Do Author Conflicts of Interest and Industry Sponsorship Influence Outcomes of Systematic Reviews and Meta-Analyses Regarding Glaucoma Interventions? A Cross-sectional Analysis

Supplemental Digital Content is available in the text. Précis: In our sample of systematic reviews focusing on treatments for glaucoma, reviews conducted by authors with a conflict of interest were more likely to reach favorable conclusions compared with reviews without conflicted authors. Purpose: Previous studies have demonstrated that authors’ conflict of interest can influence outcomes of systematic reviews. Therefore, we aimed to determine whether the presence of 1 of more conflicts was associated with more favorable results and conclusions in systematic reviews of glaucoma interventions. Materials and Methods: MEDLINE and Embase were searched for systematic reviews of glaucoma treatments published between September 1, 2016 and June 2, 2020. Author conflicts of interest were located using multiple databases (eg, CMS Open Payments Database, Dollars for Profs, Google Patents, the United States Patent and Trademark Office USPTO) and previously published disclosure statements. Study sponsorship was determined using each review’s funding disclosure statement. Results: Our study included 26 systematic reviews conducted by 108 authors. Of these reviews, 9 (35%) were conducted by at least 1 author with an undisclosed conflict of interest. Of those 9, 3 (33%) reported results favoring the treatment group, and 5 (56%) reported conclusions favoring the treatment group. Of the 17 systematic reviews with no conflicted authors, 1 (6%) reported results favoring the treatment group, and 2 (12%) reported conclusions favoring the treatment group. The Fisher exact tests demonstrated that these differences held a statistically significant association between author conflicts and the favorability of the reviews’ conclusions toward the treatment group (P=0.04). Conclusions: We found that systematic reviews conducted by 1 or more authors with conflicts of interest were more likely than those with no conflicted authors to draw favorable conclusions about the investigated intervention.

G laucoma is one of the leading causes of blindness in the United States, 1 and it accounts for US $2.86 billion in medical costs each year. 2 To date, the exact cause of glaucoma is not well understood. Current treatments (eg, eye drops, oral medications, and/or surgical interventions) can slow disease progression but are not curative. 1 Given the high prevalence and the significant health care costs associated with glaucoma, high-quality evidence regarding treatment and management options is vital for making clinical decisions policies. 1 In the field of ophthalmology, evidence from high-quality research designs, such as systematic reviews (SRs), often influences these critical decisions.
SRs use comprehensive and systematic strategies to search bibliographic databases and other resources to locate and synthesize data from multiple, individual studies to better understand the magnitude and direction of an intervention's effect. 3 Given the comprehensive nature of SRs, their results are often used by physicians and professional medical societies to develop clinical practice guidelines. 4 For example, the American Academy of Ophthalmology's Preferred Practice Patterns for primary open-angle glaucoma 5 rates high-quality SRs that synthesize results from clinical trials as I++, the highest level-of-evidence rating per the Scottish Intercollegiate Guideline Network criteria. 6 Given their role in directing patient care, SRs must be objective and free of bias such that results are reliably and accurately represent the totality of available evidence on a specific research topic.
Author conflict of interest (COI) and industry sponsorship have been shown to influence outcomes of medical literature. [7][8][9] For example, Hansen et al 4 found that compared to SRs with no author COIs, SRs with 1 or more confirmed author COI report favorable results more often. Lundh et al 10 determined that industry sponsorship is positively associated with the likelihood of reporting favorable results for sponsor's product. Ahn et al 7 found that financial relationships between principal drug investigators and drug manufacturers are associated with positive results and conclusions of clinical trials. Based on these studies, we sought to determine if a similar association exists within the ophthalmology literature. Our primary objectives were to evaluate for the presence of disclosed and undisclosed author COIs in SRs of glaucoma interventions and to determine whether their narrative results and conclusions were influenced by the presence of those COIs. Our secondary objectives were to determine whether an association exists between risk of bias and author COI and determine whether study sponsorship influences the reported results and conclusions.

Transparency, Reproducibility, and Reporting
To ensure transparency and reproducibility, our a priori protocol and materials are available on the Open Science Framework. 11 While composing our manuscript, we referred to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines 12 and Murad and Wang's 13 guidelines for meta-epidemiological studies.

Training
All researchers completed online and in-person training developed by J.M.A. and M.W. Training included an outline of the study design, objectives, materials, and data extraction training for one example SR. This training session was recorded and is available for reference. 11

Search Strategy
We searched MEDLINE (Ovid) and Embase (Ovid) on June 2, 2020, to identify SRs of glaucoma treatment with or without a meta-analysis. Supplemental Table 1 (Supplemental Digital Content 1, http://links.lww.com/IJG/A525) lists our search strategies. Records from these search strategies were uploaded to the SR screening platform Rayyan (https://rayyan.qcri.org/) for title and abstract screening.

Screening
Two of us (A.W. and D.M.) screened the search results by titles and abstracts in a double-blind manner to include all SRs on the treatment of glaucoma with and without meta-analysis. Following this initial screening, full-text versions were evaluated to determine final eligibility according to the criteria outlined in the next section. Third party adjudication was available, if necessary.

Eligibility Criteria
To be included, a study must have: (1) met the PRISMA-P definition of a SR or meta-analysis 14 ; (2) been a direct comparison of one treatment to another single treatment, another combination treatment, a placebo, or a standard of care; (3) investigated a treatment for glaucoma; (4) been published between September 1, 2016, and June 2, 2020; (5) been published in the English language; and (6) included studies of human participants. The rationale behind the prespecified date range was based on the International Committee of Medical Journal Editors (ICMJE) recommendation to disclose any financial interest within 36 months of journal submission, regardless of whether the financial interest relate to the submitted study. 15 Starting the prespecified date range at September 1, 2016 forward, allowed 36 months from the time of CMS Open Payments Database (OPD) when online in September 2013. Consequently, all eligible SRs in our study could be checked for author COI.

Data Extraction
Two investigators (A.W. and D.M.) extracted data in a masked, duplicate fashion using a pilot-tested Google form. The investigators analyzed the full-text of each SR for the following characteristics: (1) PubMed identification number and/or DOI; (2) journal name; (3) publication date; (4) author names; (5) treatment interventions being compared; (6) author institutional affiliation(s); (7) author country affiliation; (8) funding source; (9) complete COI statement; (10) whether the SR or meta-analysis addressed risk of bias; (11) the verbatim risk of bias statement; (12) whether an author also authored 1 or more of the primary studies included in the SR; (13) total number of self-cited primary studies; (14) primary outcome of the SR; (15) whether an overall pooled effect estimate was calculated; (16) statistical significance of the pooled effect estimate; and (17) whether narrative results and conclusions favored the treatment or comparison group (eg, placebo, standard of care, control). We used the term "conclusion" to describe both the discussion and conclusion sections of the SRs.

Favorability of Narrative Results of Conclusions
We appraised the SR results and conclusions as "favorable," "unfavorable," or "mixed/inconclusive." SRs were coded as favorable if the narrative results only presented positive outcomes, unfavorable if they presented only negative outcomes, and mixed/inconclusive if they presented both positive and negative outcomes. SR conclusions were coded as favorable if they directly stated or implied preference for the treatment group, unfavorable if they stated or implied preference for the comparison group, and mixed/inconclusive if neither a favorable nor unfavorable rating applied (eg, positive subgroup analyses but negative overall population outcomes).

Identification of Undisclosed Conflict of Interest
Our search strategy to identify undisclosed COI was performed using the stepwise algorithm provided in Figure 1. We used the methodology of Mandrioli et al 16 to develop this search strategy; with the inclusion of 3 additional databases: CMS Open Payments, Dollars for Profs, and the United States Patent and Trademark Office (USPTO). Supplemental Table 2 (Supplemental Digital Content 2, http://links.lww.com/IJG/ A526) provides a description of these databases. We searched all authors using the stepwise algorithm, regardless of whether they disclosed a COI within the SR. To ensure accuracy and consistency of searches between investigators, A.W. and D.M. created standardized database search strings for the Google Patents, USPTO, and PubMed databases using the Python programming language (Python Software Foundation, https:// www.python.org/). In accordance with ICMJE COI disclosure policies, we searched PubMed for studies published three years prior to the publication date to assess if the author had disclosed a COI in a previous study that was not disclosed in the SR from our sample. For searches yielding more than 10 manuscripts, random numbers were assigned to each, and then data were extracted from the first 10 publications of the randomized list. If at any time we were unable to verify a corresponding COI for an author in our sample, we considered the search inconclusive and continued our search process. If an undisclosed COI was identified, the search was terminated and the author identified as having an undisclosed COI. A similar search termination strategy was used by Mandrioli et al. 16

Risk of Bias Evaluations
To assess risk of funding bias, we applied the Cochrane Collaboration's criteria combined with the following four items from Mandrioli et al 16 : (1) whether explicit and "welldefined" criteria that could be replicated by others were used to select studies for inclusion or exclusion; (2) whether an adequate study inclusion method (eg, a minimum of 2 assessors selecting studies) was used; (3) whether search strategies were comprehensive (eg, included > 1 database, used complete search strategies, had accessible key terms); and (4) whether methodological differences with the potential to introduce bias were controlled for. Each item was assigned a "yes," "no," or "unclear" response. Overall risk of bias was considered low if at least 3 of the 4 criteria were met and high otherwise.

Statistical Analysis
Results were quantified using descriptive statistics, and relationships were evaluated using The Fisher exact tests, when possible. Stata 16.1 (StataCorp, LLC, College Station, TX) was used for all analyses.

RESULTS
Our searches of MEDLINE (Ovid) and Embase (Ovid) returned 1467 records. Following removal of duplicates, 1089 records were screened by title and abstract. After this initial screening, 213 records were included for full-text analysis. Further review resulted in an additional 187 exclusions, leaving 26 systematic reviews and meta-analyses meeting inclusion criteria (Fig. 2).

Relationship Between COI and Favorability of Results and Conclusions
Of the 9 SRs with 1 or more conflicted authors, 3 (3/9; 33%) reported narrative results favoring the treatment group, and 5 (5/9; 56%) reported conclusions favoring the treatment group. Of the 17 SRs with no conflicted authors, only 1 (1/17; 6%) reported results favoring the treatment group, and only 2 (2/17; 12%) reported conclusions favoring the treatment group. Although the Fisher exact test demonstrated a statistically significant association between author COI and the favorability of the SR conclusions toward the investigated treatment group (P = 0.04), a similar association between author COI and favorability of narrative results was not found (P = 0.21). Table 2 summarizes the results.

Relationship Between Risk of Bias and Industry Sponsorship or Conflicts of Interest
Of the 3 industry-sponsored SRs, 1 (1/3;33%) had a high risk of bias. None of the nonindustry sponsored SRs had a high risk of bias. Of the 17 SRs without a conflicted author, only 1 (1/17; 6%) was considered to have high risk of bias. Similarly, only 1 (1/9; 11%) SR with 1 or more conflicted authors was found to have a high risk of bias. We found no association between risk

Search of Open Payments Database
Was an undisclosed COI found?

Search of ProPublica Dollars for Profs
Was an undisclosed COI found?

Search of United States Patent and Trademark Office
Was an undisclosed COI found?
No Yes Yes

Search of Google Patents
Was an undisclosed COI found?

Search of PubMed
Was an undisclosed COI found?

Yes No
Author marked as having undisclosed COI; search terminated No undisclosed COI were found for this author of bias and the presence of either industry-sponsorship or author COI (Table 3).

DISCUSSION
We found that SRs conducted by 1 or more authors with a COI were more likely to draw favorable conclusions about their studied interventions more often than SRs with no conflicted authors. Furthermore, our results suggest that COIs are common among authors of ophthalmology SRs, though these COIs are often incompletely disclosed at the time of journal submission. Despite multiple SRs in our sample receiving industry support, the presence of industry sponsorship did not appear to influence the nature of their results or conclusions. Here, we provide further discussion of how our findings contribute to the current literature surrounding author COI and offer future recommendations to better understand the influence that industry may have on outcomes of SRs within the field of ophthalmology.
Our findings contribute to the current literature on author COI in several ways. First, our results confirm those from previous studies suggesting that authors disclose all potential competing interests so that readers are fully aware of all conflicts when interpreting a study's conclusions. For example, although we did not observe an association between author COI and a favorable narrative result, we did observe a positive link between an author COI and a favorable conclusion. These findings align with a previously published study investigating the influence of financial ties with pharmaceutical companies on the favorability of results and conclusions of meta-analyses investigating antihypertensive drugs. Yank and colleagues found no association between financial ties and the favorability of results in these meta-analyses, but they did find a positive association between financial ties and conclusions favoring the company's drug. 17 In addition, in a sample of review articles regarding the health effects of passive smoking, Barnes and Bero 18 determined that author COI with the tobacco industry was the only factor associated with the review concluding that passive smoking is not harmful (odds ratio = 88.4; 95% confidence interval, 16.44-76.5; P < .001). Our finding similarly indicates that authors who have a COI may be more likely to frame their results in a favorable way, even if their data do not fully support such a conclusion.
Our findings suggest that COIs are regular occurrences in SRs investigating glaucoma interventions, though they are often not completely disclosed. Specifically, one-fifth of the authors in our sample were found to have a COI, most of which were omitted from the SR's published COI disclosure statement. Previous studies have demonstrated a high rate of incomplete COI disclosure in the biomedical literature. For example, a 2018 study investigating the completeness of oncology drug trialists' COI disclosures revealed that nearly one-third of trial authors failed to disclose payments received from the trial sponsor. 19 Another study published in the Journal of General Internal Medicine found a 69% discordance rate between industry and authorreported COI disclosure statements. 20 Given the influence that author COI may have on study outcomes, several influential stakeholders in academic medicine have tried to improve the accuracy of author COI disclosure; however, further efforts at addressing poor rates of COI disclosure are warranted.
Many influential bodies in the medical literature (eg, ICMJE, 21 Institute of Medicine, 22 World Association of Medical Editors, 23 and Committee of Publication Ethics 24 ) have made efforts to preserve the integrity of academic literature by increasing the transparency of author COI. For example, the ICMJE developed a standardized author COI disclosure form used by all of its member journals. 25 Although many medical journals now use this standard form, others have different COI disclosure policies and requirements. A study of 130 high-impact journals across a variety of medical specialties found that 99% of these journals required authors to disclose COIs, but only 45% of journals required authors to adhere to ICMJE policies and recommendations. 26 Moreover, the definition of a COI varies greatly across disclosure policies. 27 Thus, authors may be unsure of what warrants disclosure at the time of journal submission. The current ICMJE COI disclosure guideline also has drawn criticism for its ambiguous language, such as "potential conflicts of interest" and "relevant financial activities." 28 In an editorial published in the American Journal of Ophthalmology, the journal's editors argue that such language is "ambiguous, problematic, and confusing" because of the inherent subjectiveness on the part of the author when deciding on the relevancy of a potential competing interest. 29 Disclosure policies also offer little benefit if authors cannot be held accountable. Indeed, Shawwa et al 30 found that only 17% of the National Library of Medicine's "core clinical journals" reported having procedures to verify the accuracy of authors' COI disclosures and less than one-fourth of these journals reported potential ramifications on the editorial process if 1 or more authors were found to have incomplete COI disclosures.
Our findings offer future recommendations to better understand the influence that industry may have on outcomes of SRs in the field of ophthalmology. Our recommendations are 3-fold. First, we recommend all journals adopt the ICMJE author COI disclosure policy to help standardize the definition of what constitutes a COI. Second, we urge ICMJE guideline writers to use more specific language (eg, authors should disclose "all" financial relationships) to eliminate any confusion surrounding which financial  relationships warrant disclosure. Finally, we recommend journals adopt a verification process to ensure authors are completely forthcoming about all financial relationships prior to publishing a submitted manuscript. As part of this verification process, we support the use of predetermined consequences for noncompliance with journal COI disclosure policies (eg, publication of notice about noncompliance, manuscript rejection or retraction, prohibition from future manuscript submissions). If implemented correctly, this three-step process will help increase the completeness of author COI disclosure and the public's trust in published ophthalmology literature.

Strengths and Limitations
Regarding strengths, we extracted data in a masked, duplicate fashion, which is regarded as the gold standard in meta research according to the Cochrane Handbook. 31 In addition, we provided our study protocol and all other study materials on the Open Science Framework, thereby increasing transparency and reproducibility of our findings. 11 With regard to limitations, our study was cross-sectional in design; therefore, selection of a different date range may have yielded varying results. Our study focused on SRs investigating head-to-head interventions for the treatment of glaucoma and thus may not be generalizable to the breadth of ophthalmology literature. Due to having a small sample size, careful interpretation of our findings as the lower bound estimate of the influence author COI may have on outcomes of SRs is warranted. Finally, we were unable to determine the accuracy of COI disclosure for non-US-based authors using the Open Payments Database, the source which we obtained most of the undisclosed COIs. It is possible some COI were missed, and therefore interpretating of our results as a lower-bound estimate of the true extent of COI within the ophthalmology field is warranted.

CONCLUSION
We found that SRs conducted by 1 or more authors with a COI were more likely to draw favorable conclusions about the investigated intervention, compared with SRs with no conflicted authors. COIs are common, yet often undisclosed, among authors of ophthalmology SRs. We contend that reforming the current COI definitions and disclosure policies is necessary to increase the transparency in the ophthalmology literature.