Evidence about medical writing practice
Medical writing practice has been measured as well as argued about: how often writers' substantial contributions go undisclosed, how often companies and journals acknowledge writers, what declared writing support goes with in retractions and in the reporting of trials, and how far the circulating estimates of ghostwriting hold up.
Early estimates of writer involvement
In the surveys of AMWA and EMWA members, ghostwriting is defined as undisclosed substantial contributions by medical writers to manuscripts published in medical journals, which AMWA, EMWA and other professional associations consider unethical. Not disclosing such contributions was perceived to be common, but its prevalence was unknown. Ghostwriting is presumed to be widespread, but a systematic review found that estimates of it had often rested on anecdotal evidence, on statements taken out of context and on confusion about authorship criteria, and the prevalence among medical writers themselves was unknown in the early 2000s. In the decade to 2014, ghostwriting of medical manuscripts received considerable media attention, out of concern that unacknowledged contributors may be biasing the information presented.
Early estimates of writer involvement were low, and appeared higher for articles from industry-initiated trials. A 1998 survey of corresponding authors at five journals suggested writer involvement in 14 of 809 articles (2%); a survey covering 1994 to 1995 found evidence within the articles themselves of writer involvement in the reports of 4 of 40 industry-initiated trials (10%); and a study of articles published between November 2004 and January 2005 in ten high-ranking journals found writing assistance declared in 6.0% of articles overall and in 9.8% of pharmaceutical-sponsored ones. Because assistance is not necessarily acknowledged, the true prevalence of writer involvement could be higher, and a case study of rofecoxib articles documented extensive company involvement in drafting without acknowledgment of the extensive writing assistance. Authors were divided: in the 1998 survey, asked whether they would use a professional medical writer if one were available, 38% of 812 corresponding authors said yes, while 35% of respondents said using one was unethical.
Undisclosed contributions among AMWA and EMWA members
How the surveys were run
AMWA and EMWA members were surveyed in 2005 and again in 2008, 2011 and 2014, to measure the prevalence of ghostwritten manuscripts among them before, during and after the implementation of educational initiatives. The 2005, 2008 and 2011 rounds set out to determine the proportion of substantial contributions by medical communicators that go undisclosed in submitted manuscripts, the proportion of medical communicators who request acknowledgment, and the effect of familiarity with publication guidelines on disclosure ; the report of the 2005 and 2008 rounds also set out to determine the proportion who request disclosure of their potential conflicts of interest , and the final report treated requests for acknowledgment and the variables associated with acknowledgment as its secondary objective. The 2008 round was announced for 13 to 25 November 2008, promised to be anonymous and to take less than five minutes to complete ; for the November 2014 round, the organisers hoped to attract a large number of participants to gauge the prevalence of ghostwriting and other unethical practices among members of both associations.
The survey was web-based, self-administered and confidential, and all members of both associations were invited by email; it asked about the manuscripts participants had themselves made substantial contributions to rather than about all published articles, and it did not define "substantial". In every survey year the largest single employment category among participants was freelance, more participants were members of AMWA than of EMWA, and participants otherwise represented a wide range of employment types, years of experience and manuscript volumes with no obvious change over time. The primary figure is the mean percentage of manuscripts containing undisclosed contributions in the last year, weighted by the number of manuscripts each participant had contributed substantially to in an average year. No formal statistical comparison was made between the survey years, because none was a prespecified objective when the 2005 survey was planned, so the change from wave to wave is described rather than tested.
The fall in undisclosed contributions
Undisclosed contributions fell over the first three surveys and then stopped falling. The weighted percentage of manuscripts with undisclosed contributions was 61.8% (95% CI 59.0 to 64.6) in 2005, 41.7% (38.6 to 44.7) in 2008, 33.0% (29.7 to 36.3) in 2011 and 34.4% (30.2 to 38.5) in 2014: a 33% decrease between the first two surveys, an overall decrease of 44% between the first and last, and an improvement over the first three waves that was not sustained into the fourth. The number of participants with valid data was 843 in 2005, 773 in 2008, 620 in 2011 and 410 in 2014. Unweighted, the mean percentage fell from 58.8% (95% CI 55.8 to 61.8) among 750 participants in 2005 to 26.4% (22.4 to 30.4) among 355 in 2014; the weighted figures for those two years came from 744 and 354 participants. The decreases between 2005 and 2008 and again between 2008 and 2011 coincided with international efforts to clarify publication guidelines and to increase awareness of them.
Acknowledgment, disclosure and guideline awareness
Writers' experience of and practice in requesting acknowledgment were generally consistent with the trend in undisclosed contributions. The proportion of participants requesting acknowledgment of their substantial contributions rose from 370 of 747 (49.5%) in 2005 to 456 of 665 (68.6%) in 2008 and 282 of 357 (79%) in 2014, and the proportion requesting disclosure of potential conflicts of interest from 361 of 598 (60.4%) in 2005 to 460 of 604 (76.2%) in 2008. The proportion whose requests were granted was 83% in 2005 and remained high, at 95%, in 2014, while between those years the proportion encouraging authors and other contributors to follow the ICMJE guidelines rose from 55% to 81%, and the proportion who perceived ghostwriting to have decreased over the previous five years from 39% to 64%. Reported familiarity with the ICMJE guidelines was 54% of participants in 2005 and 85% in 2014.
The high level of familiarity with the ICMJE guidelines among 2014 participants is consistent with other recent surveys of publication stakeholders. The Medical Publishing Insights and Practices Initiative surveyed nearly 500 people in four groups and found most medical writers (88%), publication professionals (97%) and journal editors (89%) aware of the ICMJE authorship criteria, against only 49% of clinical investigators; reliance on the criteria ran at 51%, 70%, 59% and 28% respectively. The Global Publication Survey, of nearly 500 stakeholders, especially employees of medical communication agencies (51%) and of pharmaceutical or device companies (30%), found 93% of both groups routinely referring to the ICMJE for guidance on ethical practice.
What goes with disclosure
In univariate analyses of every survey year, participants familiar with more guidelines had fewer undisclosed contributions, the coefficient per additional guideline running from -6.6% (95% CI -8.5 to -4.8) in 2005 through -7.7% in 2008 and 2011 to -10.6% (-13.1 to -8.0) in 2014. In the 2005 and 2008 surveys, familiarity with publication guidelines was also associated with a higher probability of requesting acknowledgment of contributions and disclosure of potential conflicts of interest. In 2008 that association was read as suggesting that education may promote good publication practices, with more research needed to confirm the findings and to evaluate the effects of educational initiatives. In stepwise multivariate analysis, in at least two of the survey years, ghostwriting was associated with contributing substantially to more than ten papers a year relative to one or two, with review-type articles relative to original research, and with freelance status relative to employment by a hospital, university or medical school; disclosure was associated with familiarity with the AMWA, EMWA, GPP, ICMJE and PhRMA guidance, of which the ICMJE association was significant in all four years, at between -14.0% and -20.5%. These associations do not prove cause and effect, but they can be used to generate hypotheses that might have practical implications: if future research confirms that certification and other variables are associated with transparency, employers, contractors and authors could use the findings to enhance their criteria for selecting writers, and the findings may inspire companies to encourage or even require their writers to take up educational opportunities and to audit freelance writers for awareness of and compliance with best practice.
Writers who contributed to more manuscripts were likelier to have undisclosed contributions, which is why the weighted proportion sits slightly above the unweighted one. Some contributions by prolific writers may not have been substantial, being limited to copy editing, say, and so less deserving of acknowledgment, and substantial contribution is undefined in most guidelines, so its interpretation can often be a grey area. Alternatively, contributing to larger numbers of manuscripts may really be correlated with ghostwriting.
How far the figures reach
The survey figures cannot be generalised to the prevalence of ghostwriting in the literature at large, because participation was restricted to AMWA and EMWA members who had made substantial contributions to manuscripts. The participation rate fell from 28% (1,537 of 5,463 invitations) in 2005 to 8% (464 of 5,664) in 2014; the fall between 2005 and 2008 was probably due to a clarified invitation that better defined the target participants, and the later falls are unexplained. Because the survey was anonymous, it is not known how many respondents in later surveys had also taken part in earlier ones, so it is not possible to tell whether the fall in ghostwriting represents individual writers changing their practice, a newer cohort of writers less likely to make undisclosed contributions, or both. Data were self-reported and based on recall; members of the two associations may be likelier than other writers to follow guidelines, and those who take time to answer a survey likelier to take time to learn the guidelines and comply with them, so the figures might underestimate the prevalence of ghostwriting. The word ghostwriting was deliberately kept out of the invitation, partly to avoid email security filters and partly because the term is often misunderstood and potentially ambiguous, and the lengthy, often awkward wording that resulted might have led to unintended answers.
The low response rate is partly attributable to the survey's restriction to a subset of the two associations' members. The proportion of members who make substantial contributions to manuscripts is unknown; taking the 26.8% (108 of 403) of AMWA members who named scientific publications as their primary area of work in a recent survey, and assuming it holds for EMWA and for the whole 2014 sample, about 1,518 of the 5,664 people invited would have been eligible, which puts participation at 28.8% (437 of 1,518) rather than the 8% calculated on the whole membership. Members of the International Society for Medical Publication Professionals were never invited: the first survey predated that society, and they were left out of later waves to keep the waves comparable, although inviting them would probably have raised the number of participants.
A literature-wide prevalence of about 3% can be estimated from the survey, with caution. A separate survey found medical writing assistance declared in 6% of publications in 1,000 high-ranking journals; assuming that writers fail to disclose a third of their contributions, so that the ratio of undisclosed to disclosed is 1:2, the combined figures suggest an overall ghostwriting prevalence of about 3%. The estimate combines data from different sources, but it is closer to the figures from earlier well-designed serial surveys of authors publishing in six prestigious journals, which found 1.4% in 1996 and 0.16% in 2008.
Citing the joint survey of the two associations' members, the 2009 AMWA keynote address read it as showing that members were doing less ghostwriting but that 42% of them were still ghostwriting, called that not good enough, and held that the figure was likely to underestimate the true problem because these writers had at least recognised the importance of joining a professional association; the target it set was to get the figure below 1%. The survey itself reported 41.7% for 2008 as the mean weighted percentage of manuscripts with undisclosed contributions, among participants who had made substantial contributions to manuscripts, without defining "substantial".
The authors' conclusions
The survey authors judged the 44% fall encouraging, a 34% rate among medical writers still unacceptable, and the failure to sustain the improvement both disappointing and perplexing, leaving no room for complacency; they challenged medical writers and professional organisations to intensify collaborative efforts to promote transparency and to research how to achieve best practices in medical publication.
Writers who refuse to ghostwrite can take heart in knowing that their requests for acknowledgment are likely to be granted. The strong correlations between guideline awareness and ethical practice suggest the need for renewed and expanded educational campaigns.
Checking the circulating prevalence estimates
The first systematic review of the prevalence of ghostwriting, presented in 2012, set out to challenge the perception that 50% to 100% of industry articles are ghostwritten, a perception that undermines advocacy efforts highlighting ethical publication practices. It included primary and secondary English-language publications reporting a numerical estimate of prevalence. Of 137 publications retrieved, 25 were eligible: nine original research reports and sixteen reviews or commentaries.
The estimates varied markedly, depending on whether a definition of ghostwriting was given at all, whether the definition used was conservative, and whether errors had been made in re-reporting data. Recent estimates from original research using conservative definitions put the prevalence at 0.2% to 4.3% and decreasing. Ghostwriting is unethical, but the evidence that it is pervasive is often misleading, misguided and mistaken, and sensationalising the prevalence will not reduce it. The review points to standardising the definition of ghostwriting and reporting results accurately, and holds that the most robust evidence, which shows ghostwriting low and decreasing, should be used in advocacy.
Company policies on using and acknowledging writers
In August 2007 a 16-item questionnaire was emailed to more than 700 AMWA members whose titles or affiliations suggested they might be involved in publication planning and development at their companies. It surveyed medical writers at medical communication companies and at developers and marketers of health care products about their companies' policies, perceptions and practices, where earlier surveys had asked authors or counted acknowledgments in published articles; independent medical writers were not included, and not everyone answered every question, so the denominator varies from item to item. Sixty-one people responded: 20 from large and 12 from small pharmaceutical companies, 13 from medical communications companies, 7 from biotechnology, 5 from device companies and 4 from contract research organisations, with 93% based in the United States.
A third of responders said their companies were more likely to use medical writers than five years earlier and 63% saw no change, and the main reasons for using medical writers were speed and quality. Responders estimated that when writers were involved in drafting an article, 51% of the content was controlled by the author, 32% by the writer and 19% by the company.
Almost half of responders (49%) said their company always acknowledges medical writers, but the figure split sharply by employer: 67% in pharmaceutical companies against 18% in medical communications companies. Sixty-four percent thought writers were more likely to be acknowledged than five years earlier, 34% saw no change, and only 1 of 53 thought acknowledgment had become less likely. The survey's author read acknowledgment as appearing to become more likely, while almost half of companies still had no policy or never acknowledged writers. Eighty-eight percent did not think acknowledgment made acceptance less likely. The survey read these results as suggesting that publicity about the ethics of undisclosed writing assistance appeared to be having an effect, and that the trend was towards more use of medical writers alongside more transparency about their involvement.
Responders were self-selected, the response rate among appropriate recipients is not known, and AMWA members may not represent medical writers as a whole; if AMWA members tend to drive transparency at their companies, that would bias the results. Whether the mix of company types in the sample matches the mix of companies that actually hire writers is also unknown.
Writers acknowledged in journal articles
Acknowledgment of medical writers in journal articles was counted in a review of 581 articles from nine medical journals: Annals of Internal Medicine, Archives of Internal Medicine, the British Medical Journal, the Canadian Medical Association Journal, the Croatian Medical Journal, JAMA, the Lancet, the New England Journal of Medicine and the New Zealand Medical Journal. Taking the International Committee of Medical Journal Editors' 2001 article on sponsorship, authorship and accountability as an arbitrary dividing point, it compared issues published in June and December 2000 with issues published in January and December 2007. An acknowledgment counted as naming a medical writer when it recorded writing assistance or support, editorial assistance, preparation of the manuscript, or assistance with preparation of the manuscript, and did not when it recorded review of a manuscript, technical support, or secretarial or administrative support; letters, editorials and news sections were left out of the sample because a writer was unlikely to have been involved in them.
The frequency of acknowledgment of a medical writer doubled, though the numbers were small. Seventeen of the 334 articles from 2000 (5.1%) acknowledged a medical writer, against 28 of the 247 from 2007 (11.3%); seven of the nine journals rose, while the British Medical Journal and the New Zealand Medical Journal fell. In 2000 three journals carried no such acknowledgment and 13 of the 17 appeared in three others, whereas in 2007 every journal but the British Medical Journal carried at least one and 24 of the 28 were in five. Over the same interval the listing of individual author contributions rose from 34.1% to 59.1% of all articles reviewed, while acknowledgment of a funding source barely moved, at 61.7% and 61.1%.
The acknowledgments seldom said who funded the writing. The 28 from 2007 split into 11 for editorial assistance, 12 for assistance with preparation of the manuscript and 5 for preparation of the manuscript, and only 5 of the 28 recorded that the writing or editorial support had been funded by or provided on behalf of a pharmaceutical company. Funding for the study itself was recorded more often: 12 of the 17 articles from 2000 that named a writer also named a funding source for the study, as did half of the 28 from 2007.
The increase may not generalise beyond the journals reviewed, because only a selected group of journals was examined and the issue dates were chosen arbitrarily; confirming it needs a wider range of journals over a longer period, such as a year rather than two months of a year. Its causes are probably more than one: the climate of transparency, journals' instructions to authors and the ICMJE statement on authorship criteria may together have provided the impetus. Annette Flanagin and colleagues, surveying corresponding authors of articles published in 1996, found an 11% prevalence of ghost-authored articles, counting as a ghost author either someone not listed whose contributions merited authorship or an unnamed individual who participated in writing the article; most ghost authors (88%) met the first definition. The review's author read the doubling as suggesting that transparency about writers' contributions is improving, while ghostwriting may still be occurring.
Writing support and the reporting of trial abstracts
Declared professional writing support was not associated with better overall adherence to CONSORT for abstracts. A research note analysed 463 abstracts of randomised controlled trials published between 2011 and 2014 in five journals, the New England Journal of Medicine, Annals of Internal Medicine, the Lancet, The BMJ and JAMA, and found acknowledged professional medical writing in 66 of them, 14.3%. Mean adherence to the CONSORT for abstracts items was 64.3% in the articles with professional writing support and 66.5% in those without; writing support was associated with lower adherence to reporting the study setting and with higher adherence to disclosing harms and side effects and the funding source. The data may not be generalisable to the biomedical literature as a whole, and although the third good publication practice guideline encourages transparency about writing support, support may not have been consistently acknowledged in the dataset.
Writing support in retracted publications
Papers that declared a medical writer accounted for very few retractions. A study of retracted publications set out to determine for the first time what proportion of them, especially those retracted for misconduct, involved declared medical writers or pharmaceutical industry sponsorship, and what factors may be associated with misconduct retractions. Papers with both declared medical writing and declared industry support accounted for very few retractions and none of the misconduct retractions; papers with one or the other also accounted for very few; and the group with no declared industry funding accounted for almost all of the retractions and almost all of the misconduct retractions. In the AMWA member survey that was the latest in 2009, ethics was "by far the greatest concern" of members.
Odds ratios from the same analysis show which features went with misconduct rather than mistake retractions. Against mistake retractions as the control group, the odds of being retracted for misconduct were significantly lower where medical writers or the pharmaceutical industry were involved, and significantly higher where the paper had a single author, where the first author had at least one other retraction, which supports the concept of the serial offender, or where the first author was affiliated with a low- or middle-income country. The study's authors concluded that the declaration of a medical writer on a paper should be seen as a good sign rather than a bad one, and asked why medical writers and the industry are treated as guilty until proven innocent.