Requirements and Definitions in Conflict of Interest Policies of Medical Journals FREE

The failure of some authors to disclose competing interests that conflict with patient care has shaken the confidence of both the general public and health science professionals in peer-reviewed medical literature.^1- 3 In an effort to protect the integrity of research and improve the public trust, the Institute of Medicine, International Committee of Medical Journal Editors (ICMJE), World Association of Medical Editors, and Committee on Publication Ethics have published ethics guidelines that include specific recommendations for disclosure of information about authors’ conflicts of interest (COIs) as a means for improving transparency.^4- 7 Similarly, journal policies calling for authors to disclose COIs have evolved.^8- 12 World Association of Medical Editors guidelines suggest that journals' COI policies should be “readily accessible to everyone involved in the publication process by publishing them with instructions for authors.”⁵

A recent study of financial COI disclosures in articles reporting on coronary stents found that author disclosure statements varied from article to article.¹³ Variation between journals in handling information about COI may contribute to these inconsistencies. Recent studies examining the content of COI policies among journals with a clinical focus are limited.^14- 15 We therefore examined the COI policies of medical journals to determine the prevalence and variability of COI disclosure requirements and definitions.

Our study included the top 10% of 2117 medical journals by impact factor in the Institute for Scientific Information’s Web of Science for 35 subject categories. We conducted a cross-sectional survey of Instructions for Authors and manuscript submission forms for journals from March through October 2008. All data examined are publicly available on journal Web sites. We chose a sample size including 10% of medical journals to provide adequate representation of journal policies and a sample size that could be analyzed within a discrete period, because journals may update policies at any time. We selected the highest-impact journals, because the disproportionate influence these journals have on the medical literature imparts particular relevance on their disclosure policies. We used the 2006 Impact Factor published by the Institute for Scientific Information Journal Citation Reports, which was current at the outset of data collection in March 2008, because publication occurs 18 months after the reference year (the indexing agency cannot calculate the impact factor until all of the preceding year's publications have been received).¹⁶ We selected Journal Citation Reports subject categories likely to contain clinical journals, including those subject categories that corresponded to specialties and subspecialties accredited by the Accreditation Council for Graduate Medical Education.

To ensure sampling from different clinical areas, we included at least the top 6 rated journals for subject categories with fewer than 60 journals listed (eTable). If a journal title overlapped between categories, it was assigned to the category in which it ranked highest. Instructions for Authors and manuscript submission documents were acquired and archived using the WebCite scholarly Web-reference archiving system (http://www.webcitation.org).¹⁷ We included any guidelines or instructions pertaining to authorship or manuscript submission (eg, “Manuscript Submission Guidelines”), along with all available documents related to manuscript submission (eg, “Authorship Form,” “Copyright Assignment,” “Disclosure Statement,” or “Submission Agreement”).

Documents for each journal were reviewed independently by 1 of 2 trained abstractors (J.A.B., K.F.), using a standard form (eAppendix). To identify all text related to COI, we searched each document electronically for the phrases “conflict of interest,” “disclosure,” “financial,” “financial relationship,” “support,” “fund,” “acknowledgement,” “competing interest,” or “dual interest.” Because coding was objective (the key words appear or do not appear), we decided that concordant review of a small number of samples would be adequate for establishing interrater reliability of data abstraction. We randomly selected 5% of journals for both abstractors to review independently and calculated interrater reliability as the simple agreement for each objective data point between 2 reviewers.

We used a descriptive approach for data analysis and present the prevalence of COI policies and the content of these policies as frequencies and proportions. Comparisons between a journal's COI policies and the journal's subject category, impact factor ranking, or endorsement of the ICMJE's conflict of interest policy¹⁸ are presented without statistical testing. We used Excel 2003 (Microsoft, Redmond, Washington) and Stata version 10 (StataCorp, College Station, Texas) for analyses.

We identified 2117 medical journals across 35 subject categories (Table 1). The median number of journals per subject category was 49 (interquartile range, 30-85), with a range of 8 to 147 journals. Our cross-sectional sample included 262 journals from these 35 subject categories (eTable). Of 262 journals, 256 (98%) provided a URL linking to Instructions for Authors, manuscript submission forms, or both. Six journals (2%) solicited manuscripts by invitation; because Instructions for Authors were not available, these journals were excluded from analysis. Within the Instructions for Authors, 25% of journals included online links to policy statements or articles on COI.

In addition, 14% of journals were part of a journal group, with shared Internet domains and similar formats for their online instructions, based on the following affiliations or owners: Nature Publishing Group (11 journals), American Medical Association (9 journals), British Medical Journal Group (6 journals), American Heart Association (5 journals), The Lancet (4 journals), and Annual Reviews (4 journals). We noted that some journals within a group had distinct definitions and disclosure requirements despite having similar wording. In other journal groups, the instructions, manuscript submission forms, or both were identical. For the purposes of this study, each journal was reviewed independently and received equal consideration in our analysis. Interrater reliability of data abstraction was 97% concordance of all objective data points recorded for each journal.

We found that 228 of 256 sampled journals (89%) had language requesting author COI disclosure (Table 1). Eighty-five percent of journals requested COI disclosure in the Instructions for Authors; an additional 4% requested COI disclosure only in manuscript submission documents, for a total of 89%. Of 256 journals, 138 (54%) required authors to submit signed disclosure statements.

We then examined specific details, examples, and definitions of COI provided by each journal (Table 2). Of 256 journals, 197 (77%) defined or gave examples of possible COIs for submitting authors. Of 197 journal COI definitions, most included direct financial relationships such as “equities interest or stock ownership” (89%) or “consultancies” (84%). A minority included other potential conflicts such as personal relationships (42%), paid expert testimony (42%), relationships with organizations (26%), or travel grants (12%).

To describe the variations in COI policies that we observed, we divided journals by subject categories. The prevalence of policies varied by subject category. For example, all internal medicine, respiratory medicine, and toxicology journals studied had comprehensive COI definitions, with 19 of these 24 journals requiring signed disclosure attestations. In contrast, only 6 of the 19 highest-impact journals in geriatrics and gerontology; radiology, nuclear medicine, and medical imaging; and rehabilitation requested that submitting authors discolose COIs.

Despite choosing the top 10% of journals by impact factor, there was a wide range of impact factors within our sample (0.9-63), so we explored differences in prevalence of COI policies as a function of impact factor. Journals in our sample set were divided into top and bottom quartiles based on impact factor; the prevalence of COI policies in the top quartile was 94% compared with 89% in the bottom quartile (Table 3). We noted that several journals with an impact factor greater than 10, including Immunology Reviews, Molecular Psychiatry, and Seminars in Immunology, did not require submitting authors to report COIs or submit signed disclosure statements.

Lastly, we examined the COI policies of the journals based on the journal's endorsement of ICMJE guidelines. At the time of this study, 69 journals in our sample had officially endorsed the ICMJE guidelines, and 187 journals had not. Nearly all journals that officially endorsed ICMJE guidelines had COI policies (68/69), compared with 84% (158/187) of those journals that did not endorse ICMJE guidelines (Table 3). We noted that some journals, such as Cephalalgia and Journal of Nuclear Medicine, appear on the ICMJE list of journals that follow the ICMJE Uniform Requirements for Manuscripts Submitted to Biomedical Journals,¹⁸ but they did not ask authors to disclose COIs in their Instructions for Authors or manuscript submission forms.

In the interest of transparency, and under intense public scrutiny, the ICMJE and editors of leading medical journals have implemented increasingly comprehensive policies regarding disclosure of authors' financial interests and other potential conflicts.^10- 12,18 Conflict of interest policies and their implementation have evolved over time. For example, JAMA initially encouraged authors to disclose conflicts, but authors were not required to sign disclosure statements until 1998. In 2006, JAMA began to require that authors not only sign a disclosure statement but also publicly disclose all potential COIs in the Acknowledgment section of submitted manuscripts.¹⁰

Our study provides new data on COI policy availability, content, and requirements for signed disclosure statements among high-impact medical journals from March 2008 through October 2008. In 1997, Krimsky and Rothenberg found that only 16% of 1396 journals across all scientific disciplines had COI policies.¹⁹ By comparison, in our sample, in 2008, 89% of medical journals had COI policies available for public review. While the journals sampled and methods were different, this suggests a substantial increase in the prevalence of COI policies over the past decade.

Even when editorial COI policies exist, the policies may not be available to submitting authors. In 2004, Ancker and Flanagin illustrated this weakness of COI policy in a survey of 84 journals across 12 scientific disciplines, finding that only 33% of journals had online published COI policies, although 80% reported having such a policy when surveyed.¹⁴ In that study, all general medical journals had published COI policies for authors, while none of the journals in biology, engineering, physics, and zoology had published policies. Another survey (published in 2006) of the editors of 91 medical journals found that 93% of medical journal editors reported having a COI policy.¹⁵

Unlike these studies, our prevalence estimate was based on having the policy available to submitting authors, not only in the Instructions for Authors but also in the manuscript submission agreement or other available documentation. We did not survey the journal editors, and our study's reliance on review of journal policies available on the Internet might be considered a limitation. This methodology could lead to underestimation of the prevalence of COI policies if COIs were only requested by editors in communication with authors after submission. However, in the interest of transparency, the COI policy should be available for review by submitting authors and also by the public readers of the journal.⁵

Our finding that most journals' COI definitions were limited to direct financial interests is not surprising. Some of the first COI disclosures in the medical literature were financial disclosures, and Ancker and Flanagin found that the author COI policies of most journals addressed financial interests.¹⁴ Conflict of interest disclosure recommendations have expanded with the recognition that conflicting interests arising from professional, intellectual, or academic competition can also lead to bias.^5,18,20 The infrequent appearance in our sample of definitions including more subtle conflicts illustrates a lack of uniformity among journal editors on what constitutes COI. The variability that authors encounter as they submit to different journals may limit the intended goal of COI policies.

We found that only 54% of journals evaluated required that each author sign a COI statement. This discrepancy between journals having COI policies (89%) and requiring signed statements (54%) is important, because the corresponding author may be the only author of the manuscript to review a COI policy provided only in the Instructions for Authors. The lack of required signed attestation for each author of a specific manuscript may limit dissemination of the policy to authors of a manuscript beyond the corresponding author, potentially leading to failures of some authors to disclose.

We explored differences in journals based on subject categories, impact factor, and the following of ICMJE guidelines. Interpretation of differences between subject categories was limited by the small sample size for many of the journals (n = 6 for many categories) and our method of sampling higher-impact journals. When compared by impact factor, the prevalence of disclosure requirements between top and bottom quartiles of journals was similar (94% and 89%, respectively). A 2006 survey of editors from 84 scientific journals (7 medical journals) found that the highest-impact journals were most likely to have a published COI policy.¹⁴ Our comparison by impact factor quartiles is again limited by our method of sampling the higher-impact journals from each category. We also compared journals based on the ICMJE list of publications that follow the ICMJE Uniform Requirements for Manuscripts Submitted to Biomedical Journals.¹⁸ The ICMJE requirements specify that journals should have COI policies in place with specific definitions. We found that overall, journals that follow the ICMJE guidelines more frequently request author COI disclosure and provide definitions than those journals not following the guidelines. However, we also found that some journals listed did not actually appear to follow the ICMJE's Uniform Requirements with regard to COI disclosure practice.

In an attempt to make the process of disclosure easier for authors and less confusing for readers, ICMJE member journals have recently published a uniform vehicle for recording and submitting COI disclosures.²¹ This uniform disclosure form has standard definitions and disclosure requirements and enables authors to save online copies of the forms that can be updated as needed. This eliminates the need to reformat disclosure information for each submission to ICMJE member journals and other journals that would eventually accept this universal disclosure form. The ICMJE member journals are testing this form and the universal disclosure process and are asking for feedback until April 2010.

In summary, we found that most high-impact medical journals had author COI policies available for public review in 2008. These findings are encouraging. However, many journals did not require authors to sign disclosure statements, and there was variability in how COI was defined. With little uniformity among journal requirements, authors may encounter differing, and at times confusing, requests for disclosure or no requests for disclosure. We only assessed the means by which the journals defined and solicited disclosures and did not determine their publication practices. Readers should consider the potential for undisclosed COIs in medical journals that lack explicit disclosure requirements. Future investigations should determine whether all disclosures of authors are published by journals. Furthermore, it is important to assess whether detailed COI policies and mandating signed disclosure statements from all authors increases accurate reporting of author COIs.