Systematic review on the primary and secondary reporting of the prevalence of ghostwriting in the medical literature

Abstract

Background

Ghostwriting of industry-sponsored articles is unethical and is perceived to be common practice.

Objective

To systematically review how evidence for the prevalence of ghostwriting is reported in the medical literature.

Data sources

MEDLINE via PubMed 1966+, EMBASE 1966+, The Cochrane Library 1988+, Medical Writing 1998+, The American Medical Writers Association (AMWA) Journal 1986+, Council of Science Editors Annual Meetings 2007+, and the Peer Review Congress 1994+ were searched electronically (23 May 2013) using the search terms ghostwrit*, ghostauthor*, ghost AND writ*, ghost AND author*.

Eligibility criteria

All publication types were considered; only publications reporting a numerical estimate of possible ghostwriting prevalence were included.

Data extraction

Two independent reviewers screened the publications; discrepancies were resolved by consensus. Data to be collected included a numerical estimate of the prevalence of possible ghostwriting (primary outcome measure), definitions of ghostwriting reported, source of the reported prevalence, publication type and year, study design and sample population.

Results

Of the 848 publications retrieved and screened for eligibility, 48 reported numerical estimates for the prevalence of possible ghostwriting. Sixteen primary publications reported findings from cross-sectional surveys or descriptive analyses of published articles; 32 secondary publications cited published or unpublished evidence. Estimates on the prevalence of possible ghostwriting in primary and secondary publications varied markedly. Primary estimates were not suitable for meta-analysis because of the various definitions of ghostwriting used, study designs and types of populations or samples. Secondary estimates were not always reported or cited correctly or appropriately.

Conclusions

Evidence for the prevalence of ghostwriting in the medical literature is limited and can be outdated, misleading or mistaken. Researchers should not inflate estimates using non-standard definitions of ghostwriting nor conflate ghostwriting with other unethical authorship practices. Editors and peer reviewers should not accept articles that incorrectly cite or interpret primary publications that report the prevalence of ghostwriting.

Strengths and limitations of this study.

• First systematic review on the reporting of the prevalence of ghostwriting in the medical literature.

• A broad search strategy was employed with few restrictions to minimise any potential for publication or language bias; there were no restrictions on language and all study designs and publication types were considered.

• Study heterogeneity in the outcomes reported and populations investigated precluded synthesis of the data.

• The retrospective and self-reported nature of the data collection increased the risk of selection bias in the studies.

• Most included studies were not broadly representative of the peer-reviewed medical literature.

Introduction

Ghostwriting occurs when paid or unpaid writing contributions to a manuscript that do not meet authorship criteria are not disclosed in the acknowledgments.¹ This practice is considered to be distinct from ghost authoring, where contributions to a manuscript that do merit authorship are not disclosed in the author byline.¹ However, part of the challenge in understanding the prevalence of ghostwriting is the current confusion surrounding these unethical practices. For example, the widely used Council of Science Editors definition of ghost authors (ie, individuals who participate in research, data analysis and/or writing of a manuscript but are not named or disclosed in the author byline or acknowledgments) does not distinguish between ghost authors and ghostwriters.²

Ghostwriting of peer-reviewed journal publications in the medical literature is believed to be common practice.^3–5 This belief is supported, to a large extent, by highly publicised cases, primarily from the 1990s and early 2000s, of pharmaceutical companies and authors who had used ghostwriters to prepare manuscripts for publication in medical journals.^6–8 Such cases are highly unethical because the role of the commercial sponsor and any other potential conflicts of interest were hidden. Estimating the prevalence of ghostwriting in the medical literature has been hindered not only by the different definitions of ghostwriting and ghost authoring, but also by the failure of authors to distinguish ghostwriting, which is unethical, from professional medical writing support, which is ethical.^{9 10} Ghostwriters keep their involvement in a manuscript hidden, whereas professional medical writers disclose their involvement and follow ethical publication practices.^{1 9–11} Consistent with the authorship criteria recommended by the International Committee of Medical Journal Editors (ICMJE), professional medical writers who provide writing assistance and do not meet all of the ICMJE authorship criteria should be acknowledged rather than listed as authors.¹² The prevalence of disclosed professional medical writing assistance in medical journals has been estimated to be between 6% and 11%^{10 13} and the legitimate value that this medical writing assistance can bring to improving the quality, timeliness and integrity of reporting in medical journals has been demonstrated.^{1 14 15} However, the exact prevalence of ghostwriting and other forms of undisclosed contributions to papers published in medical journals is unknown.

The primary objective of this study was to conduct the first systematic review on how the prevalence of ghostwriting is reported in the medical literature. The secondary objectives were to assess the variability of the reported estimates of the prevalence of ghostwriting and investigate the source for these estimates.

Materials and methods

Literature search strategy

A search strategy was developed to retrieve publications from the medical literature that reported quantitative estimates of the prevalence of ghostwriting. The following databases were searched on 23 May 2013: MEDLINE via PubMed (1966+); EMBASE (1966+); The Cochrane Library, including the Cochrane Database of Systematic Reviews and Cochrane Central Register of Controlled Trials (1988+); The AMWA Journal (1986+, journal of the American Medical Writers Association); Medical Writing (1998+, journal of the European Medical Writers Association, formerly known as The Write Stuff); Council of Science Editors annual meetings (2007+); and the International Congress on Peer Review and Biomedical Publication meetings (1994+). General text was searched using the following search terms: ghostwrit*, ghostauthor*, ghost AND writ*, ghost AND author*. Truncation symbols and Boolean operators (AND, OR) were used wherever possible.

Two reviewers (SS, Sericka McGee) independently screened the title and abstracts of all retrieved publications using prespecified eligibility criteria; the full text of potentially eligible publications was screened to confirm eligibility for inclusion. Any discrepancies between the reviewers were resolved by consensus. The reference lists of relevant reviews and other publications were screened by hand to identify any additional publications for inclusion.

Eligibility criteria

The eligibility criteria were prespecified. Publications were included if they reported a numerical estimate of the prevalence of possible ghostwriting. Publications were excluded if they were duplicate publications from different databases or abstracts of subsequently published full-text articles, did not report any outcomes related to ghostwriting or reported qualitative estimate(s) of the prevalence of possible ghostwriting. No restrictions on language were included in the eligibility criteria. If needed, the abstracts of publications written in non-English language were to be translated to assist in screening for eligibility. If publications were eligible for inclusion, the full texts of the publications were to be translated to English.

Data extraction and analysis

The primary and secondary outcomes and data to be collected were prespecified. Data to be collected included publication type and year, study design and sample population, definitions of ghostwriting reported, reported prevalence of possible ghostwriting and the source of the reported prevalence of ghostwriting. The prevalence of possible ghostwriting was reported as published. No unpublished data from the retrieved literature were reported and values for prevalence were not extrapolated from published data.

The primary outcome measure for the systematic review was a numerical estimate of the prevalence of possible ghostwriting. The reported definitions of possible ghostwriting were categorised according to the following standard definitions, irrespective of the term used to describe the practice in the publication. Undisclosed writing contributions to a manuscript were defined as: (1) ghostwriting if they were described as not meriting authorship and were not listed in the acknowledgments;^{9 16} and (2) ghost authoring if they did merit authorship and were not listed in the author byline.¹ Consistent with recommendations from international medical journal editors,^{12 14} Good Publication Practice guidelines¹¹ and professional medical writing associations,^{17 18} disclosed writing contributions to a manuscript that did not merit authorship and were disclosed in the acknowledgments were not categorised as ghostwriting.

The key factors considered when assessing study quality were study design, the population assessed and how ghostwriting was defined. While there is no consensus on the best practice for reporting survey research,¹⁹ the quality of the cross-sectional surveys was determined by an assessment of the following factors: validation or pretesting of the survey questions, anonymity of the response, sample size and response rate.

Results

Publication selection

A total of 848 articles were retrieved from the literature search, 800 were excluded and 48 met the eligibility criteria and were included in the systematic review (figure 1). The main reasons for exclusion were publications not relevant to ghostwriting (n=539), duplicate publications from different databases (n=129) and publications not reporting a numerical estimate of the prevalence of possible ghostwriting (n=124). The titles or abstracts of the eight publications that were excluded because the full text or abstract could not be retrieved were reviewed; none were considered to report a numerical estimate of the prevalence of possible ghostwriting. Overall, eligible publications included 16 primary publications that reported original research on the prevalence of possible ghostwriting and 32 secondary publications that cited published or unpublished evidence of possible ghostwriting.

Figure 1.

Selection of publications reporting a numerical estimate of the prevalence of ghostwriting. Electronic databases were searched on 23 May 2013.

^a Journal of the European Medical Writers Association, formerly known as The Write Stuff.

^b Journal of the American Medical Writers Association.

^c Council of Science Editors Annual Meetings.

^d International Congress on Peer Review and Biomedical Publication.

Primary publications

Publication characteristics

Of the 16 primary publications (table 1), there were 13 full-text publications^{6 8 20–30} and three conference abstracts^31–33 that reported a numerical estimate of possible ghostwriting. Twelve publications^{20–28 31–33} reported findings from cross-sectional surveys and four publications^{6 8 29 30} reported findings from descriptive analyses of published articles. Most cross-sectional surveys were reasonably well reported. The surveys used were not validated but most included pretested questions, required an anonymous response, were conducted in targeted populations (ie, all individuals invited to participate were involved in the preparation of peer-reviewed manuscripts), and reported a sample size and response rate. Of the cross-sectional surveys conducted in targeted populations (eg, corresponding or first authors), six had response rates greater than 50%, two had response rates less than 50%, and one did not report a response rate (table 1). Of the cross-sectional surveys conducted in non-targeted populations (eg, individuals invited to participate who may or may not have been involved in providing medical writing assistance for peer-reviewed publications), three had low (12% to 28%) response rates and one did not report a response rate (table 1).

Table 1.

Characteristics of primary publications reporting original evidence of ghostwriting

Publication	Population	Survey description	Data collected	N	RR
Cross-sectional surveys—authors or corresponding authors
Flanagin et al27	Six general medicine journals*	Self-administered, postal, anonymous, pretested, targeted	1996	809	69%
Mowatt et al23	Published Cochrane reviews	Self-administered, online, anonymous, pretested, targeted	1999	362	63%
Hao et al32 (Abstract)	Chin Med J	Self-administered, email, not anonymous, targeted	2008†	220	86%
Dotson and Slaughter28	Three pharmacy journals	Self-administered, online, anonymous, pretested, targeted	2009	112	25%
Mirzazadeh et al24	Three Iranian journals	Self-administered, email, not anonymous, targeted	2009–2010	NR	NR
Wislar et al20	Six general medicine journals‡	Self-administered, online, anonymous, pretested, targeted	2008	622	70%
Ghajarzadeh26	Arch Iran Med (based on student theses)	Self-administered, email, anonymous, targeted	2005–2007	30	49%
Vinther and Rosenberg21	Ugeskr Laeger and Dan Med J	Self-administered, online, anonymous, pretested, targeted	2010	272	62%
Cross-sectional surveys—healthcare professionals
Price et al22	Health academic staff	Self-administered, postal, anonymous, pretested, targeted	NR	166	59%
Rees et al31(Abstract)	Registered users of EPG Online	Self-administered, online, anonymous, non-targeted	NR	295	NR
Cross-sectional surveys—members of medical writing associations
Jacobs and Hamilton25	EMWA/AMWA members	Self-administered, online, anonymous, pretested, non-targeted	2005 2008	843 773	28% 14%
Hamilton and Jacobs 201233(Abstract)	EMWA/AMWA members	Self-administered, online, anonymous, pretested, non-targeted	2011	620	12% §
Publication reviews and descriptive analysis
Healy and Cattell6	Articles on sertraline	NA	1998–2000	96	NA
Gøtzsche et al30	Articles on Danish industry-initiated trials approved 1994–1995	NA	NR	44	NA
Ross et al8	Reviews on rofecoxib associated with Merck support	NA	1996–2004	72	NA
Suda et al29	Non-inferiority clinical trials	NA	1989–2009	583	NA

*Annals of Internal Medicine, JAMA, the New England Journal of Medicine, American Journal of Cardiology, American Journal of Medicine, American Journal of Obstetrics and Gynecology.

†Submission date.

‡Annals of Internal Medicine, JAMA, the New England Journal of Medicine, Lancet, Nature Medicine, PLoS Medicine.

§Personal communication, C Hamilton.

AMWA, American Medical Writers Association; EMWA, European Medical Writers Association; NA, not applicable; NR, not reported; RR, response rate.

Not all cross-sectional surveys were broadly representative of the peer-reviewed medical literature. The cross-sectional surveys were conducted in single populations of academic staff or healthcare professionals,^{22 31} members of medical writing associations,^{25 33} or corresponding or first authors. The corresponding or first authors were surveyed from single journals (Cochrane reviews, Chinese Medical Journal, Archives of Iranian Medicine),^{23 26 32} two Danish journals,²¹ three Iranian journals,²⁴ and three pharmacy journals.²⁸ Two cross-sectional surveys of corresponding authors were conducted in six general medicine journals.^{20 27}

The descriptive analyses of published articles were conducted in single populations and included an analysis of sertraline publications from 1998 to 2000,⁶ rofecoxib reviews from 1996 to 2004,⁸ publications from industry-initiated trials in Denmark from 1994 to 1995,³⁰ and from published non-inferiority trials from 1989 to 2009.²⁹

Reported prevalence of ghostwriting

The reported prevalence of possible ghostwriting in the primary publications varied markedly and was not suitable for meta-analysis because of the different populations assessed, the different methods used to generate the estimates and the various definitions that were used (tables 1 and 2). All cross-sectional surveys reported a definition of ghostwriting or ghost authoring, but most definitions did not differentiate contributions that merited authorship from those that did not merit authorship (table 2).

Table 2.

Primary publications reporting estimates of the prevalence of possible ghostwriting

Publication	Measure of possible ghostwriting reported by authors	estimate % (n)
Cross-sectional surveys reporting ghostwriting
Flanagin et al27	Unnamed individual who participated in the writing	1.4% (11/809) of articles
Wislar et al20	Unnamed individual who participated in the writing	0.2% (1/622) of articles*
Jacobs and Hamilton, Hamilton and Jacobs 25 33	Undisclosed medical writing assistance not qualifying for authorship	2005: 61.8% (NR) 2009: 41.7% (NR) 2011: 33.0% (NR) of articles†
Cross-sectional surveys reporting combined ghost authoring and ghostwriting
Flanagin et al27	Failure to name, as an author, individuals who made substantial contributions to the research or writing or an unidentified medical writer	11.5% (93/809) of articles
Price et al22	Failure to name, as an author, individuals who made substantial contributions to the research or writing	24.1% (40/166) of authors
Mowatt et al 23	Individual merited authorship or had assisted with drafting but not listed as an author or acknowledged	8.8% (32/362) of articles
Hao et al32	English-language speakers assisted with writing but not identified as authors or acknowledged	10.4% (NR) of authors
Dotson and Slaughter28	Failure to name, as an author, individuals who made substantial contributions to the research or writing	0.9% (1/112) of articles
Wislar et al20	Failure to name, as an author, individuals who made substantial contributions to the research or writing of the article or an unnamed individual who participated in the writing	7.9% (49/622) of articles
Cross-sectional surveys reporting ghost authoring
Mirzazadeh et al24	Failure to name, as an author, individuals who made substantial contributions to the research	21.4% (25/NR) of authors
Ghajarzadeh26	Failure to name, as an author, students who made substantial contributions to the research‡	0.7% (2/296) of articles
Vinther and Rosenberg21	Individual merited authorship but not listed as an author	2.4% (6/245) of articles
Rees et al31	Individual merited authorship but not listed as an author	70% (NR/202) of published authors
Publication reviews and descriptive analysis reporting possible ghost authoring or ghostwriting
Healy and Cattell6	Published articles coordinated by a medical information company, including acknowledged medical writing support§	57.3% (55/96) of articles
Gøtzsche et al30	Individuals who wrote the trial protocol, conducted the statistical analyses or wrote the manuscript but were not listed as authors, not members of a study group or steering committee or not disclosed in an acknowledgment	75.0% (33/44) of trials
Ross et al8	Published reviews associated with Merck support and with a single external author¶	69.4% (50/72) of reviews

*Available as online supplementary data.

†Values represent the mean weighted percentage of publications that were ghostwritten by respondents. Findings were weighted in proportion to the number of manuscripts the respondent wrote per year.

‡Students were classified as ghostwriters if the student was not named as an author and if the results reported in the publications were based on the results of their theses.

§Authors conclude data provide quantification of the possible extent of ghostwriting based on a single drug. Of the 55 published articles that were coordinated through a medical information company, 2 included medical writing assistance that was acknowledged in the published article.

¶Published review articles had been identified from correspondence between Merck and a medical publishing company, from Merck publication status reports, or were affiliated with an author named within the correspondence or publication status reports. The authors did not report whether medical writing assistance was acknowledged in the published article.

NR, not reported.

Four cross-sectional surveys reported the prevalence of ghostwriting where the definition provided could be categorised as undisclosed contributions that do not merit authorship.^{20 25 27 33} Findings from these surveys, which were repeated at different time points, suggested that the prevalence of ghostwriting is low and decreasing. Flanagin²⁷ and Wislar²⁰ conducted two similar cross-sectional surveys of corresponding authors of articles published in six general medicine journals in 1996²⁷ and in 2008.²⁰ The surveys, which included a core set of three journals (Annals of Internal Medicine, JAMA, the New England Journal of Medicine) at each time point and in targeted populations, used pretested questions that required an anonymous response and had response rates greater than 65%. The prevalence of ghost authoring (which included ghostwriting) was estimated to be 11.5% in 1996 and 7.9% in 2008. The prevalence of ghostwriting from these surveys was estimated to be 1.4% in 1996 and 0.2% in 2008. Hamilton and Jacobs conducted a cross-sectional survey of members of two medical writing associations in 2005, 2008 and 2011.^{25 33} The survey, which was not conducted in targeted populations, used pretested questions that required an anonymous response and had response rates from 12% to 28% of the total population surveyed (ie, not all participants surveyed were involved in the preparation of peer-reviewed manuscripts). Findings from these surveys showed that the mean weighted percentage of publications that were ghostwritten by respondents had decreased from 61.8% in 2005 to 33% in 2011. Findings were weighted in proportion to the number of manuscripts the respondent wrote per year.

Six cross-sectional survey publications reported the prevalence of possible ghostwriting (ie, the prevalence of ghostwriting and ghost authoring combined).^{20 22 23 27 28 32} The prevalence of possible ghostwriting reported in these cross-sectional surveys varied from 0.9% to 24.1% of publications or authors.

Four cross-sectional survey publications reported the prevalence of ghost authoring.^{21 24 26 31} The prevalence of ghost authoring reported in these cross-sectional surveys varied from 0.7% to 70% of publications or authors.

The descriptive analyses of published articles on sertraline and rofecoxib did not include a prespecified definition of ghost authoring or ghostwriting^{6 8}; in these studies, possible ghostwriting or ghost authoring was assumed in publications associated with industry-sponsored support (table 2). The descriptive analysis of industry-initiated trials in Denmark used a non-standard definition of ghost authors; in this study ghost authors were defined as individuals, not named as authors, who were involved in writing the protocol, conducting the statistical analyses or writing the manuscript.³⁰ The prevalence of possible ghostwriting in the two descriptive analyses of single drugs that did not include a prespecified definition of ghostwriting was 57% of articles on sertraline published from 1998 to 2000⁶ and 69% of reviews on rofecoxib published from 1996 to 2004.⁸ The prevalence of possible ghostwriting in the descriptive analysis of Danish industry-initiated trials, which used a non-standard definition of authorship, was 75% of Danish initiated trials approved in 1994 to 1995.³⁰ The descriptive analysis of non-inferiority trials²⁹ was not considered further as the definition of ghostwriting used was consistent with disclosed medical writing assistance. In this study ghostwriters were defined as acknowledged individuals, other than authors, who contributed to the writing and were affiliated with an industry-sponsored study.²⁹ The prevalence of disclosed medical writing assistance in this descriptive analysis was 17.3% (101/583) of clinical trials.

Secondary publications

Publication characteristics

Of the 32 secondary publications, there were 13 review articles, 10 editorials, 5 commentaries, 3 news articles and 1 government report, with most being published after 2008 (table 3). Most publications cited primary sources as evidence of the prevalence of possible ghostwriting (table 3), with the cited prevalence of possible ghostwriting varying from 6% of publications to 100% of publications involving drugs. In most secondary publications, the information on the prevalence of ghostwriting was not reported consistently compared with the cited evidence (table 3). Misleading and mistaken information was reported in many publications that: (1) did not distinguish between ghostwriting and ghost authoring; (2) included acknowledged medical writing assistance or a combined estimate of guest authorship and ghost authorship as ghostwriting; (3) generalised findings from publication reviews and analyses of specific data sets to wider populations of publications or industry-sponsored trials; or (4) cited personal communications or informal surveys where the original source was unpublished and could not be verified (table 3).

Table 3.

Characteristics of secondary publications citing evidence of possible ghostwriting

Publication	Cited evidence	Consistent with cited source(s)	Comment
Elliot34	Primary	No	Did not distinguish GW from GA
Moffat and Elliott35	Primary	No	Did not distinguish GW from GA
Schiefe36	Primary	No	Did not distinguish GW from GA
MacLennan et al37	Primary	No	Did not distinguish GW from GA
Nahai38	Secondary	No	Did not distinguish GW from GA
Ngai39	Primary	Yes	Did not distinguish GW from GA
Bosch40	Primary	Yes	Did not distinguish GW from GA
Wiwanitkit41	Primary	No	Did not distinguish GW from GA or GA from guest authorship
Krimsky42	Primary	No	Did not distinguish GW from GA or GA from guest authorship
Langdon-Neuner 200843	Primary and secondary	No	Did not distinguish GW from GA or GW from disclosed medical writing support
Tharyan44	Primary	Yes	Did not distinguish GW from GA or GW from disclosed medical writing assistance
Paul45	Primary	Yes	Did not distinguish GW from GA from disclosed medical writing assistance
Bavdekar46	Primary	No	Did not distinguish GW from GA and generalised evidence to a wider population
Górski and Letkiewicz47	Primary and secondary	No	Did not distinguish GW from GA and generalised evidence to a wider population
Matias-Guiu and García-Ramos48	Primary and secondary	No	Did not distinguish GW from GA and generalised evidence to a wider population
McHenry49	Gov. report	No	Generalised evidence to a wider population
Healy50	Primary and secondary	No/ND	Generalised evidence to a wider population and reported personal opinion of GW prevalence
Abbasi4	Secondary	No	Secondary publication cited named individual
Mitrany51	None	ND	Cited named individual
Collier5	None	ND	Cited named individual
Kmietowicz52	None	ND	Cited named individual
Matthews3	None	ND	Cited unpublished data
Bonita et al53	Primary	Yes	Consistent with cited source
Anon 201154	Primary	Yes	Consistent with cited source
Anon 200755	Primary	Yes	Consistent with cited source
Editors15	Primary	Yes	Consistent with cited source
Baethge56	Primary	Yes	Consistent with cited source
Flanagin 201057	Primary	Yes	Consistent with cited source
Murray et al58	Primary	Yes	Consistent with cited source
Moore59	Secondary	Yes	Consistent with cited source
Hargreaves60	Primary	Yes	Consistent with cited source
Jones61	Primary	Yes	Consistent with cited source

CSE, Council of Science Editors; GA, ghost authoring; Gov, government; GW, ghostwriting; ND, not determined.

Case study of misleading and mistaken evidence

An analysis of secondary publications citing the possible prevalence of ghostwriting in sertraline publications provides an illuminating case study on how misleading or mistaken evidence can enter and remain in the medical literature. In 2003, Healy and Cattell⁶ showed that 57% (55/96) of articles on sertraline published from 1998 to 2000 were coordinated through a medical information company and only two acknowledged medical writing assistance. The authors concluded that these data provided information on the ‘possible extent of ghostwriting based on a single drug’. On October 14, 2004, Healy also provided evidence on the influence of the pharmaceutical industry on key groups at a UK House of Commons Health Select Committee investigation.⁶² In answer to Question 197, Healy stated the following:

My estimate is that, even in journals like the BMJ, the Lancet, the New England Journal of Medicine and JAMA, the leading journals in the field, if these articles have to do with therapeutics, with drugs, it may be worse perhaps for psychiatry than elsewhere, but I doubt it, 50% of these articles are ghostwritten. It may be higher.

Healy and Cattell's original evidence has been cited incorrectly and interchangeably with Healy's statement to the House of Commons Health Select Committee investigation (figure 2). Although findings from the primary publication have been cited and interpreted correctly in two secondary publications,^{34 35} many secondary publications incorrectly cited Healy's original evidence and statement to the House of Commons Health Select Committee investigation (figure 2).

Figure 2.

Case study of original versus cited evidence of ghostwriting prevalence. Contrast between original evidence on the prevalence of possible ghostwriting in sertraline publications from 1998 to 2000 and the subsequent citations of this evidence. Source of the original published evidence: Healy and Cattell⁶ showed that 55 of 96 (57%) articles published on sertraline from 1998 to 2000 were coordinated through a medical communications company and concluded that these data provided information on the ‘possible extent of ghostwriting based on a single drug’. Inaccurate reporting from the cited source is marked with a cross.

Discussion

This systematic review on the reported prevalence of ghostwriting in the medical literature showed that the estimates of the prevalence of ghostwriting in primary publications varied markedly and were influenced by the definitions used, the types of study designs and the type of population or sample assessed. In addition, secondary publications often cited outdated, misleading or mistaken evidence of the reported prevalence of ghostwriting, with many publications not distinguishing ghostwriting from ghost authoring.

Although evidence from descriptive analyses can highlight the extent of ghostwriting in single populations, evidence from well-conducted cross-sectional surveys have the potential to provide estimates of the prevalence of ghostwriting that may be generalised to the majority of peer-reviewed publications. Despite this, many of the cross-sectional surveys retrieved in this systematic review were conducted in limited populations that were not broadly representative of the peer-reviewed medical literature. In addition, many of the cross-sectional surveys did not differentiate between contributions that merited authorship from those that did not merit authorship and provided, at best, an estimate of possible ghostwriting. The reported prevalence of ghostwriting, where ghostwriting was defined as undisclosed contributions that did not merit authorship, was retrieved from the two cross-sectional surveys of corresponding authors from several general medicine journals and by cross-sectional surveys (repeated on three separate occasions from 2005 to 2011) of members of two major medical writing associations. Together the findings from these surveys suggested that the prevalence of ghostwriting has decreased in recent years. However, while the findings from these surveys may be considered more broadly representative of the peer-reviewed medical literature than surveys focused on single journals, single countries or single subject areas, interpretation of these findings should take into account that respondents were required to retrospectively self-report potentially unethical or unprofessional behaviour.

Unethical authorship practices are a major concern and are an increasingly recognised problem in the medical literature.⁶³ As the findings from this systematic review suggest, some of these perceived problems may arise from the considerable, but unnecessary, confusion and disagreement surrounding the definitions of ghostwriting. As recognised by the World Association of Medical Editors, professional medical writers can have a legitimate role in assisting authors to communicate their research findings in the peer-reviewed literature.¹⁴ Professional medical writers can and do improve the timeliness and quality of reporting and can assist investigators and industry sponsors to meet their ethical commitments to the disclosure and publishing of clinical trial results.^{1 64–66} As such, the misleading and mistaken reporting of the prevalence of ghostwriting that was evident in the secondary publications retrieved in this systematic review is disappointing. Specifically, the confusion surrounding definitions of ghost authorship and ghostwriting and the unbalanced focus on industry as the source of unethical authorship practices in the secondary publications takes attention away from the need to focus on all types of unethical contributions in peer-reviewed publications, irrespective of the source of the unethical practice.

The various definitions of ghostwriting and ghost authoring found in this systematic review highlight the considerable disagreement in the medical literature with regard to the definitions of ghostwriting and ghost authoring and whether a distinction should be made between these unethical practices. Although the Council of Science Editors uses a straightforward and broad definition of ghost authors that does not distinguish between ghost authors and ghostwriters,² the distinction between ghost authorship and ghostwriting used in this manuscript is consistent with those used in the cross-sectional surveys conducted by Flanagin²⁷ and Wislar²⁰ and with the authorship criteria recommended by ICMJE.¹² Distinguishing between these practices is important because it differentiates between undisclosed contributions that do and do not merit authorship. However, part of the reason for the disagreement about definitions of ghost authoring and ghostwriting stems from differences of opinion in the medical literature over how authorship is defined, particularly with regard to what constitutes ‘substantial’ contributions to a publication and the role of disclosure of contributions to a manuscript in the acknowledgments.⁶⁷ The ICMJE criteria for authorship¹² are the most well-recognised authorship criteria available, but leave much about authorship undefined.^{63 68} Despite this, the recent emphasis on author accountability in the ICMJE's fourth criterion is an attempt to better define authors as those persons who are responsible and accountable for the content of a published work. While a professional medical writer can take responsibility for the writing and how research is reported in a manuscript, the professional medical writer cannot take responsibility for the integrity of the research or be accountable for the clinical interpretation of the findings, unless he or she was involved in the generation of the research or its analysis. Responsibility for the integrity of research and accountability for the clinical interpretation of the findings is, and should always be, the responsibility of the authors. Until a more definitive model of authorship is universally accepted, the controversy and disagreement over definitions of authors, ghostwriters and ghost authors will remain. However, no matter what term is used, the practice of misleading readers about potential competing interests and hiding contributions to a published work, no matter whether these contributions should be most appropriately disclosed in the author byline or acknowledgements section, is unethical and totally unacceptable.

The strengths of this systematic review are that a broad search strategy was employed with few restrictions to minimise any potential for publication or language bias. All study designs and publication types were considered and there were no restrictions on language. Although the full text or abstracts of eight publications could not be retrieved, omission of these publications was unlikely to have biased the findings. Review of the abstracts or titles of these publications suggested that none reported numerical estimates of the prevalence of ghostwriting. The major finding of this review was the limitations of the reported evidence of ghostwriting in the medical literature. These limitations included the heterogeneity among the studies in the outcomes reported and populations investigated, the observational study designs and the retrospective nature of data collection. Given the nature of unethical authorship practices, it may not be feasible to conduct a prospective study on ghostwriting. However, based on the findings from this review, recommendations can be made to help researchers, authors, editors and peer-reviewers apply the same rigorous standards that are applied to the conduct and assessment of all clinical research and actively improve the quality of reporting of the evidence of unethical authorship practices.

Researchers should use a standard definition of ghostwriting so that the confusion around ghostwriters and ghost authors is not perpetuated. For example, the descriptive analysis of Danish industry-initiated trials used a non-standard definition of authorship, which is likely to have contributed to the very high prevalence of possible ghostwriting reported in this study (75%). In this study, Gøtzsche et al³⁰ suggested that individuals who write the trial protocol, conduct the statistical analyses or who contribute to the writing of a publication should be included as authors. Indeed, the prevalence of ghost authorship in this study was 91% when the analyses included these individuals, irrespective of whether they had been appropriately acknowledged elsewhere. Given that the Gøtzsche et al³⁰ study has been downloaded more than 3000 times since publication⁶⁹ and cited repeatedly as evidence not only of ghost authorship but also of ghostwriting,^{40 44 46 48} it would have been illuminating if the authors had included an estimate of the prevalence of ghostwriting using standard definitions for comparison. In addition to using standard definitions of ghostwriting, researchers and authors should ensure that cited evidence of ghostwriting is reported accurately without unwarranted generalisations. Publications based on personal commentary should be avoided and studies that use non-standard definitions, specific populations or that were conducted before a change in practice (eg, before the adoption of the Good Publication Practice guidelines for communicating company sponsored research in 2003 (GPP)⁷⁰ and in 2009 (GPP2)¹¹) should be described in context.

Editors and peer reviewers have a duty to prevent outdated, misleading or mistaken evidence on the prevalence of ghostwriting from being published and perpetuated. Close attention should be paid to the internal and external validity of the study, the definitions used, how the data are reported and whether the data are interpreted within the context of current practices. When assessing a submitted manuscript on ghostwriting, editors should consider using peer reviewers with expertise in the appropriate study designs (eg, survey research) and ethical publication practices. Collectively, these actions could help prevent further questionable evidence on the prevalence of ghostwriting from being published.

In conclusion, the findings from this systematic review showed that reports of the prevalence of ghostwriting in the medical literature are limited by the varied definitions used to describe unethical authorship practices, the types of study designs employed and the populations assessed. To improve reporting, researchers should not inflate estimates using non-standard definitions of ghostwriting nor conflate ghostwriting with other unethical authorship practices. While open and transparent debate should be encouraged, editors and peer reviewers should not accept articles that incorrectly cite or interpret primary publications that report the prevalence of ghostwriting.