval of an inferior drug by comparing it with placebo than they can by testing it against a serious competitor. Smaller studies are also cheaper. Unfortunately, the costs saved by the drug company are borne by patients, who receive placebos instead of effective treatments, and by the public at large, which is supplied with a drug of undetermined efficacy. -- Rothman and Michels (1994).
Others have argued that studies without a placebo-control lack scientific merit.
A comparison between new drug and standard is convincing only when the new remedy is superior to standard treatment. If it is inferior, or even indistinguishable from a standard remedy, the results are not readily interpretable. In the absence of placebo controls, one does not know if the "inferior" new medicine has any efficacy at all, and "equivalent" performance may simply represent a patient population that cannot distinguish between two active treatments that differ considerably from each other, or between active drug and placebo. Certain clinical conditions, such as serious depressive states, are notoriously difficult to evaluate because of the delay in drug effects and the high rate of spontaneous improvements, and even known remedies are not readily distinguished from placebo in controlled trials. -- Lasagna (1979).
Both viewpoints have some validity, but the arguments are quite subtle. The problem is that it is often difficult to combine two pieces of evidence to come up with a third piece of evidence.
Suppose there are two studies. The first shows that X is equivalent to S. A second study shows that S is superior to placebo. You would like to conclude, then, that X is also superior to placebo, but you can't. From the perspective of X, the placebo group is a historical control. It was run at a different time and was drawn from a different pool of patients.
Also, in the world of Statistics, X=S plus S>P does not imply that X>P. Consider the following analogy. Xavier lives within one mile of Sarah. Sarah lives more than one mile away from Pamela. You can't conclude that Xavier also lives more than one mile away from Pamela.
For the same reasons, showing that X is superior to placebo and that S is superior to placebo does not necessarily imply that X is equivalent to S.
Although it is easy for me to raise these questions, the answers are difficult. Do clinicians need to know whether X is equivalent to S? Do they need to know if X is superior to placebo? It depends a lot on the context of the problem. If both questions are important, then only a three arm trial (X versus S versus placebo) would have scientific merit.
A three arm trial has an additional benefit. Suppose that S has already been shown to be superior to placebo, but in this trial you find out that S is equivalent to placebo. This tells you that failure to show superiority for X over placebo is probably due to the fault of the research design rather than any failing of treatment X itself.
Hiding information from patients
There are good scientific reasons to hide information from patients during the conduct of a clinical trial, but these need to be balanced against the needs of the patient. The key issues are whether hiding that information has scientific merit and whether this hiding of information can be shown to cause no harm to the subjects.
There are indeed examples of research generally perceived as ethical where information was withheld from patients. For example, a study of intercessory prayer recruited patients without informing them (much less obtaining their consent).
That the hiding of information has scientific merit is usually straightforward enough. There is ample empirical data to suggest that failure to hide information about treatment status can cause serious biases in research.
It is also usually easy to establish that hiding information will not cause any harm. You should insure, for example, that hiding information will not interfere with other care that the patient might receive. A patient who is about to start using birth control pills, for example, would probably want to know whether she is receiving an antibiotic treatment or a placebo of the antibiotic. A patient who needs to undergo an emergency appendectomy will need to know whether he is taking a blood pressure medication or a placebo. These cases can be often handled by setting appropriate restrictions on the patient pool and/or by allowing subjects to drop out of the research for medically necessary reasons.
A special problem of consent arises where informing subjects of some pertinent aspect of the research is likely to impair the validity of the research. In many cases, it is sufficient to indicate to subjects that they are being invited to participate in research of which some features will not be revealed until the research is concluded. In all cases of research involving incomplete disclosure, such research is justified only if it is clear that
- incomplete disclosure is truly necessary to accomplish the goals of the research,
- there are no undisclosed risks to subjects that are more than minimal, and
- there is an adequate plan for debriefing subjects, when appropriate, and for dissemination of research results to them.
Information about risks should never be withheld for the purpose of eliciting the cooperation of subjects, and truthful answers should always be given to direct questions about the research. Care should be taken to distinguish cases in which disclosure would destroy or invalidate the research from cases in which disclosure would simply inconvenience the investigator. -- Belmont Report
Withholding information is not the same as intentional deception. You can and should tell your patients that they will be randomized into either a treatment or a placebo group and that they will not know which group until after the study is completed. Failure to inform your patients that they may receive a placebo raises some serious ethical problems.
Many trials, however, are preceded by a "placebo run in," in which all patients are given placebo. The practice is common within the pharmaceutical industry and recommended by standard texts as a means of weeding out non-compliers before randomisation, eliminating placebo responders, ensuring that patients are stable, washing out previous treatment, or simply to provide a period for baseline measurement. This is incompatible with informed consent, since a doctor is hardly likely to say: "Take this ineffective substance for the next month and record your symptoms daily in this diary." -- Senn (1997).
You should carefully consider the situations that might warrant disclosing information about whether a patient received the treatment or placebo.
To balance the need for scientific objectivity with the concern for subject safety, investigators should consider in advance the conditions in which a blind may be broken to treat an adverse event. Specifically, they should include a description in the protocol of where the code is located, the circumstances (if any) in which the code will be broken, who will break it, how the information will be handled (i.e. will the investigator, the subject, the IRB, and the treating physician be informed?), and how breaking of a blind will influence analysis of the data. The subject should also have information about whom to notify in the event of an emergency. The IRB should be satisfied that the plan provides for adequate protection of subject privacy. -- OHSR (1998).
Informed consent requires that you tell your patients that information is intentionally being withheld.
In studies that employ a randomized blinded approach, both randomization and blinding should be explained in simple terms in the consent document. The explanation should include the meaning of randomization, placebo (if used), and blinding; why these methods are being used; who has the ability to identify treatment assignments (who has the code); when and under what conditions the blind may be broken; and when information about treatment assignment will be shared with the subject. It should be made clear whether information is to be shared at the completion of the study rather than when the subject personally completes the study. -- OHSR (1998).
Uncertainty, indifference, and equipoise
Use of a placebo means that a treatment is being deliberately withheld from a random fraction of the research patients. This sounds terrible, but a lot depends on the context. After all, withholding treatment is often appropriate. In Oncology, there is even a term for this: "watchful waiting".
Whether withholding a treatment is ethical or not depends on what is known about that treatment. If there is knowledge that the treatment is superior to placebo, then it would be inappropriate to include a placebo group (though some would argue that there are exceptions, see the discussion below about special placebo designs). It is lack of knowledge (uncertainty) that allows you to ethically use placebo controls. But uncertainty comes in different flavors.
"How much uncertainty can we accept before entering a patient into a trial and by whom (patients, physicians, and community)?" -- Lilford and Djulbegovic (2001)
The answer to this question depends, unfortunately, on who you are talking to. Also, there is a lot of ambiguous terminology. The first term you will see is "equipoise." The general rule is that equipoise makes an experiment ethical; when equipoise is lost, the experiment becomes unethical.
Merriam Webster defines equipoise as a state of equilibrium or balance. A placebo group is ethical if there is balance between treatment X and placebo (taking into consideration all that is known about the efficacy of treatment X and its side effects). It is important to emphasize that there is never perfect balance or perfect equipoise in the real world. The FDA has stated:
It is clear, despite the uncertainty, that the investigational intervention is intended to be beneficial and that there is conceptual, preclinical, and possibly clinical (e.g., other settings, preliminary results) evidence that the hoped for benefits outweigh the potential risks, all of which leads the investigator (and the pertinent IRB) to hope for, even anticipate, benefit. Such anticipation is compatible with the state of clinical equipoise needed to allow a clinical investigation. Indeed, true neutrality is rarely present at the start of an investigation; in the absence of expectation that an intervention may represent an improvement, or a belief that a standard therapy might not work, there is little incentive to proceed. The experienced clinical investigator, however, also knows that expectations are not the same as knowledge and that disappointments are too common to ignore. Therefore, despite optimistic expectations, one can be in the state of equipoise needed to allow a clinical investigation to be conducted. -- www.fda.gov/opacom/morechoices/fed996.html.
Others, however, describe equipoise as being in a position where you would accept an even odds bet. This strikes me as somewhat naive, because in every medical situation there would be a range of responses where the clinician would be indifferent between two treatments (or between a treatment and a placebo).
Others have used terms like "genuine uncertainty" which I think is reasonable. Others have said that equipoise means that the researcher is "indifferent" between the treatment and the control. I dislike this term because indifferent implies a sense of detachment or boredom. You can, however, interpret this in a different sense. A medical professional is indifferent if he/she believes that the range of uncertainty about a treatment is wide enough to incorporate part or all of the range of clinical indifference.
There is even more confusion on this issue. A "hunch" may not be enough to upset the balance of clinical equipoise:
Controlled studies are possible and necessary, however, because even though clinicians usually have hunches that one treatment arm is more effective than another, they are often not certain that their hunches are correct. The boundaries (confidence interval) on their hunch may range from much better, through marginally better, down to ineffective, or even frankly harmful. When this is the case "it is time for a trial, and that trial is ethical." -- Weijer et al (2000).
The same authors, however, argue that information "on personal experience, on anecdote, on tacit understanding, or rules of thumb" are sufficient to upset equipoise.
Community equipoise versus individual equipoise
Equipoise or uncertainty can be held at several different levels:
- Lack of consensus in the medical community, or
- Uncertainty of the individual physician or medical professional.
The preferences of the individual patient are also relevant (see discussion below in the patient choice section).
The language is not standardized. Some people refer to lack of community consensus as "macro uncertainty" and others call it "the uncertainty principle".
Some researchers have proposed that community equipoise rather than individual equipoise should dictate whether a trial is ethical.
The ethics of clinical research requires equipoise--a state of genuine uncertainty on the part of the clinical investigator regarding the comparative therapeutic merits of each arm in a trial. Should the investigator discover that one treatment is of superior therapeutic merit, he or she is ethically obliged to offer that treatment. The current understanding of this requirement, which entails that the investigator have no "treatment preference" throughout the course of the trial, presents nearly insuperable obstacles to the ethical commencement or completion of a controlled trial and may also contribute to the termination of trials because of the failure to enroll enough patients. I suggest an alternative concept of equipoise, which would be based on present or imminent controversy in the clinical community over the preferred treatment. According to this concept of "clinical equipoise," the requirement is satisfied if there is genuine uncertainty within the expert medical community--not necessarily on the part of the individual investigator--about the preferred treatment. -- Freedman (1987)
Clinical equipoise would exist whenever at least a reasonable minority of medical professionals believe the experimental treatment would be as good as, or better than, the standard treatment. -- www.fda.gov/opacom/morechoices/fed996.html.
Others believe instead that is the individual doctor's beliefs and not community consensus that determines whether a trial is ethical.
An ethical physician must do what is best for his or her patients. She cannot participate in a controlled trial if she is certain that one arm is superior to the others and that some of her patients will receive an inferior treatment by participating in the trial. It does not matter whether her certainty is based on formal scientific studies, on personal experience, on anecdote, on tacit understanding, or rules of thumb. Whether her certainty is in accord with or diverges from the view of the medical community is irrelevant. Uncertainty is a moral prerequisite for a controlled study. If we know what we should do, we should do it, not study it. Controlled studies are possible and necessary, however, because even though clinicians usually have hunches that one treatment arm is more effective than another, they are often not certain that their hunches are correct. The boundaries (confidence interval) on their hunch may range from much better, through marginally better, down to ineffective, or even frankly harmful. When this is the case "it is time for a trial, and that trial is ethical." -- Wejer et al (2000).
A patient can be entered if, and only if, the responsible clinician is substantially uncertain which of the trial treatments would be most appropriate for that particular patient. A patient should not be entered if the responsible clinician or the patient are for any medical or non-medical reasons reasonably certain that one of the treatments that might be allocated would be inappropriate for this particular individual (in comparison with either no treatment or some other treatment that could be offered to the patient in or outside the trial). -- Peto and Baigent (1998).
Equipoise in one or both directions?
Often a therapy can be considered new, untested, and/or innovative. The newness can introduce some asymmetries into the considerations of equipoise.
Comments just above by the FDA (experimental treatment would be as good as, or better than, the standard treatment) imply that as long as a treatment is considered experimental, testing can ethically continue, even if the researchers believe the the experimental treatment is indeed superior.
Rothmann and Michaels have a similar view:
... every patient, including those in a control group, should receive either the best available treatment or a new treatment thought to be as good or better. -- Rothman and Michels (1994)
Stephen Senn states the argument even more directly.
Equipoise is an irrelevance. Patients are entered onto the trial because the trialists believe that the experimental treatment is better. The control group treatment should be as good as that available outside the trial. Experimentation continues until either the trialists are convinced they are wrong (equipoise is reached) or they convince Society they are right. -- Senn (2000).
Others have argued (see some of the arguments in William Silverman's book) that you need to randomize from the very first patient, before there is a chance for non-randomized data to upset equipoise. This strikes me as a bit naive. Some therapies require a bit of preliminary data before they can even be considered equivalent to placebo, because they have serious side effects. Furthermore, some therapies like surgery require some training and experience before they can be incorporated into the context of a clinical trial. Besides, if equipoise is such a fragile thing that a few preliminary non-randomized patients can upset it, then why wouldn't promising results in an animal model also upset equipoise?
Changes in equipoise
Equipoise can change from one patient to another. Your research protocol should specify inclusion and exclusion criteria that would avoid randomizing patients when there is clear evidence that treatment X is the only appropriate option (or conversely that treatment X is clearly inappropriate). The medical professional responsible for recruiting patients, of course, always has the option of declining to enroll a patient and should be encouraged to use this discretion appropriately. The patient should also be provided with enough information so that they can decide if they are in equipoise with respect to the treatment and the placebo.
Information about the current standard of care can also change over time, based on information collected during the study itself. This information might lead you to stop a study before the planned completion date if there is convincing early evidence of:
- efficacy ,
- lack of efficacy, or
- harmful side effects,
though you should specify detailed criteria in your protocol about when you would expect your trial to stop early. You should also continually review developments in the field to see if these developments make further research unethical.
Interim analyses (also known as group sequential trials) provide a specific framework for early stopping. One of the best approaches, the method of O'Brien and Fleming method uses a very strict cut-off at first, then relaxes this cut-off over time. If you wanted two interim and one final analysis. The two interim analyses would occur after one-third and after two-thirds of the patients have been evaluated. At the first interim analysis, you would conclude that the new therapy is effective if the p-value is less than 0.005. At the second interim analysis, you would compare the p-value to 0.014. At the end of the study, you would compare the p-value to 0.045.
Interim analyses offer a lot of intriguing advantages, but I'm not convinced yet that they are uniformly superior to the traditional analyses. There was a lot of controversy over the early stopping of the Tamoxifen trials. Take a look at Bruzzi 1998 for an example of some of the downside with early stopping.
Early stopping rules do not come without their costs. It is pretty easy to construct scenarios where the conventional test would have rejected the null hypothesis, but interim analysis would not because some of the alpha level was "spent" in early tests. On the other hand, some interim analyses can be shown to have superior power to the conventional test.
The logistics of interim analyses might also create costs greater than the expected savings. On the other hand, if the interim design has a good shot at reducing the number of patients receiving the inferior treatment, that would usually be more important. Note that sometimes it is the active treatment and not the placebo which ends up being the inferior treatment.
Most importantly, you need to compare the length of accrual of subjects and the length of time to determine an outcome for each subject. It makes no sense, for example, to design an early stopping rule when you are accruing subjects for only one year and your outcome is three year survival rates.
Another attractive approach is adaptive randomization. Peter Thall describes a study of soft tissue sarcoma that involved four arms:
- surgery with no chemotherapy agent,
- surgery with one chemotherapy agent,
- surgery with a different chemotherapy agent,
- surgery with both chemotherapy agents.
As the trial progresses, the statistician monitors the median survival time in each of the arms. These patients are quite ill; historical data placed the median survival time at eight months. Rather than allocate to each arm of the trial with 25% probability, the probability would increase for the arm(s) with larger medians and decrease for those with smaller medians. As the sample size and the total follow-up time increases, the inferior arms of the trial should have randomization probabilities that shrink to zero.
Originally, Dr. Thall had also included an early stopping rule, but the physicians balked. They pointed out that if one arm emerged as an early winner, the adaptive randomization probability would effectively insure that all of the later patients would get the most effective treatment.
It is indeed the physician's responsibility to protect his/her patients. Even if a patient is willing to participate in a trial, the physician cannot abdicate his or her responsibility.
Medical research involving human subjects should be conducted only by scientifically qualified persons and under the supervision of a clinically competent medical person. The responsibility for the human subject must always rest with a medically qualified person and never rest on the subject of the research, even though the subject has given consent. -- Declaration of Helsinki (2000).
Rothman and Michels (1994) make a similar argument.
Informed consent is always desirable, but investigators should not put patients in a position in which their health and well-being could be compromised, even if the patients agree. There are several reasons. Despite the best efforts to inform patients, they will rarely if ever be as well informed about their treatment options as their physicians. Moreover, even informed patients may not be disinterested enough to decide rationally whether it is tolerable to be deprived of an accepted treatment. Finally, patients are given the choice of participating in a trial or not, but they are given no choice about which treatments will be studied. It may be more desirable to a patient to be a part of the trial than to decline to participate, but it might have been preferable to be in a different trial that did not have a placebo arm. -- Rothman and Michels (1994).
But sometimes this protection can cross the line and become paternalism.
"Two years ago I was diagnosed as having breast cancer. I had recently started a PhD investigating the application of randomised controlled trial methods to my specialty and had just returned from a course at which the ethics surrounding a randomised controlled trial in breast cancer were discussed. I had decided, before I was told my diagnosis, that if my lump was malignant I wanted to be entered into a trial if I was eligible for one. On receiving my diagnosis I told my consultant that I would like to be considered for a trial. The reply was, "You mustn't let academic niceties get in the way of the best treatment for you." I already knew the uncertainties surrounding treatment for breast cancer and the academic arguments in favour of and against randomised controlled trials, and I felt belittled by this statement. I was fortunate in being able to have my care transferred to an academic unit, where my request to enter a trial was welcomed. The trial was discussed at length, and I was informed of the short term and long term implications, benefits, and risks of the treatments in each arm." -- Harrison (1996)
There is obvious value in seeking patient input and using that input. They may have a different perspective on equipoise than you do.
Lay people have helped researchers to think through the implications of the results of research. For example, I was pleased when a large randomised trial showed that intensive monitoring of babies during labour reduced their likelihood of having seizures after delivery, because this hypothesis had been derived from my first attempt to prepare a systematic review of controlled trials. Women's comments on the trial helped me to put the confirmed hypothesis into perspective. For many of them, increasing the chances of a baby not having seizures from 996 per 1000 to 998 per 1000 (with no evidence that this would be reflected in any more substantive beneficial effect in the longer term) was simply not an adequate incentive to accept the encumbrance of being connected to intensive fetal monitoring equipment during labour. -- Chalmers (1995).
An area of special controversy is the use of placebos in developing countries. A better treatment may be available, but it may not be available locally. An example was a series of trials in the 1990s in Africa that evaluated short term dosage of zidovudine to prevent mother to child transmission of HIV. Half of the patients received placebo, and that was consistent with the local standard of care at the time. But in developed countries, it would have been unethical to conduct this experiment, because the standard of care already in place was superior to placebo.
In communities where there is no access to treatment of any sort, the concept of best available treatment (as defined by rich countries) is meaningless. Health research in poor countries should be designed and conducted pragmatically, taking into account local health needs and priorities. -- http://bmj.com/cgi/content/full/325/7368/796/a
Another example was the testing of Malarone. The testing was done on natives of Kenya, Indonesia, and Gabon, but the benefits would be restricted to people travelling to countries with endemic malaria.
The concern is that developing countries might exploit undeveloped countries to circumvent ethical restrictions on the research. At a minimum, these studies need to involve the views of the people in the countries where the research is taking place.
Special placebo designs
There are also special situations where the use of placebo might be appropriate, even when there is no equipoise. These situations include:
- short term studies where a delay in treatment poses no substantial risk, or
- studies where patients who deteriorate on the placebo are allowed to switch to the treatment.
It is often possible to design a successful placebo-controlled trial that does not cause investigator discomfort nor raise ethical issues. Treatment periods can be kept short; early "escape" mechanisms can be built into the study so that subjects will not undergo prolonged placebo-treatment if they are not doing well. In some cases randomized placebo-controlled therapy withdrawal studies have been used to minimize exposure to placebo or unsuccessful therapy; in such studies apparent responders to a treatment in an open study are randomly assigned to continued treatment or to placebo. Subjects who fail (e.g., blood pressure rises, angina worsens) can be removed promptly, with such failure representing a study endpoint. -- U.S. Food and Drug Administration
Rothman and Michels (1994) disagree strongly with the concept that a short term delay in treatment is acceptable.
First, one can argue that withholding an accepted treatment may not lead to serious harm. For example, treating pain or nausea with a placebo may cause no long-term adverse effects, and the patient can call attention to any treatment failure or even choose to drop out of the study. Nevertheless, although withholding an accepted treatment may occasionally seem innocuous, allowing investigators to do so runs counter to the ethical principle that every patient, including those in a control group, should receive either the best available treatment or a new treatment thought to be as good or better. Instead, it concedes to individual investigators and to institutional review boards the right to determine how much discomfort or temporary disability patients should endure for the purpose of research. Ethical codes in medical experimentation have been developed expressly to shield patients from such vulnerability. -- Rothman and Michels (1994).
The World Medical Association, the body which drafted and updated the Declaration of Helsinki, recently issued a clarification that addresses the issue of harm:
The WMA is concerned that paragraph 29 of the revised Declaration of Helsinki (October 2000) has led to diverse interpretations and possible confusion. It hereby affirms its position that extreme care must be taken in making use of a placebo-controlled trial and that in general this methodology should only be used in the absence of existing proven therapy. However, a placebo-controlled trial may be ethically acceptable, even if proven therapy is available, under the following circumstances:
- Where for compelling and scientifically sound methodological reasons its use is necessary to determine the efficacy or safety of a prophylactic, diagnostic or therapeutic method, or - Where a prophylactic, diagnostic or therapeutic method is being investigated for a minor condition and the patients who receive placebo will not be subject to any additional risk of serious or irreversible harm. -- World Medical Association (2001a).
A recent letter to the British Medical Journal criticized this clarification.
I think that patients have a right to be treated. They go to doctors to have their problems alleviated, not to be guinea pigs. -- Bland (2002).
Further clarification of the above clarification came from the president of the World Medical Association, Dr. Delon Human
A good example is where add-on treatment is being tested for the treatment of cancer, while the patients still receive their standard therapy. Another situation is where research is done to find more effective treatments for a minor condition, such as baldness or allergic rhinitis. For this type of clinical situation there would be no additional risk or irreversible harm for the control group, who would be receiving placebo (no treatment). -- World Medical Association (2001b)
There are other situations where some would argue that a placebo might be acceptable. For example, a placebo would be ethical when the study is restricted to patients who are not responsive to the standard therapy. This choice is not without problems. It is harder to extrapolate from such a study to the general population.
Still another possibility exists when it is unethical to withhold a standard treatment (S). The add-on trial, which compares S plus X to S plus placebo, would still be appropriate. This approach has limitations, of course, in that such a study only measures the effectiveness of X above and beyond the effectiveness provided by S.
A final alternative worth considering is the use of a dose-response design. If it is unethical to withhold a treatment X, then demonstrating the treatment X has increasing efficacy with increasing dosages provides indirect evidence of the general efficacy of X. It would be unethical, of course, to use such a design if it involved dosages that were known a priori to be ineffective.
Over 5,000 HIV negative men and women are participating in a randomized clinical trial of AIDSVAXX, a possible vaccine against the AIDS virus.
Participants receive seven injections, of either the vaccine or a placebo, over 3 years. Participants also receive intense safe-sex counseling each time they come in for an injection. Here lies the paradox, on a personal level, for the counselors themselves. How do the counselors balance their personal desire to keep their clients practicing safe sex with their knowledge that mistakes will happen, and will be good for the study? -- (Klach 2001).
There is a potential conflict of interest here, which needs to be discussed, but from the perspective of equipoise, this is an ethical study. All patients are getting the best proven treatment (counseling). The effect of counseling is to minimize the number of opportunities for the vaccine to demonstrate its efficacy.
The current study is set up to observe a certain number of participants who will theoretically practice unsafe sex. A previous study following a group of predominantly gay men over 18 months using only prevention counseling resulted in 1.5% of the group becoming HIV positive. The AIDSVAX trial is based on observing a similar number of "seroconversions." -- (Klach 2001).
Another way of looking at this is that counseling will dilute the ability of the experiment to demonstrate efficacy (if efficacy does indeed exist). This forces the researchers to rely on a larger sample size. This has consequences beyond financial. A larger sample size means that a greater number of patients receiving a placebo vaccine and a longer period of time to demonstrate efficacy.
In most research, there is an implicit tension between the rights of the individual and the needs of society. The evidence base of a research study is weakened in order to reduce the chance that some individuals are randomized to a treatment that is known or suspected to be inappropriate at the time of randomization. This may mean a delay in discovering an effective treatment for society as a whole. In general, I feel that the rights of the individual to get the best available treatment should trump the need of society for a strong evidence base.
A good example of this tension was a recent study of cellular implants for patients with Parkinson's disease. Blinding is important because some of the outcome measures are subjective and drug studies have shown a very strong placebo effect. So the researchers proposed (and got IRB approval) for a study where some patients were randomly assigned to a placebo operation group. These patients underwent the same operation where a hole was drilled in the skull, but no material was inserted into the brain. This made for a strong evidence base, but were the rights of the individuals ignored? Can you justify exposing a patient to the risks of a placebo operation when there is no direct benefit to that patient?
Lung volume reduction surgery
Making an experimental or unproven treatment available outside the realm of a randomized trial is a controversial issue. The New England Journal of Medicine recently released early the results of a trial of lung volume reduction surgery for patients with severe emphysema (National Emphysema Treatment Trial Research Group 2001). This is a treatment where there was evidence in 1996 of efficacy based on a comparison of 150 consecutive surgeries to historical controls. But the high cost of the surgery and the uncertain evidence basis of the 1996 study caused the NIH and HCFA to bar the use of Medicare funds for this surgery unless the patient agreed to participate in a randomized trial. Patients who wanted the surgery had to either take their chances or pay for the surgery themselves (Drazen 2001). Should patients who trust the results of a non-randomized study be denied the opportunity to a therapy until others have collected more definitive randomized evidence?
It turns out that, at least for some patients, the surgery was actually harmful. Even so, the concept of restricting patient choice seems to be a difficult issue to resolve. William Silverman, in a recently published book, argues strongly against allowing unproven therapies to be available outside a randomized trial.
Cholecystectomy is the surgical removal of gallstones. A surgery group in Germany pioneered the use of laparoscopic surgery for this condition. Laparoscopic surgery involves a narrow incision and uses a video monitor to observe the progress of the surgery.
This surgery group made the decision to defer a the start of a randomized trial comparing laparoscopic to conventional cholecystectomy until the group obtained some experience with the new approach. Instead they conducted an observational study of 100 consecutive patients and compared these patients to historical controls (Neugebauer et al 1991).
Some of the factors that influenced their decision to defer a randomized trial were
- to audit outcome by prognostic factors such as age, sex, diagnostic categories, and severity of the condition and to identify the therapeutic uncertainty justifying a randomized controlled trial,
- all factors contributing to the 'learning curve' had to reach an acceptable degree of variation or plateau which was unlikely to change greatly.
The observational data showed that laparoscopic cholecystectomy provided strong benefits (less post-operative pain and nausea, and a shorter convalescence). Unfortunately, the novelty of the surgery attracted patients from a wide geographic region compared to the historical controls. In particular, 77% of the new patients had a low severity classification, compared to 10% of the controls.
Although it was documented on imperfect data, the surgeons also factored in their personal experiences with the surgery and other published data about the surgery and decided that a randomized trial was not justified.
The authors propose the use of a comprehensive surveillance system as the only tenable option to a randomized trial.
What would have been the consequences if the surgery group had decided to initiate a randomized trial with the very first patient? What would have been the consequences if they had decided to initiate a randomized trial after the first 100 non-randomized patients?
If you want to use a placebo, you need to examine the scientific relevance of the findings. You also have to consider situations where you need to reveal blinding information when a specific patient may need to know whether he or she is receiving the placebo or the active treatment. You also need to consider the medical and ethical implications of denying a treatment to some of your patients.
Equipoise, or uncertainty about what the best treatment might be, must be present for a trial to be ethical. There is considerable controversy about whether this uncertainty should be at the community level or at the level of the individual physician.
There are some design modifications for placebo experiments. An interim analysis will examine at pre-specified intervals whether a research study should be terminated early. Adaptive randomization changes the randomization probabilities as information accumulates during the trial.
Finally, as a health care professional, you are responsible for providing the best standard of care for each patient. This needs to be balanced, however, against patient wishes and desires.
This page was written by Steve Simon while working at Children's Mercy Hospital. Although I do not hold the copyright for this material, I am reproducing it here as a service, as it is no longer available on the Children's Mercy Hospital website. Need more information? I have a page with general help resources. You can also browse for pages similar to this one at Category: Placebos in research.