We found that 53% of guideline authors had conflicts of interest, corresponding to 43 out of 45 guidelines being written by one or more authors with conflicts, and that only 2% disclosed them, in just one guideline. Most guidelines did not state how they were developed or graded the evidence, and many did not include references in the text.
Our study was based on very comprehensive data from the Danish Health and Medicines Authority, and physicians and drug companies are required by law to report their collaboration to the authority. Our findings demonstrate that reliance on voluntary disclosure underestimates the prevalence of conflicts of interest substantially.
Using a publicly available disclosure list made it possible to identify undisclosed conflicts of interest, but one limitation of the list is that drug company affiliations are deleted as soon as the collaboration ends. As judged by our sensitivity analysis, this seemed to have had little impact on our results, but we suggest that such lists include affiliations up to 3 years prior to the current date, similar to the ICMJE criteria.
It is therefore likely that previous studies of conflicts of interest among clinical guideline authors have underestimated the prevalence of conflicts of interest, as they have relied on voluntary disclosure. We note, however, that the underreporting of conflicts of interest in our study of Danish guidelines may have been atypical, e.g. US guidelines have more disclosures. We found that only six societies had instruction manuals for guideline preparation and none included policies on conflicts of interest. Thus, lack of policies on conflict of interest disclosure may have caused the extraordinary underreporting of conflicts of interest found in our study. On the other hand, this lack of transparency cannot be excused by the lack of explicit policies, as it is well known that it is important to declare conflicts of interest in biomedical publications, especially in clinical guidelines.
Albeit our study was based on Danish guidelines, our sample represents 14 different specialties and we have thereby obtained more comprehensive information on the extent of conflicts of interest among guideline authors than in other studies, which have usually only included guidelines from a few specialties.
In a recent systematic review by Norris and colleagues, the prevalence of conflicts of interest ranged widely among the different studies included, from 18% to 100%. The large variation in prevalence may have several explanations.
Firstly, the sources used for identifying conflicts of interest may have been important. Many studies identified conflicts of interest solely based on authors’ disclosures in guidelines, which will generally underestimate the prevalence[7, 21]. Other studies have identified conflicts of interest based on authors' disclosure in their additional journal publications[11, 22, 23] or by surveys[24, 25]. However, such strategies are often inadequate, e.g. many journals do not include disclosure statements in their articles or have only started recently, response rates in surveys were often low, and authors often do not to disclose conflicts in their scientific publications[26, 27]. A few studies have used other sources such as US patent databases[22, 28].
Secondly, what constitutes a conflict of interest may also influence its prevalence. We coded a conflict of interest to be present if authors had an affiliation with a company up to 3 years prior to the published guideline. This interval differs among studies as do perceptions about what constitutes a conflict of interest. For example, some studies included paid travel fees[9, 21], which we did not, as such conflicts are not listed on the disclosure list. We coded any financial tie as a conflict of interest, although some ties might be related to companies producing drugs not relevant to the guideline. However, this will likely be less important, as guideline authors are usually affiliated with companies producing drugs in areas where they are experts and write guidelines in the same areas.
Thirdly, the overall extent of conflicts of interest among physicians varies between countries. In Denmark, the prevalence is approximately 8.5%, whereas it is 14.1% among US physicians.
Public disclosure lists may assist guideline issuing bodies in ensuring that all conflicts of interest are disclosed. In order for guideline end-users to judge fully the possible influence of the conflicts of interest, information on individual authors' income from drug companies is also relevant. This information will be available with the implementation of the US Sunshine act in 2014, which will provide a public and comprehensive register of data about physicians’ financial relationships with the drug and device industries[15, 30]. Similarly, the Danish disclosure list will in future include financial information. Apart from the improved transparency in relation to guideline users, such registers may also assist researchers studying conflicts of interest.
While disclosure improves transparency, it does not remove the potential bias related to conflicts of interest[5, 6, 32], and a better strategy is to prohibit authors with conflicts from guideline production or minimise their influence on formulating recommendations. It has been argued that authors without conflicts of interest lack the content area knowledge and skills necessary to interpret the scientific data. But this is a flawed argument. Industry relations, such as being on a company’s speakers bureau or advisory board, does not serve any academic purpose and authors can choose to avoid these relationships. Furthermore, there is a substantial pool of authors without conflicts. Lastly, content area experts often have preconceptions about treatment effects, which may bias their interpretation of the evidence[34–36]. A better strategy would be to include more methodologists as guideline authors as they are often free from preconceptions and rarely have conflicts of interest.