In order to reduce subjective interpretation of what constitutes a guideline publication we developed strict eligibility criteria. Eligible for the study were all articles that (I) contained in their titles, heading or abstracts (or introductory / summary paragraph when an abstract was missing) key words that were characteristic of guidelines ("guidelines" or "recommendations", "consensus [panel-statement-conference]", "clinical synthesis conference", "guidance", "policy statement", "practice parameter" and "position [paper-article-statement]"; and (II) had main focus on preventive and /or therapeutic interventions including health care delivery. We excluded papers where the main focus was on descriptive epidemiology, research design, diagnosis (diagnostic performance rather than clinical impact of diagnostic methods), risk assessment, legal issues, biology and/or pathophysiology. We specifically excluded editorials, commentaries, original randomized controlled trials, as well as systematic reviews and meta-analyses unless they were part of the guideline. Both full-length and shortened versions qualified, but comments and discussion items concerning guidelines were excluded. The last search for eligible guidelines was performed in mid-October 1999.

Data extraction, including journal, year, main subject focus, main country of origin, authorship, type(s) of interventions and funding, was performed on standardized forms. We also recorded whether conflicts of interest were mentioned at all, and if so, how much space was given for such disclosures. The amount of space given for conflict disclosure does not necessarily guarantee the completeness of disclosure or the quality of the conveyed information, but it can be used as a surrogate of the importance given to this information. This parameter has been previously used for other aspects of reporting in journal publications [2]. Finally, we recorded the nature of specific disclosed conflicts. We considered all reported conflicts of interest, financial and non-financial, that were judged to be relevant for publication by each journal. For each selected journal we examined the instructions to the authors over the last 20 years to examine if and when specific disclosure policies were available for conflicts of interest.

Only 7 guidelines (3.7%) disclosed potential conflicts of interest (table ​(table1).1). The disclosure rate was 6.1% when estimated on the basis of the 115 guidelines published when specific disclosure policies were in place. All disclosures had been made in 1999 and for this year the disclosure percentage was 17.5% (7/40). Reporting was also related to specific journals (p=0.026), but the rate of disclosure was very low even in the journal with the highest disclosure rate (JAMA, 4/35 [11.4%]). The country of origin, focus (therapy, prevention or both), group authorship and emphasis on medication vs. other interventions were not related to reporting of conflicts of interest (p=1.00, 0.14, 1.00 and 1.00, respectively) suggesting the deficiency is universal. Guidelines funded mostly by the government, universities or major professional organizations almost never reported on potential conflicts of interest while guidelines funded by private or mixed sources were more likely to disclose potential conflicts (p < 0.001). Among guidelines where specific authors assumed primary responsibility for writing the manuscript, reporting of conflicts of interest was more common when there were more authors (p=0.004) (table ​(table11).

Further analysis revealed that all guidelines reporting conflicts of interest dedicated less than 1/20 of a page for this reason and less than 1/100 of the total length of the article - with one exception where disclosure took 13 lines to report. In the entire database of 191 guidelines, a mere 18 authors disclosed a total of 24 potential conflicts of interest (table ​(table2).2). The majority of the alluded items of conflict were for relatively minor issues (such as travel fees, speaker funds and consultation) while no author disclosed possessing stock in a company.

It is estimated that the pharmaceutical industry in the United States spends $8000 to $13000 per year on each physician simply for marketing purposes [6,7] and this phenomenon is probably international [8]. It is unknown how these figures should be extrapolated to field experts who are authors of clinical guidelines. More studies are needed to examine the extent of financial and other connections of experts to the industry in their field of expertise. One study [9] found that 15% of lead authors from Massachusetts publishing in 14 major biomedical journals in 1992 had at least one major potential conflict of interest, such as serving on scientific advisory boards of biotechnology firms in their state; being officers, directors or major shareholders in biotechnology firms; or being listed as inventors in a relevant patent or patent application. In all, 34% of the sampled articles had at least one author with a financial conflict of interest directly relevant to the publication [9]. The proportion might have been larger if financial interest data could have been retrieved from additional out-of-state and international connections and if more non-financial conflicts of interest could have been captured as well. In another study of authors of 70 studies of calcium channel blockers for treating cardiovascular disorders [10], 96% of the authors of supportive studies had financial relationships with the manufacturers, as compared with 60% of neutral authors and 37% of critical authors. Only 2 of the 70 articles disclosed the authors' potential conflicts of interest. Finally, one study of faculty members at the University of California [11] found that 7.6% of investigators reported personal financial ties with sponsors of their research. Given the large heterogeneity in the policies of reporting faculty conflicts of interest [12,13], it is difficult to arrive at exact estimates of the frequency of serious competing interests in field experts.

In this study, it was not possible to determine exactly what might have been undisclosed, since the authors were sampled from all over the world and guidelines covered very diverse topics. Sorting out financial connections in a consistent fashion would have been practically impossible. Non-financial conflicts of interest would be even more intangible to the outsider. Nevertheless, one may speculate that competing interests may have been at least as common as what has been described in the other studies mentioned above. For example, the 1999 BMJ guideline about the management of hypertension [14] did not declare any competing interests. In a MEDLINE search, we were able to identify at least 66 randomized controlled trials of anti-hypertensive interventions authored by the guideline authors up to 1999 (range 1 to 34 per author). These trials evaluated a total of 27 different anti-hypertensive drugs (range 0 to 14 per author) in various doses and formulations and 9 other non-pharmaceutical interventions. In any case, funding of trials by the pharmaceutical industry does not mean that the investigators would be biased and it is expected that a panel of leading scientists would be able to balance on objective conclusions during the development of consensus. Still, it would have been useful to know what might have been the potential competing interests, including non-financial ones [15], of the panel members.

We found that conflicts of interest were rarely, if ever, published in clinical guideline reports funded by the government, universities or professional physician organizations. Although it may be anticipated that such sponsors may make more stringent efforts to assure objectivity in guideline development, it is still important that potential conflicts of interest of the individuals participating in the process should be acknowledged. This holds true even if group authorship is assumed. Conflicts of interest may be more worrisome when there is individual authorship. One would expect the objectivity of the guideline to be more vulnerable when only one or two individuals are authoring it. Nevertheless, the likelihood of disclosing conflicts of interest was smaller when few authors took responsibility for a guideline than when several authors were involved.