Editorial peer review stands as the gateway to scientific publication. The process was established to ensure that research papers are vetted by independent experts before they are published, although it is recently being increasingly questioned due to beliefs that it is flawed.1 2 Despite efforts over the last 30 years to ‘make peer-review scientific’, its impact is still considered suboptimal.3
Peer reviewers, who are the pivotal actors in this process, are requested to write a review report evaluating the submitted manuscript. A peer-review report helps authors to improve the quality of their manuscripts, and it also helps editors make an informed decision about the outcome of the manuscript. However, evidence shows that these peer-review reports are often of poor quality.4 5
Tools for assessing the quality of peer-review reports have been proposed, of which we have conducted a systematic review and identified 24 tools: 23 scales and 1 checklist.6 However, none reported any definition of peer-review report quality, only one described the scale development, and 10 provided measures of reliability and validity. Further, the development and validation process resulted from a small consensus of people, and the concepts evaluated by these tools were quite heterogeneous.
In 2016, Bruce et al published a review evaluating the impact of interventions to improve the quality of the peer-review process.5 The authors showed that it is essential to clarify the outcomes (such as, the quality of peer-review reports), which should be used in randomised controlled trials to evaluate these interventions.
A validated tool is direly needed to clearly define the quality of a peer-review report in biomedical research. This tool could be used regularly by editors to evaluate the reviewers’ work, and also as an outcome when evaluating interventions to improve the peer-review process. In the present study, we report on the development of a new tool to assess peer-review reports in biomedical research.
We formed a steering committee of five members (CS, DH, AR, IB and JAG), whose expertise include clinical epidemiology, biostatistics, social science and editorial peer review. The steering committee agreed on how to define peer-review report quality; they agreed on the survey questionnaire based on the results of a previous systematic review6; they interpreted the results of the survey and they agreed on the final version of the tool.
The tool aims to assess the quality of peer-review reports in biomedical research. We defined the quality of a peer-review report as ‘the extent to which a peer-review report helps editors make a fair decision and authors improve the quality of the submitted manuscript’.
A systematic review allowed the identification of 24 tools, aimed at assessing the quality of peer-review reports.6 We extracted 132 items from such tools. After removing the redundant items, we obtained 17 items. We then eliminated two items and incorporated five new ones that met our definition of peer-review report quality, after piloting the survey questionnaire and discussing with the steering committee. Overall, 20 items were identified to assess peer-review report quality (table 1).
We conducted an online survey of editors and authors in order to: (1) determine if they endorse the proposed definition of peer-review report quality; (2) identify the most important items to include in the final tool and (3) identify any new items that should be included.
The questionnaire was constructed using the online survey software SurveyMonkey.7 It was structured into four main parts and included both open and multiple-choice questions. First, the participants were asked to agree (‘yes/no/partially’) on the definition we provided for peer-review report quality. They were also invited to add any comments or ideas on how to improve the definition. Second, they were asked to rate the importance of the 20 items for assessing the quality of peer-review reports we identified. Their responses were based on a 1–5 Likert scale (1 being not important and 5 very important). In particular, we asked the participants if the item should be included in a tool for assessing the quality of peer-review reports. Moreover, they were invited to comment on the importance and wording of each item. In order to eliminate the question order effect, the items appeared in random order for each respondent. Third, the participants were invited to suggest any additional items missing that they considered important for assessing the quality of peer-review reports. Finally, the questionnaire included nine demographic questions related to sex, age, education level, job title, referring institution and job experience as biomedical editor and/or author. We developed two versions of the questionnaire because biomedical editors and authors were recruited differently, despite the fact that some of them could play both roles (see online supplementary file 1). The two versions were structured in the same way; they only differed in some questions related to the demographic characteristics. The questionnaire was piloted among six experienced scientific editors and authors, followed by a subsequent revision based on their feedback.
We targeted biomedical editors and authors using a purposive sampling approach to recruit a heterogeneous sample of information-rich cases.8
By means of standardised email, we invited two groups of editors to participate in the survey: 586 biomedical editors from 43 journals in the BMJ Publishing group; and 478 editors from 235 journals identified in a previous cross-sectional bibliometric study9 (see online supplementary file 2). The survey was also distributed to 27 editors from 48 journals in BMC (part of Springer Nature), using internal email and to members of the European Association of Science Editors (EASE) through their newsletter. In the invitation email and newsletter, the editors were encouraged to forward the survey to colleagues who might be interested in issues related to peer-review. This recruitment strategy, known as snowballing, allowed us to identify ‘information-rich key informants’ among biomedical editors.8 On the first page of the survey, participants were informed that the collected data would be anonymous, and they were further asked if they would agree to share their deidentified data in an open access repository. Two reminder emails were sent to non-respondents. Finally, the survey was promoted on Twitter and on the EASE blog10 and Methods in Research on Research11 websites.
Searching the top 30-biomedical journals with the highest impact factors, we identified 4396 corresponding authors of articles that reported original research and which were published in Medline between 1 February and 31 October 2018 (see online supplementary file 3). We used the R package easy PubMed to extract the email contacts.12 The corresponding authors received a standardised email that explained the purpose of the study and included a link to the survey (see online supplementary file 2). The first page of the survey informed participants that the data were collected anonymously and also asked if they would agree to share their deidentified data in an open-access repository. Two reminder emails were sent to non-respondents.
We did not use a snowballing strategy to recruit authors. However, since the survey directed to biomedical editors was promoted on Twitter by different users who sometimes did not provide thorough instructions, we included in the first page of the survey, also the link to the questionnaire addressed to authors. This was done so that a researcher, who was not an editor and mistakenly opened the link to the survey questionnaire, was still able to participate to the study as biomedical author.
We described the demographic data in terms of frequencies and percentages. The importance of the 20 items to assess peer-review report quality is described in means and proportions of editors or authors who rated the importance of the items from 1 to 5. The items were also sorted according to the mean raking of all participants and either editors or authors. We also calculated Pearson correlations among items. The calculations and graphical representations were all obtained using the statistical software R 220.127.116.11
We conducted a principal component analysis (PCA) to examine item redundancy among the 20 items to assess peer-review report included in the survey. PCA is a multivariate statistical technique used to reduce the number of variables in a dataset to a smaller number of dimensions.14 The new dimensions (or principal components (PC)) are mutually independent and are determined by choosing the directions that explain the most variation in the data. The first PC (PC1) accounts for the largest possible variance in the data, and each succeeding PC accounts for decreasing amounts of the remaining. This exploratory analysis helps reveal simple underlying structures in complex datasets. We performed PCA using the R package FactoMineR.15
We used a general inductive approach for qualitative data analysis. In particular, we followed the five steps of inductive analysis proposed by David R. Thomas: (1) Preparation of raw data files; (2) Close reading of text; (3) Creation of codes; (4) Overlapping coding and uncoded text and (5) Continuing revision and refinement of themes system.16 In the third phase, two investigators (CS and DB) created independently the initial codes from the responses of the first 100 participants for each open-ended question. In order to ensure consistency and credibility, the initial codes were discussed with a third investigator (DH) and a codebook was developed and was used for analysing the remaining responses. In case new codes were successively created from the remaining responses, the emerging codes were added to the codebook and applied to entire dataset. Two investigators (CS and DH) reviewed and refined the codebook and further clustered the codes into major themes. We used the software NVivo V.12 for data management and analysis.17
The steering committee reviewed all items and, ultimately, drafted and refined the final version of the tool. Based on the participants’ qualitative and quantitative answers, redundant items were combined, existing items were modified and/or expanded on, and new items proposed by survey participants were added.
Patients and members of the public were not involved in the study.
Between 7 November 2018 and 4 February 2019, 198 biomedical editors and 248 authors completed the survey. Of the 1134-biomedical editors and 3633 corresponding authors invited via email, 89 (7.8%) and 238 (6.5%) completed the survey, respectively. In addition, 109 editors and 10 authors completed the survey using the web link.
Participants were mainly male (263/399, 65.9%) with a PhD degree (225/399, 56.4%), and their ages were equally distributed across ranges (mean=50.3, SD=13). They were mainly located in Europe (219/389, 56.3%) and North America (118/389, 30.3%). More than half of the editors had work experience of more than 5 years (91/165, 55.2%), while over one-third of the authors had work experience of more than 20 years (84/224, 37.5%) (see table 2). Editors were mainly associate editors (63/165, 38.2%) and editors in chief (50/165, 30.3%), primarily involved in making decisions on the submitted manuscripts (144/165, 87.3%). Most of them worked in specialty journals (126/165, 76.4%) and they were used to contribute as authors in scientific papers (141/165, 85.5%). The corresponding authors were mainly professors (63/224, 28.1%), but also PhD students, postdocs or lecturers (49/224, 21.9%) or researchers (47/224, 21%). The majority of them worked in public universities (134/224, 59.8%) and they were not employed as editor (161/224, 71.9%) in biomedical journals. Among those who also work as biomedical editors (63/224, 28.1%), 88.9% of them are involved in making decision on the manuscript (online supplementary file 4).
Overall 84% (362/431) participants, precisely 85% (160/188) editors and 83% (202/243) authors, agreed on the definition of peer-review report quality that we provided in the survey. The definition was slightly modified to take into account participants comments (online supplementary file 5). The quality of a peer-review report is now defined as ‘the extent to which a peer-review report helps, first, editors make an informed and unbiased decision about the manuscripts’ outcome and, second, authors improve the quality of the submitted manuscript’.
We created a web application that is publicly available at https://www-eio.upc.edu/redir/ReportQuality. Through the application, the readers can easily access and explore the quantitative results of the survey.
The items were generally highly rated, with a mean score ranging from 3.38 (SD=1.13) to 4.60 (SD=0.69). All the items were scored 4 or 5 by >50% of the participants (see web application). The three items rated as the most important were: (1) Knowledgeability; (2) Methodological quality and (3) Fairness. The three least important items were: (1) Originality, (2) Presentation and organisation and (3) Adherence to RG.
A peer-review report aims to help authors improve their submitted manuscripts and assist editors in taking editorial decisions. Due to this dual objective, we compared editors’ and authors’ mean scores in order to investigate whether any difference is found in their perceptions regarding the importance of the 20 items that assess peer-review report quality. We found little discrepancy in the mean scores between biomedical editors and authors, with only two items indicating any difference: (1) Timeliness and 2) Detail/Thoroughness. The Timeliness of the peer-review report was considered more important to authors than to editors (respectively, in the 12th and 16th rank positions). Meanwhile, editors rated the Detail/Thoroughness of the reviewer’s comments higher than did authors (respectively, in the 11th and 16th rank positions).
Overall, we found relatively weak positive correlations among items. The largest positive correlations were found between Relevance and Originality, and between Fairness and Objectivity (r=0.55 and 0.43, respectively).
The first PC1 accounted for 22.1% of data variability. The next two dimensions (PC2 and PC3) accounted for 38.5% of the cumulative variability and contributed gradually, that is, they increased at only small increments. PC1 was positively correlated to all items (or variables), and it showed correlations higher than 0.4—which is the figure commonly used as a threshold reference for factor loadings—for 16 out of 20 items (see web application). These results illustrate that the data variance was not concentrated in a few components but distributed across all of them; hence, reducing the number of items is not recommended, since this would imply an important loss of data information.
The study of the supplementary variables did not reveal any differences between authors and editors in terms of items rating. However, we found that female participants above the age of 55 years old generally provided higher rating for the items, compared with younger male participants.
Out of 446 survey participants, 267 (59.9 %) made at least one comment on the importance and/or wording of the items. Based on the initial coding of the comments, we were able to identify eight general themes that they addressed: Peer reviewer; Wording; Importance; Dependency; Responsibility; Item; Structure and content; and Improvement. Table 3 reports the eight themes together with their definition and the most frequent codes (n>5), with example quotes. The entire codebook is found in online supplementary file 5.
The steering committee met on the 19 July 2019 to discuss the selection of items to include in the final version of the tool. Their decisions were based on the participants’ quantitative and qualitative answers. The flow of the items is summarised in figure 1.
The items Relevance and Originality were merged into a new item named Contribution (of the study). This decision was based on the high positive correlation found between the two items (0.55) and on the participants’ opinions. Furthermore, participants suggested in their comments that the item Relevance was ‘highly subjective’, because ‘each reviewer’s decision on relevance reflects what is relevant to them, which may not reflect relevance to the journal’. They also believed that the Originality of a study is not always an important aspect for comments in a peer-review report, because some manuscripts ‘are trying to duplicate findings from previous studies’. They, therefore, suggested reformulating the two items by asking the reviewer what the study ‘adds to our knowledge’.
The steering committee decided to include the item Interpretation of results as a domain of the tool instead of a single item, changing the name into Interpretation and discussion of the study results. This decision resulted from the addition of two new items (Study conclusions and Study limitations), based on the suggestions of survey participants. The domain Interpretation and discussion of the study results now encompasses three items: (1) Study conclusions; (2) Study limitations and (3) Applicability and generalisability.
Overall, survey participants believed that the items Strengths and weaknesses (general) and Strengths and weaknesses (methods) were ‘confusing to separate’. Additionally, the steering committee agreed that Strengths and weaknesses (methods) and Methodological quality were also redundant; thus, it was ultimately decided to merge the three items into a new item named Study methods.
The items Objectivity and Fairness were merged because of both the moderate correlation between them (0.43) and the participants’ opinions. Participants suggested that the total objectivity of the reviewer’s comments is not possible because ‘all decisions contain some personal biases and subjectivity’ and they also believed that the term fairness was ‘very subjective’ and difficult to define. Additionally, the steering committee agreed to also combine these two items into Supported by evidence. The committee finally decided to merge all three items into Objectivity, and this was defined as ‘comments provided in a peer-review report should be as objective as possible and, if considered appropriate, include references to support the reviewer’s statements’.
The steering committee agreed to merge Structure of reviewer’s comments and Clarity, because participants considered both important for making the peer-review report easy ‘to read for both editors and authors’. Moreover, participants suggested that the Detail/Thoroughness of a peer-review report was mostly associated with the quality of a manuscript, because in certain occasions a study can be so poorly conducted that ‘a reviewer can highlight one or two major methodological flaws’ without conducting a detailed review. They, therefore, believed that a detailed report is not ‘always necessary’ and instead preferred a succinct report that ‘cuts straight to the critical points’. Taking into account the participants’ opinions, the steering committee finally decided to include a single item named Clarity, which is defined as ‘a peer-review report should be clear, succinct and well organised in order to be understood correctly by editors and authors’.
The items Tone and Constructiveness were merged into Constructiveness, which is defined as ‘a peer-review report should contain constructive and polite comments that allow the authors to improve the quality of their work’. This decision was based on the participants’ opinions that ‘the comments should be polite and constructive’.
The item Adherence to RG and the new item Reproducibility suggested by survey participants were merged into Reporting based on the steering committee decision. The item Reporting was defined as ‘the reviewer should comment if the reporting of the study is clear, complete and transparent enough for facilitating its reproducibility by verifying the adherence of the manuscript to the corresponding reporting guideline’.
The items Timeliness and Knowledgeability were not included in the final version of the tool. Survey participants suggested that Timeliness was not ‘directly tied to review quality’ because ‘some of the best reviews come in past the deadline’. Furthermore, the steering committee agreed that the item Knowledgeability was generally difficult to assess, because it implied that anyone using the tool would have enough competence to evaluate the reviewer’s knowledge and expertise. Five new items suggested by survey participants (Data availability, Study protocol, Study conclusions, Study limitations and Relevant literature) were finally included in the tool.
The Assessment of Review reports with a Checklist Available to eDItors and Authors (ARCADIA) tool was finally developed. The tool is a checklist that includes five domains and 14 items (table 4). Brief explanations of the items included in the five domains are provided in online supplementary file 7.
This study resulted in a checklist of items to assess the quality of peer-review reports in biomedical research. The checklist constitutes the first tool that has been systematically developed to assess the quality of peer-review reports.
The checklist is simple, applicable to any biomedical field, and consists of five domains covering 14 items, each of which is phrased as a question. Each item should be ticked as ‘yes’ or ‘no’. An item could be also checked ‘NA’ if it is not covered in the study (e.g., there are no data or other materials attached to the manuscript) and/or the peer reviewer is not qualified to comment on that specific aspect (e.g., statistical methods). The ARCADIA tool has several strengths. It is the first tool ever developed based on an exhaustive review of the literature6 and on empirical data from a large sample of both biomedical editors and authors. Further, it is the only tool that clearly defines the quality of peer-review reports, as its definition was based on the perspectives of 446 authors and editors.
To develop the tool, we recruited a large sample of biomedical editors and authors with varying experience and backgrounds. We found the percentage of female participants who took part in the survey to be quite low (129/399, 32.3%). This is in line with evidence showing that gender equity in academic medicine careers remains far behind.18 Moreover, we recruited corresponding authors (who are usually first authors) from the top 30 biomedical journals. Evidence also shows that women are under-represented as first authors among biomedical journals with high impact factors.19
Overall, we did not find any differences between authors and editors in terms of item rating by conducting PCA. Only two items, Timeliness and Detail/thoroughness, presented a difference according to the separate mean score rankings of authors and editors. Timeliness was considered more important for authors and this could be justified by the fact that authors are usually more interested in receiving decisions about their manuscript as soon as possible. Whereas, editors rated detail/thoroughness as more important to them, given thorough and detailed peer-review reports help them make a better editorial decision on any given manuscript.
The present study also has some limitations. The survey questionnaire included some open-ended questions, which allowed participants to voluntarily express their opinions. However, we were not able to inquire further to clarify and verify some information provided by the study’s participants. Therefore, the interpretation of some information could be affected by the perception of the three investigators who conducted the qualitative analysis. Additionally, since participants could comment voluntarily on the importance and wording of each item, the number of comments among items differed greatly. Furthermore, the majority of editors and authors were from Europe and North America, which may limit the generalisability of the results. This result may be due to the recruitment strategy we used, especially to identify biomedical editors. Although we also used a snowballing strategy, we mainly contacted editors through European biomedical journals. Finally, the present study reports on the first version of the ARCADIA tool, which has not yet been validated.
The tool is a general checklist available to all biomedical editors and authors. It could be regularly used by editors to evaluate the reviewers’ work, and it can also be used as an outcome when evaluating interventions in order to improve the peer-review process.