Of the 18,297 individuals listed in the database, 384 were identified by ProPublica as earning $100,000 or more during the time period 2009 to 2010 (Figure
). Of these 384 individuals, 373 (97%) were physicians; the remaining 11 were pharmacists, nurses, and other healthcare providers. The majority (89%) of these physicians were male and the average age was 53 years (median 53, range 36 to 75). The specialties most frequently represented in the cohort were internal medicine and psychiatry (each 31%).
Study flow diagram. See text for details. Authors refer to individuals for whom we identified a publication within the period January to March 31, 2011.
The total payments to this physician cohort during the reporting period were $52,600,624, with a mean payment of $141,020 per physician (median $126,724, range $100,047 to $303,558) (Table
). The mean total payment per physician per quarter for each company ranged from $21,305 for GlaxoSmithKline to $3596 for Johnson & Johnson. Payments were for consultation, continuing education presentation, and other services. None of the payments to this cohort of payees were categorized as research funding.
Total payments and payments per physician by specialty
For specialties with five or more physicians represented in our cohort, mean payments ranged from $125,733 for psychiatry to $174,666 for anesthesiology. Four specialties (family medicine, internal medicine, surgery, and psychiatry) had 20 or more physicians represented in our cohort and the mean total payments per physician per quarter were significantly different among these specialties (P
0.0001). In particular, the mean total payment to physicians in internal medicine was significantly higher than to physicians in family medicine (mean difference $21,507; 95% confidence interval (CI) $1,533 to $41,482) and to psychiatrists (mean difference $25,623; 95% CI $13,176 to $38,069). The mean total payment to surgeons was also significantly higher than to psychiatrists (mean difference $27,990; 95% CI $6,750 to $49,229). The mean total payment to surgeons was not significantly different from that to physicians in internal medicine ($153,722 vs. $151,356). Based on results from the linear regression, the total payment per physician per quarter was not associated with sex (P
0.1730) or age (P
147 of the 373 physicians published a total of 1223 articles between January 2009 and the end of March, 2011 (mean 8 publications per physician, median 4, range 1–125). Of the 134 publications by physicians in the first quarter of 2011, 103 (77%) provided author disclosures of COI in the publication while the remaining 23% did not. Among publications with disclosures, 42 (41%) reported that the index author had nothing to disclose, while in 29 publications (28%) the author disclosed a conflict other than the payment in the Dollars for Docs database. In the remaining 32 publications (31%), the payment in the database was disclosed. The percentage of publications with the payer in the Dollars for Docs database disclosed, stratified by the 7 companies, is displayed in Figure
Figure 2 The relationship between disclosures in publications and known payments by pharmaceutical companies to physician authors. Abbreviations: AZ, AstraZeneca; GSK, GlaxoSmithKline; J&J, Johnson & Johnson. Both the grey and black bars represent (more ...)
In summary, among a cohort of physicians who were known recipients of large total payments from a select group of 7 pharmaceutical companies, nearly one-quarter of their concurrent publications do not report any COI disclosures. Furthermore, of the publications with disclosure statements, over two-thirds do not report payments listed in the Dollars for Docs database.
Several prior studies have examined the accuracy of COI disclosures in journal publications, with similar results. Two recent studies compared disclosures in publications to payment information from an orthopedic device company. Chimonas and colleagues
] reported high rates of nondisclosures: 50% among publications directly or indirectly related to payments. Okike and coauthors
] reported rates of nondisclosure in published abstracts from an orthopedic meeting of 20.7% for payments that were directly related, and 50% for payments that were indirectly related to the topic of the abstract. Other studies comparing abstracts and subsequent publications or concurrent publications also suggest a high prevalence of inaccurate author disclosures
Our study has important strengths. Our data originated with payer databases, which were placed in the public domain either voluntarily or by mandate of legal settlements. The use of physician payment data from pharmaceutical companies is novel, to our knowledge, and our study provides insights into the methods, feasibility, and limitations of using such datasets for studies of discrepancies between payments known to have occurred and disclosure in peer-reviewed journals. When datasets are more complete and journal policies more uniform, these approaches can be used to determine the accuracy and completeness of disclosures.
Our study findings have important implications. Journal editors and readers rely on accurate disclosures by authors in order to assess the credibility and risk of bias in journal publications, and our findings suggest that the accuracy and completeness of disclosures cannot be assumed at present. As industry payment data increasingly become publically available, however, journal staff and readers will be able to verify the accuracy of author disclosures. Though in the US, until the Sunshine Act is implemented and there is a unified public reporting system for industry payments to physicians, it may not be feasible for journal editors and staff to check author disclosures. For authors, the increasing availability of payment data may promote more accurate disclosures or may discourage physicians from accepting industry payments, although there is as yet no evidence to support these possibilities. It is hoped that the standardized reporting form of the International Committee of Medical Journal Editors
] will improve the accuracy of disclosures, however such data are not yet available.
There are a number of limitations to this pilot work. First, these are exploratory analyses on a small cohort of American physicians, whose characteristics and behaviors are likely not representative of US physicians in general, including those who receive much smaller total payments or those who do not receive funds from industry. Second, the database that we examined was limited to the payments of only 7 pharmaceutical companies during variable intervals in 2009 and 2010. The exact dates for payments were not available, and the categorization of types of payments varied across companies, making comparisons of the reasons for payments difficult.
Third, we faced challenges identifying publications by our physician cohort: exact author names vary (e.g., use of middle initial or not) and can resemble names of non-cohort authors. If we failed to identify all of the author’s publications, we may bias our assessment of rates of discrepancies if missing publications have a different disclosure rate than the rate in the publications that we identified. There is a delay in indexing publications in Medline thus more recent publications may have been missed.
Fourth, our assessment of rates of discrepancies cannot be equated with inaccurate disclosures because of several assumptions that we made. We did not examine journal COI policies, but rather assumed that all journals required disclosures. We also did not examine how journal COI policies address relevance of the topic of the publication to the disclosure. We assumed that all payments in the database should have been disclosed in the identified publications.
It was difficult to determine the relationship between the time of disclosure and payments as it was not possible to know exactly when authors completed disclosures relative to the time of publication, as this varies widely across journals. Although some journals request an update of disclosures at the time of publication, it is unknown how frequently authors comply and how accurate those updates are. In addition, we only had information on the quarter or group of quarters when payments occurred, and not the exact dates. Although unlikely, we could have overestimated rates of discrepancies if disclosure occurred well prior to publication and prior to drug company payment, or if a disclosure was close to the publication date and the required time period of disclosure was short, with payment preceding the disclosure period. We feel that our approach of examining 2011 publications should minimize these biases, as disclosures in publications likely encompass payments in the prior 2 years. Chimonas and colleagues
] used a similar approach when they compared payments to physicians by orthopedic device companies to disclosures in orthopedic publications 1 to 2 years after the payments were made.
Finally, we were unable to determine the reasons for discrepancies, which are undoubtedly multifactorial, including forgetting, carelessness, misreading or misinterpreting COI policies, lack of clarity in disclosure forms, or purposeful nonreporting. From these retrospective, observational data we clearly cannot determine causality or impute motive.
Future research can build on this work. Studies could further describe the characteristics of physicians and other healthcare professionals who accept payments across a broad range of values and payment types. The effectiveness of specific COI policies and disclosure forms for achieving accurate and complete disclosures should be explored for journals, institutional review boards, academic institutions, and other organizations. Predictors of the accuracy of COI disclosures can be examined when more complete payment data are available with implementation of the Sunshine Act
]. The effect of public disclosure of industry payments may be linked to provider behavior such as the future acceptance of industry payments or the prescription of specific medications when individual prescribing data are available.
Little is known about the effect of disclosure policies and practices on the intended outcomes of improving transparency and aiding in the interpretation of journal articles. The dominant opinions in the literature are that disclosures are essential because they enable the “readers to form their own opinions on whether a conflict exists and what relevance it has for the study”
]. However the evidence base for this opinion is scant. Readers may find an article less interesting and valid
] and a trial less believable
] if relevant industry relationships are disclosed. On the other hand, when presented with a direct comparison of an abstract with and without an industry conflict disclosed, physician readers did not differ significantly in their reported likelihood of prescribing the new drug
]. Further work is thus needed to determine how disclosures are interpreted and the information used by the reader, and what constitutes optimal presentation, level of detail, and relevance.