Shifting the burden of proof regarding placebo controls

[Blog post I wrote for the Bill of Health, a blog from the Harvard Law School]

Placebo controls usually get special treatment. Almost all ethical guidelines for research specify how to use placebo controls, in particular when an established or proven effective treatment exists. To list just a few: guidance issued by the World Medical Association, CIOMS and the Council of Europe all have provisions dedicated to the use of placebo controls. But is there any reason to consider placebo controls a special case? I believe there isn’t, and I also think the burden of proof is on those who hold that placebos should be evaluated differently from other research interventions.

Using a placebo when an established effective treatment exists deprives the control group of the benefits of that treatment. This poses relative net risks to participants – that is, risks from delaying or foregoing a treatment that participants may have received outside the trial. Some placebo interventions also pose risks themselves, but these risks are typically so low – taking a “sugar pill” or receiving a saline injection through an existing i.v. line, for example – that we rightly focus on the risks of foregoing or delaying treatment. (There are other cases, though, such as certain forms of sham surgery. I won’t consider these cases here.)

So how should we evaluate the risks of using a placebo when an established effective treatment exists? We have general criteria for evaluating the risks of research interventions, in particular:

  • The study in which interventions are to be used must meet some minimum threshold of social value, or else it would not be justified to impose risks on participants; and
  • The risks of each research intervention must be reasonably reduced and justified by the potential clinical benefits for the participant and / or the social value of the research; and
  • In aggregate, the risks of research interventions that are justified by the study’s social value should be no greater than minimal if participants cannot give their own consent; and
  • Some level of risk should not be exceeded even if  participants can give valid informed consent.

The logical thing would be to use the same criteria for evaluating the risks of placebo controls. But placebo controls are treated as a special case in most existing ethical guidance, seen to require special consideration and limits on risks.

The 2002 CIOMS guidelines are an example. CIOMS Guideline 11 states that placebos can only be used if there is no established effective intervention or withholding the established intervention would either result in “at most temporary discomfort or delay in relief of symptoms” or “not (…) any risk of serious or irreversible harm to subjects”. On most standards of minimal risk in clinical research, temporary discomfort and the absence of serious or irreversible harm would qualify as minimal risk. But the general criteria for evaluating research risks apply the minimal risk threshold only to research with participants who cannot give their own informed consent, not to all research.

It seems to me that the burden of proof is on those who believe that the use of placebo controls requires specialconsideration, i.e. consideration that differs from those general criteria for evaluating research risks.

So what might justify treating placebos as a special case? The standard argument is to invoke the doctrine of clinical equipoise, which holds that investigators can only randomize participants to an intervention if there is “honest, professional disagreement among expert clinicians about the preferred treatment”. This doctrine excludes the use of placebo controls when there is a proven or established effective treatment (and hence it excludes any level of “relative net risk” from withholding or delaying established effective treatment). But, as Frank Miller has forcefully argued in many articles, the doctrine of clinical equipoise extends the obligations of clinicians to the context of research, even though the norms of clinical practice do not apply in this context.

Worse: even if equipoise could be defended, it would be inconsistent with most existing guidance. Current guidelines, such as CIOMS, hold that placebo use is acceptable when foregoing established effective treatment poses no more than minimal risks. But if investigators know that the risks of foregoing treatment are minimal, they know that the study intervention is not as effective as standard treatment. Hence there is no honest, professional disagreement about the preferred treatment, and the requirement of clinical equipoise is not met.

In addition, it is not clear how proponents of equipoise can defend the use of study procedures that offer participants no prospect of clinical benefit – for example, non-clinically indicated blood draws or biopsies that are often an essential component of clinical trials. If an investigator can only use placebos when this poses no “relative net risks” to participants, why would it be acceptable for her to impose risks that are not offset by any potential clinical benefits for the participant?

This suggests that the standard argument for considering placebo controls a special case fails. The burden of proof is on those who support “placebo exceptionalism”.

Importantly, this is not just an interesting theoretical question of but one of real practical import. Existing guidelines set out requirements for placebo controls that seem difficult to justify, and – perhaps rightly – these requirements are often ignored by sponsors and investigators. This weakens ethical guidance and raises concerns about adherence to requirements that rest on a more robust moral foundation. It is high time that we either identify a sound argument for why placebo controls should be evaluated differently from other research interventions, or that we stop treating placebo controls as a special case.