Policy statement on the use of the randomized trial
A paper prepared by John Heron for the Research Council for Complementary Medicine, London, 1983.
It has been classically acknowledged (Hill Br Med J 1963 ii 1043-1049) that it is only appropriate to use the randomized trial in medical research when it is uncertain which of two treatments being compared is the more efficacious, when there is a genuine 50-50 chance that either is as effective as the other. There are two issues here. Firstly, how often is this condition met within orthodox medical research? Secondly, is it a condition that can be met when you are wanting to compare not simply two treatments within the same overall medical paradigm, but two treatments each of which belongs to a different medical paradigm? A positive answer to the second question at least requires that the two different medical paradigms, for example allopathy and homoeopathy, share the same criteria of effective outcomes. But there are deeper moral problems.
The randomized trial is easiest to implement, and most valid methodologically, when patient consent to enter the trial is not sought. For then there is no loss of accrual to the trial by patient dissent, no problem in recruiting sufficient doctors to participate, no issue about disturbing patient confidence in the doctor-patient relationship, and no interfering test variable arising from the information-giving activities involved in seeking consent. Hence patient consent in this country is often not sought, or is at best thought to be “desirable but not obligatory” (Cancer Research Campaign Working Party Br Med J 1983 286 1117-1121).
We consider that patient consent is obligatory before entry into a randomized trial. Respect for personal autonomy has priority over the claims of statistical argument. And respect for patient autonomy entails that practitioners honour the right of each patient to fully informed self-determination in the selection of available treatments. Hence prospective entrants to a trial should be told that participants in the trial will have their treatments selected by a randomizing process, and told what the alternative treatments involved in the trial are. The patient can then either agree to randomized selection, or decide not to enter the trial and request one of the treatments involved in it.
It is sometimes argued that since doctors do not know which is the better of two or more treatments in the trial, nothing material is withheld if patients entering the trial are not told about the random selection of treatments. But this rather shockingly assumes that it is only the medical view about the status of the treatments in the trial that counts. We consider that patients have an important right to exercise their reasoning, hunches, preferences in relation to the treatments being tested, even if this leads to the loss of accrual to the trial and to the exercise of irrational preference (Cf Schafer N Eng J Med 1982 307 719-724). Often patients are not told about randomization precisely because it is feared that they may opt for self-determination in the selection of treatment; but such a practice of not telling makes the researchers morally subversive of patient autonomy.
If you use randomized trials to compare conventional with complementary therapies, you are asking entrants to the trial to forgo the exercise of their intelligent self-determination in the selection of available treatment to a rather extreme degree. For you are not simply asking them to agree to be randomly allocated to different treatments within the same medical paradigm or overall medical philosophy; you are asking them to agree to be randomly allocated to treatments within quite different medical paradigms. It would seem educationally and medically immoral not to explain fully to prospective entrants to the trial the difference between the two paradigms; but the result of such explanation might be massively to disrupt and skew accrual to the trial.
We acknowledge that if patient consent to entry into a randomized trial is sought on the basis of full information about the treatments to be tested and the randomized selection of them, then this tends to be subversive of the trial. If many prospective trial patients opt out, then accrual to the trial is undermined and accrual can no longer claim to be random but is biased in favour of those who opt in. Prospective participating doctors may drop out at the thought of all the consent-seeking activities involved and their possible effects on doctor-patient relationships. Furthermore, all the information-giving and consent-seeking may constitute a treatment and test variable that itself clouds the experimental treatment effect. We think these subversive tendencies should be accepted: the moral principle of patient autonomy has precedence over the statistical rationale for the randomized trial. If morality and methodology conflict, it seems to us that the onus is upon us to develop methodologies that harmonies with our morality, rather than compromise with morality on the probably false assumption that we are dealing with an immaculate methodology.
We consider that the randomized trial does have further limitations. Firstly, randomization generates statistical information by trying to make patients as homogeneous as possible, and this by controlling for all sorts of “irrelevant” variables other than the one primary variable being studied. But patients, of course, are highly heterogeneous: some patients with certain personal characteristics may benefit from one kind of treatment, and other patients with other sorts of characteristics may benefit from another kind of treatment. But a randomized study of these two treatments might show no statistically significant difference in their efficacy. Thus randomization tends to “obscure, rather than illuminate, interactive effects between treatments and personal characteristics” (Weinstein N Eng J Med 1974 291 1278-1285). Now the practical question is often what is the treatment of choice for this individual and idiosyncratic patient. But the data derived from randomized clinical trials cannot help with this question, since randomization is designed to eliminate any effect from idiosyncratic variables.
Secondly, just as the patient’s right to self-determination in the selection of available treatments is an embarrassment for the randomized trial, so too is the patient’s potential as a consciously self-healing agent. Biofeedback, autogenic training and related approaches indicate that persons can exercise conscious self-regulation of physiological states in ways that promote well-being. If some patients in a randomized trial are privately exercising such self-regulation, then its effect will simply be eliminated by the randomized design since it will fall within the general class of “nuisance” variables. But if you set up arandomized trial to test the relative efficacy of conscious self-regulation as against some other kind of treatment, then you are in a quandary indeed. On the one hand it is psychologically odd for a patient to be randomly allocated to the practice of conscious self-regulation since this is surely an activity whose effectiveness depends in part upon its being freely chosen. On the other hand, it is morally dubious, if you practice the seeking of informed consent, to ask patients to agree to the possibility of being randomly allocated to atreatment in which they are not to practise conscious self-regulation, and this when there is already evidence that in some other fields such self-regulation is efficacious. Would, and should, any intelligent ill person agree not to attempt intentional self-healing? And should any ill person ever be asked not to attempt such mental self-help?
Thirdly, for all the above reasons, and as long as the randomized trial is practised as the central paradigm of medical research, it will tend to sustain and breed aculture of patient alienation, in which patients are conditioned to see their diseased bodies as cut off from the exercise of self-determination and of intentional self-healing, and from the influence of idiosyncratic personal characteristics.
Special difficulties arise when considering the use of randomized trials in the field of complementary medicine, at any rate with respect to their use on patients (non-patients are adifferent case, see below). Suppose you want to use arandomized trial to compare the effects of a conventional form of treatment or of no treatment with some complementary form of treatment. In the case of osteopathy and acupuncture, you cannot do the trial blind, and the results will not enable you to differentiate between the effects of the physical osteopathic or acupuncture technique used and the effects of the osteopath or acupuncturist qua person. You may say this does not matter, and that you simply want to know whether the complementary sort of treatment is better or not, however its outcomes are achieved. Yet this seems less than satisfactory.
In the case of homoeopathy and herbalism, you can certainly do the trial double blind and thus claim to be getting at an independent physical effect, but if the trial is done likethat of an allopathic drug, all the patients in the experimental group would get the same homoeopathic or herbal remedy. It could then be argued that this is no test of either homoeopathy or herbalism. Indeed it has already been argued (Ives Br Hom J 1983 72 224-228) that a study by Shipley et al (Lancet Jan 15 1983 97-98) comparing Rhus Tox 6x with phenoprofen and placebo in the treatment of arthritis “only poorly approximated to normal homoeopathic practice, in that it would be unusual to treat a chronic condition with only a single remedy in low potency” (Ives op.cit.). For in homoeopathy especially, and to a lesser extent with herbalism, the prescription for a chronic condition is tailored to the idiosyncratic patient as a whole: with regard to patients with the same disease, each may receive a significantly different series of prescriptions. You could deal with this by having all entrants to a trialundergo thorough homoeopathic history taking, then randomize some into aplacebo group and some into a group whose members receive idiosyncratic individually tailored remedies. If the trial is administered double blind, and the experimental group show a significantly better outcome, then it is reasonable to infer there is a homoeopathic effect that is due neither to elaborate history taking nor to placebo; although no one homoeopathic drug in only one potency is being tested. However, “in certain acute conditions the choice of remedy may be sufficiently narrow, or the criteria for selecting a remedy sufficiently obvious, for a double blind trial along traditional lines to be feasible”
We consider it would be entirely improper to argue that the randomized trial has no place in research into complementary medicine using groups of patients. But we do consider it is a method which tends to contradict much of the legitimate ethos of complementary medicine – its emphasis on self-determination, self-help and self-healing, on treatment tailored to the patient as idiosyncratic whole person. And it is a method which within any kind of medicine is morally and medically problematic.
We think therefore there is a strong case for considering the claims for using non-randomized controls in research on complementary medicine. Such selected controls may be historical controls, chosen from the literature, from preceding trials, or matched from a previous study. Or they may be selected from matching patients being treated concurrently. The case for the use of selected, non-randomized controls in comparative studies in cancer clinical trials in conventional medicine has already been argued for (Gehan and Freireich N Eng J Med 1974 290 198-203). And other papers in the medical literature have pressed the case (Weinstein N Eng J Med 1974 291 1278-1285; Dudley Br Med J 1983 287 957-960). But where selected controls are patients being treated concurrently, we do think their informed consent should be sought. The most non-problematic use of non-randomized concurrent controls from a moral point of view is where, for example, astudy compares treatment given to patients who are voluntarily attending a conventional medical centre with treatment given to patients who are voluntarily attending a complementary medical centre. We think that researchers and statistician should work at legitimating such a study rather than rule it out dogmatically on the grounds of non-randomization.
The strength of the randomized trial, especially if it is double blind, is that it is an elegant way of demonstrating whether the treatment variable does have a physical effect relatively independent of other variables including psychosocial variables. And it is certainly reasonable to ask whether homoeopathy and herbalism do have such an effect, however important psychosocial variables in the treatment may also be. We think there may well be a case for the use of randomized trials with non-patients in order to test for relatively independent physical effects of homoeopathic and herbal substances. The argument is as follows.
The claim of complementary therapies such as homoeopathy and herbalism, osteopathy and acupuncture, is that they work on disease indirectly by means of a catalytic effect on the body’s self-regulatory functions, whereas conventional treatments using drugs and surgery directly confront diseased part or process seeking to eliminate it, restructure it, subdue it, re-educate it or whatever. Hence homoeopathic and herbal remedies may well show an effect on the physiological functions of non-patients, and for some remedies that effect may be relatively independent of other variables. Hence given suitable assessment techniques for regulatory changes in physiological functions, with pre- and post- testing on a randomized trial with non-patients, double-blind using a placebo control group, such a relatively independent physical effect may be demonstrated.
Such a trial is ethically sound, since the participants are non-patients, provided their consent is sought; and provided that it is reasonable to infer from all available evidence that the remedies used in this context would have no undesirable physical consequences or side-effects. If the trial were successful it would certainly show that the remedies used in it had more than a placebo effect on physiological functions in non-patients. It would demonstrate a homoeopathic or herbal effect. But it would not per se entail any practical consequences for treating patients. And it would be false to assume that any general effect of a given remedy on non-patients would detract from the importance of particular, idiosyncratic prescribing for patients.
Randomized trials on consenting non-patients can also be used to compare the effects of attempts at conscious self-control on physiological processes with the effects of doing nothing or doing something different. This still suffers from the anomaly of randomly allocating a person to practise conscious self-regulation – which is presumably most effective when it is chosen out of a free commitment. On the other hand this anomaly may be much less critical with a person who is well than with a person who is ill.
An interesting use of the double blind randomized trial is internal to the belief-system of homoeopathy. It is in relation to the practice of provings: non-patients take somewhat larger doses of a drug, note down what disease elements it tends to produce in them, then the principle of similia holds that patients who show similar disease elements will tend to be cured by minute or potentised doses of the drug. In a trial on a proving, some healthy persons would take moderate doses of a drug, others would take a placebo, on a random double blind basis, in order to see whether the “disease elements” produced by the drug are independent of a placebo effect (see Clover et al Br Hom J 1980 69 134-149). This inquiry is problematic: the sort of full explanation involved in seeking informed consent for entry into the trial may well on its own induce a placebo effect. Again those already committed to homoeopathy are most likely to assent to enter the trial, and this prior commitment may again induce a placebo effect by itself. If you push the dos of the drug up high enough to transcend placebo effects demonstrated on earlier trials with lower doses, then you start to need heroic and highly committed entrants to the experiment who are willing to undergo possibly very unpleasant physical and mental discomfort.
Finally, randomized trials could be used with animal subjects in the laboratory and in veterinary practice, to test for a homoeopathic, herbal or acupuncture effect as against no treatment or conventional treatment. The complementary therapies used will not, it is reasonable to suppose, do any harm to the animals; there are no ethical problems to do with self-determination and informed consent of the subjects (although the consent of their owners is another matter). And while extrapolation from animal studies to human beings remains problematic, at least a determinate effect will or will not be demonstrated.