A recently published paper discusses placebos used in randomised controlled trials, referring to the placebo medication as “the Unknown Variable in a Controlled Trial”. The short but important paper is worth reading. It made me reflect on the nature of the control in rehabilitation trials – sometimes even referred to as a placebo by the authors. That reflection, coupled with the major difficulty most authors have understanding the simple logic of a randomised trial, made me consider the very title, a randomised controlled trial. Many authors seem to have no idea what the word control means. This blog explores these ideas.
A randomised controlled trial design’s logic is that only one specific item differs between the two groups, and that the control controls for all the other factors associated with the item, such as attention from doctors, the colour of the tablet and so on. But the paper raises one central, and crucial question – what does one actually control for.
To take an example, suppose the active ingredient is an antibiotic that is supposed to be much more effective against some bacterium. The control could be another antibiotic, the most effective one known, which avoids any ethical issues. Suppose that the new antibiotic makes urine green, but the old one does not. Do you add something to the other, control pill to give the same effect. And if the new antibiotic also gives most people very disturbed sleep, or gastric pain, how is that controlled for?
So the first point made by the paper is that pharmaceutical firms should disclose both what is being controlled for, and the nature and content of the control treatment, but they often do not and will not. The paper gives examples of some ‘placebos’ that may have worsened a patient’s state (not intentionally), giving the mistaken outcome that the trial medication was effective.
The second point made was that control tablets that are completely inert will not control for all aspects of the active preparation. Obvious incidental effects associated with the active ingredient may reveal which drug someone is taking. Or the experimental drug may be heavier, or have a distinct colour or taste or consistency.
The conclusion is that a randomised. controlled trial needs to state what aspects of the experimental treatment are being controlled for, and how. It also needs to state what other actions the controlling treatment has that the experimental treatment does not have. In the example given in the paper, the control ingredient, a mineral oil, controlled for the fish oil in the active preparation. Unfortunately the mineral oil itself had adverse effect that the fish oil did not.
In most trials of drugs, the patient is usually unlikely to be exposed to the active ingredient in any other way. One can at least be reasonably certain that the experimental group will not be having more of the ingredient, and also one can be reasonably certain that the control group will not be exposed to the active ingredient.
In rehabilitation the problems are greatly magnified. I will outline three of the many problems.
It is not obvious what the ‘active ingredient’ of a rehabilitation intervention is. For example if someone is asked to walk three miles extra every day, is the ‘active’ ingredient:
- the cardiorespiratory effects associated with walking for one hour or so?
- the practice of walking over rough ground?
- the exposure to the countryside?
- talking for an hour with a friend while walking?
- relaxing and thinking while walking?
- the exposure to sunlight (vitamin D)?
- the imposition of a routine upon an otherwise unstructured day?
- the feeling of control and achievement?
- the simple realisation that the asking person thinks the patient can safely walk three miles without risk?
How would you control for all of those? Or even for one!
Furthermore, almost all rehabilitation interventions are simply doing a bit more of what someone usually does. Most interventions are measured in terms of maybe 3-5 hours per week at most of some activity that will already be occurring. Some people may already walk for an hour every day, and they will be doing other things such as going up and down stairs. So for some people the extra will be proportionally quite large and for others quite small.
In many respects, it is surprising that trials of rehabilitation interventions ever show a benefit, given the small proportional changes that are made and the small proportional differences there will be between the groups.
It also must be recognised that rehabilitation can have unexpected negative effects. Education is often used as a control intervention in rehabilitation trials. But, in a randomised trial involving 114 people with progressive neurological conditions, which compared giving a standardise information leaflet with a visit by an occupational therapist who gave personalised education and information and set a 12 month education plan, it was the active educational group who did worse. (see here).
One important step might be to refer to randomised comparison trials, which recognises that one is comparing two (or more) options. This would lead to the logical conclusion that one intervention is better than, or equivalent to, or worse than another intervention rather than the usual conclusion that one treatment is (or is not) effective. The reader will need to judge whether one intervention caused relative harm, or the other caused relative benefit. It would prevent the oft-stated conclusion that “both treatments were effective“, because there is no comparison with any third alternative.
A second important step in rehabilitation research, just as is now being proposed for trials of medication, is to give an equally full description of both (or all) interventions, without assuming that any is ‘inactive’. This might be done using the TIDieR framework (see here), though this would be quite extensive and one still does not know what is active.
Interestingly, as I wrote this blog, I discovered that I wrote about control in rehabilitation trials in 2009! What I wrote then (see here) is not dissimilar to what I am saying now, except I did not propose changing controlled into comparison. However I did propose a rather simpler descriptive framework, possibly more relevant to rehabilitation.
In summary, randomised studies of rehabilitation interventions should refer to Randomised Comparison Trials. They will involve two or more interventions, and each intervention should be described in a similar way to the same level of detail. The conclusion should be relative, saying whether the interventions had a similar net effect, or one had a better outcome when compared with the other. No absolute conclusion should be drawn from the trial alone.