Anyone who has tried to appraise a randomised controlled trial critically will be aware of the frustration that arises when a key piece of information is missing. To understand the results of a randomised controlled trial a reader must understand its design, conduct, analysis, and interpretation. That goal can be achieved only through complete transparency from authors. The original and revised CONSORT (consolidated standards of reporting trials) statements were designed to help authors improve reporting by using a checklist and flow diagram and have been well cited.1 These have now been extended to include cluster trials (p 702).2 Cluster trials randomise interventions to groups of patients rather than to individual patients and have their own problems. Using the extended CONSORT statement should help reduce bias and help readers to understand a cluster trial's conduct and to assess the validity of its results.
A website provided by the Medical Research Council gives guidelines for the design and analysis of cluster trials.3 These trials are particularly useful in general practice where the cluster is the general practitioner or the practice.4 For example, in the Diabetes Care from Diagnosis Trial, general practitioners were randomised to be trained in patient centred care or not.5 All patients under the care of one general practitioner receive the same treatment and so cannot be considered to be independent items. One of the main reasons for conducting cluster trials is fear of contamination, whereby patients used as controls are exposed to the intervention. For example, it would be difficult for a general practitioner to switch from a patient centred approach to a more paternalistic approach between successive patients. Patients in one practice may discuss what their general practitioner has given them, and patients used as controls may demand the same treatment as those given the intervention.
The main problem associated with their design, conduct, analysis, and interpretation, compared with individually randomised trials, is that two different units of measurement—the cluster and the patient—are used. Each needs to be reported carefully. The key statistic is the intracluster correlation coefficient, which is the ratio of the between cluster variation of the outcome variable to the total variation. The startling fact is that even with apparently low values of the intracluster correlation coefficient, such as 0.05 (which is commonly found in general practice trials), when there are reasonable numbers of patients in each cluster (say 20), then the usual methods of analysis, which fail to take into account clustering, can seriously underestimate the standard error of treatment effects and so provide spuriously narrow confidence intervals. Compared with individually randomised trials cluster trials therefore are inefficient in terms of power for a given effect size and sample size. Other problems are that randomisation has to occur at the start of the trial, and blinding these trials is more difficult, thus increasing the potential for recruitment biases. Cluster leaders have to consent to the trial on behalf of the potential cluster members, which raises ethical issues. Several surveys have highlighted problems in all these areas in the past, although there is evidence that more recent trials are better reported, perhaps because of recent efforts by medical statisticians to make the research community aware of the difficulties of cluster randomised trials.6,7
The extension to cluster trials is timely since the number of trials reporting a cluster design has risen exponentially since 1997. That the revised statement should appear in the BMJ is fitting, since a recent review of cluster trials published since 1997 in the Lancet, New England Journal of Medicine, and the BMJ, showed that 24 of the 36 trials found had appeared in the BMJ.7
The checklist items relate to the content of the title, abstract, introduction, methods, results, and discussion. Similar to the statement for individually randomised trials the checklist includes 22 items, chosen to reflect important aspects of cluster trials. A failure to report an item is important because it may be associated with biased estimates of treatment effect or because the information is essential to judge the reliability or relevance of the findings. The flow diagram emphasises that the important unit is the cluster, and reporting of how both the cluster as well as the individuals progress through the trial is important. On a more pragmatic level, hopefully, investigators reading the checklist will be guided as to the correct way to calculate the required sample size, to randomise to minimise bias, to analyse the data at the end, and to report the intracluster correlation coefficient.