The de-randomised clinical trial (dRCT): how we might be fooled by seemingly rigorous research

Controlled clinical trials are methods for testing whether a treatment works better than whatever the control group is treated with (placebo, a standard therapy, or nothing at all). In order to minimise bias, they ought to be randomised. This means that the allocation of patients to the experimental and the control group must not be by choice but by chance. In the simplest case, a coin might be thrown – heads would signal one, tails the other group.

In so-called alternative medicine (SCAM) where preferences and expectations tend to be powerful, randomisation is particularly important. Without randomisation, the preference of patients for one or the other group would have considerable influence on the result. An ineffective therapy might thus appear to be effective in a biased study. The randomised clinical trial (RCT) is therefore seen as a ‘gold standard’ test of effectiveness, and most researchers of SCAM have realised that they ought to produce such evidence, if they want to be taken seriously.

But, knowingly or not, they often fool the system. There are many ways to conduct RCTs that are only seemingly rigorous but, in fact, are mere tricks to make an ineffective SCAM look effective. On this blog, I have often mentioned the A+B versus B study design which can achieve exactly that. Today, I want to discuss another way in which SCAM researchers can fool us (and even themselves) with seemingly rigorous studies: the de-randomised clinical trial (dRCT).

The trick is to use random allocation to the two study groups as described above; this means the researcher can proudly and honestly present his study as an RCT with all the kudos these three letters seem to afford. And subsequent to this randomisation process, the SCAM researcher simply de-randomises the two groups.

To understand how this is done, we need first to be clear about the purpose of randomisation. If done well, it generates two groups of patients that are similar in all factors that might impact on the results of the study. Perhaps the most obvious factor is disease severity; one could easily use other methods to make sure that both groups of an RCT are equally severely ill. But there are many other factors which we cannot always quantify or even know about. By using randomisation, we make sure that there is an similar distribution of ALL of them in the two study groups, even those factors we are not even aware of.

De-randomisation is thus a process whereby the two previously similar groups are made to differ in terms of any factor that impacts on the results of the trial. In SCAM, this is often surprisingly simple.

Let’s use a concrete example. For our study of spiritual healing, the 5 healers had opted during the planning period of the study to treat both the experimental group and the control group. In the experimental group, they wanted to use their full healing power, while in the control group they would not employ it (switch it off, so to speak). It was clear to me that this was likely to lead to de-randomisation: the healers would have (inadvertently or deliberately) behaved differently towards the two groups of patients. Before and during the therapy, they would have raised the expectation of the verum group (via verbal and non-verbal communication), while sending out the opposite signals to the control group. Thus the two previously equal groups would have become unequal in terms of their expectation. And who can deny that expectation is a major determinant of the outcome? Or who can deny that experienced clinicians can manipulate their patients’ expectation?

For our healing study, we therefore chose a different design and did all we could to keep the two groups comparable. Its findings thus turned out to show that healing is not more effective than placebo (It was concluded that a specific effect of face-to-face or distant healing on chronic pain could not be demonstrated over eight treatment sessions in these patients.). Had we not taken these precautions, I am sure the results would have been very different.

In RCTs of some SCAMs, this de-randomisation is difficult to avoid. Think of acupuncture, for instance. Even when using sham needles that do not penetrate the skin, the therapist is aware of the group allocation. Hoping to prove that his beloved acupuncture can be proven to work, acupuncturists will almost automatically de-randomise their patients before and during the therapy in the way described above. This is, I think, the main reason why some of the acupuncture RCTs using non-penetrating sham devices or similar sham-acupuncture methods suggest that acupuncture is more than a placebo therapy. Similar arguments also apply to many other SCAMs, including for instance chiropractic.

There are several ways of minimising this de-randomisation phenomenon. But the only sure way to avoid this de-randomisation is to blind not just the patient but also the therapists (and to check whether both remained blind throughout the study). And that is often not possible or exceedingly difficult in trials of SCAM. Therefore, I suggest we should always keep de-randomisation in mind. Whenever we are confronted with an RCT that suggest a result that is less than plausible, de-randomisation might be a possible explanation.