Randomised Controlled Trials
- A randomised controlled trial (RCT) is a type of experiment.
- RCTs involve randomisation, a control group and manipulation of the independent variable.
- RCTs attempt to control bias.
- RCTs are the most effective way of examining cause and effect relationships, including the effect of treatment on patients.
- Explanatory trials possess high internal validity and establish what principle, mechanism or theory accounts for a change in patient condition.
- Pragmatic trials possess high external validity and establish how well a principle, mechanism or theory translates into the real clinical world.
- RCTs are essential in providing patients with accurate information about therapeutic interventions.
In healthcare, the randomised controlled trial (RCT), first adopted in medicine, has come to be regarded as the gold standard in investigating treatment effectiveness. Its use has spread from simple medical and surgical interventions to embrace nursing, allied health professions and social care, and the investigation of complex multi-method healthcare interventions and aspects of service delivery. There are two reasons for this increased use of RCTs. First, in healthcare, it is important for us to have a clear idea about whether treatments we offer are effective. To do this, we need to be equally clear that something other than the treatment we are investigating is not, in fact, causing any improvement in the patient’s health status. If we cannot be clear about these two things, we have no justification in offering the treatment to patients. Second, RCTs are currently the best way of achieving that clarity, because no other research method is as good at controlling for these possible alternative sources of patient improvement. In this chapter, we will explore what RCTs are, how they go about establishing whether or not an intervention works, the issues which researchers need to consider in using RCT methodology and the practicalities of designing and undertaking RCTs.
What is an RCT?
An RCT is a special kind of experiment which investigates the effectiveness of therapeutic interventions with patients. To understand this further, we need to know a little about the nature of experiments, which we examined in detail in Chapter 15. Essentially, experiments attempt to explore whether there are cause and effect relationships between particular events in the world and are regarded as doing a good job of this because they are good at controlling bias. The RCT brings together the elements of experimental research in the context of healthcare research; because the RCT is so central to healthcare research, many of the issues of experimentation discussed earlier are revisited in the current chapter.
As we saw in Chapter 15, bias is controlled by manipulating the independent variable (the thing we think may be causing some effect in the world) between two or more groups and seeing whether this manipulation has an effect on the dependent variable (the thing we think might be being affected). At the same time, bias is further controlled by trying to remove from the picture unwanted intervening or confounding variables which might interfere with our ability to attribute changes in the dependent variable to our manipulation of the independent variable.
In RCTs, the independent variable is some form of health intervention which is varied by the researcher over two or more conditions. Condition here does not mean an illness or health problem. This is simply an unfortunate jargon carry over from experimental method in general and loosely means group. Thus, if we speak of a treatment condition, we mean a treatment group – those people receiving a particular intervention. Now, in an RCT, the intervention is consciously applied to one group of participants (usually, patients) rather than another. This is what is meant by manipulation. It differs from simply observing naturally occurring variations, because we decide what the variation is to be,define it in precise terms and apply it to one group rather than another.
One alternative to manipulation would be to observe naturally occurring variations. We might, for example, look at patients from two different wards where two different models of care were in use and compare the two groups. However, this would be less useful than the manipulation of such models of care between different groups because of the increased potential for bias. The two wards might differ in all sorts of ways which had nothing to do with the model of care employed. For example, all the staff on one of the ward might be more experienced than those on another, or the patient groups might be different in terms of the complaints from which they suffered, their age, sex, or whatever.
These are very obvious differences, and easy to spot, but many such differences are more subtle. Let us say we worked in a setting where a multidisciplinary care pathway was to be introduced sequentially onto two different wards (Wards 1 and 2) with similar staff groups, both of which dealt with identical patients. We wanted to see whether the care pathway had any effect on patient improvement and decided to compare the two wards. It seems as if two major sources of unwanted difference between the wards have been sorted out. However, suppose that, unknown to us, the person responsible for introducing the pathway had chosen to introduce it onto Ward 1 rather than Ward 2 because the staff on Ward 1 were more enthusiastic and open to change. Clearly, their enthusiasm would be a powerful potential confounding variable, and might be just as much the source of any difference between the wards as the care pathway.
This, then, is the first requirement of an RCT – deliberate, conscious manipulation of the independent variable, rather than observation of naturally occurring variability. By contrast with observation, manipulation allows us to control bias by reducing the likelihood that some other difference between the two conditions, rather than the difference between the conditions we claimed was responsible, might be causing any observed differences in response between the conditions. When we manipulate the variable to be studied, we do so precisely in order to isolate, as far as possible, the entity we believe might be responsible for changes in patient status. This factor – the ability to isolate the independent variable from the effects of unwanted variables – is the sole factor which drives the act of manipulation.
Manipulation rather than observation also allows us to stipulate precisely what the independent variable will consist of. So, in the care pathway example, we can develop precise definitions of what constitutes the pathway, what the health professionals concerned will be required to do as part of implementing it and what training they will require in order for us to be confident they are implementing it adequately. At the same time, we can specify what care patients who are not receiving the pathway will receive, and part of that specification will involve ensuring that their care truly is different from that involved in the care pathway. This process is often referred to as specifying, defining or operationalising the independent variable, but all these terms mean the same thing – stipulating exactly what the manipulation will consist of and ensuring that such a manipulation will take place.
Often, the goals of excluding confounding variables and adequately operationalising the independent variable are very hard to achieve, but as a general rule, a manipulated variable (as in an RCT or other experiment) offers far greater possibility for precise definition than a variable in an observational study.
The second element of the RCT is the notion of randomness of the manipulation of the independent variable (intervention) between groups involved. This is closely linked to the idea of manipulation, because it is a way of avoiding the problem we encountered above – the possibility that uncontrolled variables might be responsible for differences between the groups. In the above example, this problem involved selection bias – our HCPs were chosen on the basis of enthusiasm, which might have affected patients. A similar thing might happen if we offered treatment to patients on a first-come, first-served basis, with first-comers receiving the experimental treatment. Unfortunately, first-comers might differ from others in all sorts of ways other than the treatment they go on to receive (more motivated, more affluent, more sick, more desperate, to name just a few). Similar problems arise in almost any kind of non-random sample, because we cannot be confident that members of the two groups are similar, and this kind of subject variability can considerably affect our results.
In the RCT, confounding variables arising from selection difficulties are typically avoided by means of random allocation to treatment or control groups. In all cases, the principal reason for randomisation is to ensure that each member of the entire group of participants has a fair chance of being allocated to either the experimental or control group, and the reason for wanting this fair chance is to ensure that possible differences between individual participants are roughly equally distributed between the groups. Randomisation does not guarantee that this will be the case, but is a necessary minimum safeguard against the possible confounding effects of subject variability.
However, even the act of randomisation is prone to potential bias, and, as a result, specific protocols for randomisation exist. Traditional methods of randomisation, such as the use of putting cards bearing allocations to the different conditions into sealed envelopes have been found to be subject to abuse, as the people carrying out the randomisation have sought to influence allocation (e.g., by holding the envelope to the light in order to see its contents without opening). Partly, in consequence, the gold standard for randomisation is via remote, telephone randomisation using computer-generated allocation and performed by someone unconnected with the clinical work and the conduct of the rest of the study. This approach is resilient to abuse, as far as we know.
The final important feature of RCTs is the ability to assign participants to a control group. In experiments, usually only one group receives the experimental procedure, whilst the other receives some procedure which is in all respects similar to the experimental one, but lacks the novel element. However, in complex studies, there may be many novel treatments in a single study. Each treatment will be clearly defined and offered to one group only, so that there will be no overlap between the treatments received by each group. Finally, there will be a group of individuals who receive some intervention which is not novel. These are the control group and are the standard against which the novel treatments are judged. Typically, the most important question in an RCT is whether or not there is a difference between the novel treatment (experimental group) and the control treatment (control group). In a well-conducted RCT, the experimental and control groups will vary only by virtue of the fact that they receive differing treatments. These different treatments will be so well specified that they represent some single, clear difference in treatment.
Confusingly, the control treatment may, in fact, be no treatment at all, although this is generally regarded as the weakest form of control. This is because no treatment controls in RCTs do little to account for non-specific effects of treatment or from unwanted effects of experiments in general. For example, it is well known in psychotherapy that such issues as the personality, social skills, enthusiasm of a therapist can exert some therapeutic influence. In healthcare and medicine, the communication skills of the clinician can affect compliance with therapeutic regimes and, therefore, indirectly, the effectiveness of such regimes. More broadly, in research as a whole, the effect of attention on research subjects is itself known to influence their responses. In no-treatment control RCTs, it is often impossible to tease out non-specific effects such as attention, from the effects of the independent variable.
One of the commonest tactics, in assigning patients to a control group, has been the use of a waiting list control. This is simply one example of a no-treatment control group, but is particularly weak, because not only are the control group receiving none of the non-specific effects of being in any treatment, but they are also expecting to receive treatment in the future, which may give rise to any number of thoughts, feelings and behaviours which would affect their scores and erroneously lead us to believe that differences between them and people in the experimental group arose from specific effects of the novel treatment. However, on the plus side, it must be noted that no-treatment controls are immeasurably better than no controls at all! Moreover, there are many situations where no-treatment controls of one kind or another are the only practical solution.
Sometimes, it is possible to introduce some form of attention control, which attempts to offer participants all the non-specific aspects of treatment without the specific component under investigation and give participants the same amount of attention (e.g., through the interviewing and the assessment process) as they would receive in active treatment. There are two problems with this tactic. The first is an ethical one, in that the researcher will not wish to waste participants’ time by offering what are essentially ‘filler’ treatments – activities which fill time in an apparently meaningful way which has no demonstrable benefit. The second problem is purely practical. Even if it were ethical to offer patients non-specific interventions to control for the non-specific and attention effects of a particular novel treatment, we might run into difficulties finding a non-specific intervention which was convincing to patients. Administering an unconvincing ‘filler’ treatment might have quite the reverse effect from our intention of offsetting non-specific therapeutic effects as it would give rise to non-specific negative effects in the control group.
In a great deal of healthcare research, but also in much health service research which examines complex healthcare interventions, a tradition has arisen of using best current available treatment as the control intervention. There are considerable advantages to this. By and large, the best current available treatment will be what the patient is currently receiving, although this may vary from place to place. It will be a credible alternative, both to patients and clinicians. It will contain broadly the same non-specific aspects as the novel treatment. Finally, it will represent a ‘gold standard’ against which to judge the novel treatment.
Explanatory versus pragmatic trials
In Chapter 13, we examined internal validity (essentially the amount of confidence we can have that changes in the dependent variable are caused by the action of the independent or that lack of such changes are caused by lack of action of the independent variable) and external validity (basically, the likelihood that things we have observed in an experiment are likely to be transferable to similar situations away from the experimental environment).
There is always a tension between these two components of validity in experiments. Increasing the internal validity of a study inevitably means decreasing its external validity and vice versa. Let us illustrate this point by returning to the example of the care pathway. We will want to be certain that patients on the pathway truly receive care in line with the care pathway, and will take certain measures to ensure we can assert this. For example, we will ensure proper training in the use of the pathway for all staff, and perhaps give them a brief examination to ensure the training has been effective. This is known as a manipulation check and is a design strength. We will also ensure that the patients on the pathway are only cared for by doctors, nurses, healthcare assistants, physiotherapists and occupational therapists who have undergone this training. All this is fine and is an important factor in establishing the internal validity of the RCT, because it increases our confidence that differences between pathway and non-pathway patients are truly caused by the presence of the pathway. However, the sceptic can easily argue that our RCT lacks external validity because it does not represent how real care pathways are practiced in the clinical world. In this world, they may argue, training is imperfect and healthcare workers swap between different care settings, they also forget what they have learnt and are not being constantly primed to remember it by being given tests of knowledge. They would further argue that, to be important, a care pathway should be sufficiently robust to withstand these interferences.
So, which view is right? The answer, as with so many aspects of clinical research, is that both are. We want certainty that it truly is the treatment that is causing changes in patient well-being (so far as anyone can have such certainty) and it is the presence of internal validity which allows this certainty. But we also want a treatment which is valuable to patients in general, not just to that small subgroup who take part in the artificial setting of RCTs. We need high external validity to ensure that. Accordingly, RCTs are generally divided into explanatory trials, which aim to identify some underlying principle of treatment, and pragmatic trials, which aim to investigate that principle in the real world of everyday practice and see if it still applies there. In our pathway example, an explanatory trial would try to do at least the following:
Explanatory RCT of multidisciplinary care pathway
Isolate the key elements of a pathway
Incorporate these into the experimental condition of the study
Ensure the control condition did not contain any of them
Ensure adequate training of staff in the experimental condition in using the pathway
Monitor use of the pathway throughout the trial
Control non-pathway elements of care (e.g. patients’ consultant) across the conditions
Control non-pathway characteristics of staff (e.g. age, qualifications, experience) across conditions
Control patient characteristics (age; severity of complaint)