In a trial in which an individual or group is allocated to receive or not to receive an intervention, 'contamination' is essentially defined as the influence of an intervention on a subject not intended to receive the intervention. For example, suppose patients within a GP practice are randomized to receive or not to receive dietary advice aimed at reducing their risk of bowel cancer. If individual patients allocated no intervention become aware that others are being treated differently, then they may seek out and act on the advice which the other group is intended to receive. If this is a major problem, then the power of the trial to detect a difference between groups is clearly reduced. A certain amount of contamination is a potential problem with many individually randomized trials, but where the danger is acute, a cluster randomized design may, if it substantially reduces the chance for cross-group contamination, be a useful alternative. Suppose, for example, one wishes to assess the impact of an education package delivered through a GP on patients with a certain condition. If individual patients were randomized to receive or not receive the information, the GP would have to 'unlearn' for some patients, and this may be very difficult. A similar difficulty applies when attempting to assess the impact of organizational changes. Moher et al.  report a cluster randomized trial comparing three methods of promoting secondary prevention of coronary heart disease (CHD) in primary care. The three interventions were (a) audit and feedback at a practice meeting on the prevalence of CHD in the practice, number of patients with myocardial infarction, angina and revascularization and the proportion of patients with adequate assessment (b) information as for (a) but with discussion on guidelines for secondary prevention, and advice on setting up a register and recall system for GP-led review of patients with CHD and (c) information and advice as for (b) but aimed at nurse-led clinics. Thus the unit of randomization was the GP practice, the experimental units were the practice staff and the observation unit was the individual patient, with the primary outcome measure being evidence of'adequate assessment' of the relevant patients at 18 months.
Was this article helpful?