Membership of the group was the first point of debate — and dissent. It was proposed that the group would comprise the rheumatology members of the two pharmaceutical advisory boards (Pharmacia/Pfizer and Merck, Sharp & Dohm), representatives from the medical and economic departments of the pharmaceutical companies, general practitioners, gastroenterologists, nephrologists and cardiologists, rheumatologists with expertise in clinical epidemiology and representatives of the Arthritis Foundation of Australia (AFA), the Australian Rheumatology Association (ARA), and the National Prescribing Service (NPS). This arbitrary membership initially raised no concerns. Later, some participants noted that opinions expressed in some sections of the article may have been different had there been broader representation in the non-rheumatological specialist areas. Others questioned whether a consensus process, claiming to be evidence-based, should involve people with potentially conflicting commercial interests (such as advisory board members), although, in such an event, just who would have undertaken the task is not clear. Not surprisingly, the main point of contention from the outset was the inclusion of the medical and economic representatives of the pharmaceutical companies. Indeed, some people declined the invitation to join the group, as they considered such heterogeneity would prejudice the perception, if not the conduct, of the process. The group as a whole took a contrary view. As some form of bias is the inevitable consequence of knowledge and involvement in any issue, it seemed best to include a broad range of stakeholders, but to strive for transparency by listing the group's composition and affiliations.

Most members of the group attended the first meeting, which agreed on the task and the method. Two documents were to be produced. First, a detailed analysis and interpretation of the relevant studies as a referenced source document for a set of recommendations or "considerations" (this will be published elsewhere); and second, a succinct and practical summary of these "considerations" (see page 328).1 The group decided that it would be more appropriate to offer "considerations" rather than "guidelines" for CSIs, as the latter could be seen as having some binding force.

Having defined the aim as the development of considerations for the safe prescribing and use of the CSIs currently available in Australia, certain working agreements were adopted. First, that there were four domains of interest — gastrointestinal, renal/hypertension, cardiovascular and important comorbidities, and coprescribed medications. Second, that data on each of these areas would doubtless continue to emerge, but, if our "considerations" were not to be indefinitely delayed, an arbitrary cut-off to input would be needed. Third, that the cut-off for data from papers in peer-reviewed journals (but not abstracts) should be the end of May 2001, and that we would also use as well the pertinent sections of the United States Food and Drug Administration (FDA) website, accessible in the public domain. Data sources are described in the Box.

The point of documenting these details relates to the basis for disagreement among stakeholders. There is an assumption that medical controversy can largely be solved by evidence, given that evidence is of appropriately high quality. The problem that remains, however, is the interpretation of evidence and the extent to which that is influenced by bias, whether subtle or bold. The CSI working group members were using evidence that, on face value, was of high quality (Level 2 or better), based on results from large, double-blind RCTs reported in prestigious journals. However, close examination of trial design, exclusions and inclusions, the fact that the CLASS report provides a six-month interim analysis, the relative weight that should be assigned to the FDA analysis of the major studies, the degree to which an endoscopic endpoint infers major clinical outcomes, the validity of pooled analysis results, and many other factors, combined to pose different problems for participants with different points of view.

After the struggle with data interpretation came conflicts over the expression of conclusions. Should an idea be conveyed with a descriptive adjective or by a number, a mean or a median, a range or a confidence interval? Should each statement be qualified to reflect the spectrum of opinion among participants or would that simply obscure the message?

After some weeks of vigorous viewpoint interchange and multiple redrafts, the working group felt the need to terminate further debate, finalise the document and offer participants four options: (i) to sign off on the final draft; (ii) to decline to sign off and remain anonymous; (iii) to decline to sign off and add their names without expressing a reason; and (iv) to decline to sign off and add both their names and reasons. In the event, all participants chose option (i) or (iv).

Twenty-three of the 31 participants agreed that the document was a reasonable statement. Those who declined authorship of the final document — the four representatives of one pharmaceutical group (Pfizer/Pharmacia), two rheumatologists with epidemiological expertise and two other rheumatologist members of advisory boards — represent a diverse range of views, listed on page 331.1

The important lesson of this exercise is that evidence, even of high quality, requires interpretation, and interpretation then requires a form of expression. Within any large group, a proportion of individuals, for reasons of their biases, whether owing to knowledge, expertise, affiliations or commercial interests, and despite genuine willingness to make reasonable concessions and seriously consider alternative points of view, will be unable to assent to statements which others find well founded on the evidence before them. Clearly, the size and composition of the group working towards consensus is a critical element: the more alike in view, the easier and more complete the agreement, but the less likely to represent the prevalent range of views. And where there is a real polarisation of views, this should not be concealed under cover of a facile consensus. In the current era of "evidence based medicine" we need to be aware of the vagaries of the interpretive part of the exercise and, where possible, bring the inevitable dissent to the fore so that those struggling with an abundance of evidence-based guidelines and summary statements are alert to their faults and fallacies.