Randomised controlled trials in general practiceBMJ 1995; 311 doi: http://dx.doi.org/10.1136/bmj.311.7017.1382 (Published 25 November 1995) Cite this as: BMJ 1995;311:1382
- Mike Pringle,
- Richard Churchill
- Professor Lecturer in general practice Department of General Practice, Medical School, Queen's Medical Centre, Nottingham NG7 2UH
Gold standard or fool's gold?
Randomised controlled trials are a widely accepted means of applying experimental methods to a clinical setting and have been advocated as the gold standard for comparing and evaluating different treatments.1 2 General practice has been promoted as an appropriate arena for evaluating interventions ranging from drug treatment to developments in services3 4: nine out of 10 health service consultations take place here; it is the point of first contact for most medical conditions; and existing population registers and computer systems allow potential access to large amounts of clinical data. An increasing number of randomised controlled trials are being performed in this setting to contribute to the culture of evidence based medicine.5 But despite the theoretical attractions, applying experimental methods in clinical practice presents problems that, if not properly addressed or acknowledged, may invalidate the findings.6 The particular problems of recruitment and randomisation merit consideration with respect to primary care.
Recruitment to randomised controlled trials is justified in situations of genuine clinical uncertainty. Sample sizes must be large enough to establish the presence or absence of a worthwhile benefit in terms of either effectiveness or cost, or both. This may mean the need for larger numbers of patients than are available to single general practices, requiring practices to club together if they are to perform meaningful research. General practitioners, however, have no contractual obligation to participate in research, and they may be unwilling to take part in studies that produce no immediate benefit for their patients while possibly disrupting the delivery of health care. Practitioners who do participate are often atypical, so that extrapolating their results to the general population may be misleading. Maintaining the motivation and involvement of participating practices can be difficult in long term studies.7
General practitioners may be unwilling to participate if they believe that experimental studies will disrupt the normal interaction between them and their patients.8 They may experience a conflict of interest between their role in promoting patients' autonomy and their wish to recruit participants to benefit future patients or to gain academic merit. The need to obtain informed consent will not necessarily erase their anxiety about such conflicts, since the long term nature of the relationship between practitioner and patient may engender loyalties that unfairly coerce patients to give their consent.9 From the patients' point of view, fears about confidentiality, the risks of the intervention, or the apparent disadvantage of being allocated to a control group may further inhibit recruitment. Failure to recruit consecutive patients introduces the potential for selection bias, something that is not often reported in published studies4 but that can make extrapolating the results to the general population inappropriate.
General practitioners may feel uncomfortable in randomising patients themselves,10 but if they delegate the task to a researcher this can further compromise the doctor-patient relationship. The alternative of randomising interventions by practice (cluster randomisation) introduces analytical problems, which ultimately require larger sample sizes.11 Furthermore, some interventions such as counselling need a high degree of involvement on the part of the patients and will succeed only if they are in line with the patients' expectations. True randomisation may lead to patients being allocated to treatments that they would not normally accept--an unreasonable test of an intervention.12 Although the problems may be mitigated by partial randomisation, this reduces the study population and jeopardises generalisability of the findings.
Randomised controlled trials are not impossible to perform in general practice. Published reports of failed trials in this setting are rare but those which do exist highlight the difficulties discussed.7 13 Those most likely to succeed, however, are ones that minimise disruption to the normal working environment and compensate general practitioners for the additional time commitment. Consecutive eligible patients should be recruited or at least recorded, and no patient should be knowingly disadvantaged by participating. When interpreting the results of published randomised controlled trials, researchers and clinicians need to be alert to the possibility of biased recruitment or incomplete randomisation.
There is no doubt that experimental methods provide a rigorous, sound basis for evaluating treatments, but their introduction may either disrupt the culture of primary care to such an extent that the findings do not reflect real practice, or the methodological problems encountered may reduce the scientific reliability of the results. General practice is not a laboratory, and our patients are not experimental animals. Case-control studies, retrospective and prospective cohort studies, and descriptive studies are all acceptable methods; we should accept alternative methods when a randomised study will be too difficult or the results too biased to be of value. It would be wrong to stick blindly to a gold standard which is likely to produce the wrong findings--methodologically pure but clinically meaningless.