Rethinking Peer Review Using the Swiss Cheese Model to Better Flag Problematic Manuscripts

Peer review represents a cornerstone of scientific and scholarly publishing. Despite many unanswered questions about the value of peer review (Tennant and Ross-Hellauer 2020), it is widely assumed that peer review improves the quality of published articles. In turn, reviewers accept peer review requests based on the interest and relevance of manuscripts, and as a service to their fields (Severin and Chataway 2021). It has been estimated that reviewers contributed over 100 million hours towards peer review in 2020, with an approximate collective value of over USD 2 billion (Aczel et al. 2021). Millions of peer reviewers participate in peer review every year (Aczel et al. 2021) through invitations from the many thousands of editors who support peer reviewed journals.

Peer review occurs through a stepwise process (Figure 1), where manuscripts are first considered by editors, with a subset of manuscripts progressing to external peer review. These stages involve different individuals with often differing expertise and hence capacities to assess research quality. Peer review has, however, paid less attention to detecting manuscripts with questionable integrity. This could reflect assumptions that relatively few manuscripts present integrity issues, and that peer reviewers may be poorly equipped to detect integrity issues within manuscripts (Stroebe et al. 2012).

The rise in manuscripts from paper mills (Behl 2021; Bricker-Anthony and Giangrande 2022; Cooper and Han 2021; Heck et al. 2021; Pinna et al. 2020; Van Noorden 2023) and the possible misapplication of large language models (LLM's) to generate and scale fast-churn manuscripts (Grimaldi and Ehrler 2023; Suchak et al. 2025) highlight the need to revisit assumptions about the capacity of peer review to detect unethical or low-value submissions. Suspected submissions from paper mills have now been described by many biomedical journals (Behl 2021; Bricker-Anthony and Giangrande 2022; Cooper and Han 2021; Heck et al. 2021; Pinna et al. 2020; Seifert 2021). Other journals have described receiving large numbers of repetitive manuscripts (Jin 2022), suggesting the possible misapplication of LLM's by paper mills or individual teams (Mainous III 2025; Munafò et al. 2024; Stewart 2025). Such changing patterns of submissions (Mainous III 2025) and, in some cases, publications (Stender et al. 2024; Suchak et al. 2025) question assumptions that unethical manuscript submissions are infrequent. As submissions from paper mills and the use of LLM's to produce derivative manuscripts become more frequent, the awareness of scaled and repetitive submissions is also expected to grow (Byrne and Stender 2025). This awareness can be leveraged through peer review processes that actively consider research and scholarly integrity in addition to quality (Abalkina et al. 2025).

There are further reasons to focus on peer review to counter submissions from paper mills. In contrast to genuine research, where the rate-limiting step towards publication is likely to be conducting the research, the rate-limiting step for paper mills is likely to be peer review (Byrne et al. 2022). It is, therefore, not surprising that paper mills attempt to manipulate this rate-limiting step (Byrne et al. 2022; Matusz et al. 2025), for example, by recruiting editors and peer reviewers to facilitate manuscript acceptance (Abalkina et al. 2025; Joelving and Retraction Watch 2024; Pinna et al. 2020), and by creating fake reviewer identities (Matusz et al. 2025). Given the critical and rate-limiting importance of peer review to paper mills, their manuscripts and submission tactics are likely to continually evolve to evade detection (Byrne et al. 2024), highlighting the need for peer review processes to keep pace (Abalkina et al. 2025).

The scale and changing nature of the paper mill problem highlights the value of rethinking peer review according to the Swiss Cheese Model of error and accident causation and prevention (Larouzee and Le Coze 2020; Wiegmann et al. 2022). The Swiss Cheese Model represents processes as different slices of cheese, where strengths are represented by the cheese, and weaknesses or failings are represented by holes in the cheese (Figure 1). The Swiss Cheese Model proposes that errors or accidents can occur when blind spots at different process stages are shared, and that errors and accidents can be prevented when processes have complementary areas of strength (Wiegmann et al. 2022) (Figure 1).

When applied to peer review, the Swiss Cheese Model allows each stage of the peer review process to be recognised as a new opportunity to detect problematic manuscripts (Figure 1). Ideally, most problematic submissions will be detected at the early stages of peer review, with fewer problematic submissions progressing to the later, more resource-intensive stages (Figure 1). Nonetheless, if problematic manuscript features can be communicated back to editors, this can tighten some ‘holes’ through which problematic manuscripts could otherwise slip. While no peer review system can successfully detect all submissions from bad actors (Cooke et al. 2024; Wykes and Parkinson 2023), continuous, real-time improvements could reduce the numbers of accepted manuscripts that, with the benefit of hindsight, should have been rejected.

Other authors have recently described the features of paper mill manuscripts that can be detected by editors and peer reviewers, and how the peer review system can better respond (Abalkina et al. 2025; Christopher 2021). We will largely not repeat previous advice but instead focus on opportunities that have not been previously discussed. In all cases, we will describe simple modifications of existing practices that could be easily implemented (Table 1). We will give some final considerations to the possible consequences of implementing these recommended changes at scale, and how some publisher and journal practices might need to adapt as a result.

Editors make final decisions as to whether manuscripts will be accepted or rejected; therefore, they play critical roles in peer review (Sever 2023; Tennant, and Ross-Hellauer 2020). Many journals have editorial assistants or in-house teams checking submissions and making decisions on initial desk rejections (Horbach and Halffman 2020). These early decisions determine whether and how manuscripts progress to the later stages of peer review, and how much time and other resources will be assigned to individual manuscripts. It is in both editors' and peer reviewers' interests that editorial review identifies low-value and potentially dishonest submissions as effectively as possible to reduce the resources that could otherwise be wasted through further consideration (Cooke et al. 2024; Sever 2023; Wykes and Parkinson 2023).

To detect problematic submissions at the editorial review stage, publishers and journals are increasingly investing in research integrity teams, manuscript screening tools and using known features of problematic submissions to highlight manuscripts for desk rejection (Abalkina et al. 2025; Alam and Wilson 2023). However, despite the best efforts of journals and publishers, not all problematic submissions will be detected through editorial review (Behl 2021; Bricker-Anthony and Giangrande 2022) (Figure 1). First, screening tools may not be available to all journals, particularly where individual journals need to pay for access. Second, most screening tools represent commercial products where the targeted features may not be disclosed, and practical understanding of the output could be limited. This could lead to uncertainties over the value of predictions and the decisions that editorial staff should take in response. Third, in-house teams and screening tools are unlikely to recognise all paper mill or mass-produced manuscript features, given our currently limited knowledge of the extent of the paper mill problem, and the predicted capacity of paper mill submissions to evolve in response to improved detection (Byrne et al. 2024). Finally, awareness of problematic submissions is likely to vary between individual editors and journals. For example, some editors might simply be unaware of paper mills or assume that their journal will not be targeted. Similarly, editors might not recognise features of problematic submissions, particularly if their journal has not been previously (or knowingly) targeted or receives few submissions.

If a manuscript is not desk-rejected, editors will invite experts to review the manuscript, with the aim of securing 2–3 reviewers as quickly as possible (Sever 2023). While the process of inviting reviewers might not always be viewed as a separate stage of peer review, many journals are likely to be spending more time and resources issuing peer reviewer invitations, as more invitations are declined (Chataway and Severin 2021; Fox et al. 2017; Sever 2023). Difficulties in identifying peer reviewers could be exacerbated through a growing reluctance to approach author-suggested reviewers, in case these individuals are linked to paper mills (Wittau and Seifert 2024).

Once a reviewer has accepted a peer review request, they will usually supply their review through the journal's editorial platform. This involves providing comments to authors, optional confidential comments to the editor, and answering a range of questions about the manuscript and themselves as reviewers (Dine et al. 2015).

Many revised manuscripts undergo further peer review, particularly if many revisions are requested (Sever 2023). Editorial and peer review of revised manuscripts therefore provide another opportunity to detect and flag integrity concerns that may have been missed earlier (Figure 1). For example, manuscript revision provides an opportunity for editors to request supporting raw data that can sometimes reveal new integrity concerns (Christopher 2021).

We recognise that taking a more adversarial approach to peer review could have unintended consequences. A proportion of integrity concerns could be unfounded; and so desk rejecting manuscripts based on integrity concerns could disadvantage some genuine research and scholarship. This is an important consideration, particularly given the higher stakes that peer review can represent for junior scholars (Meibauer et al. 2025).

High rates of desk rejections could also impose additional stress on editors that could adversely impact decision making (Bazerman and Sezer 2016; Wykes and Parkinson 2023). High desk rejection rates could be particularly challenging for smaller or new journals, where editors may be encouraged to send more submissions for peer review. Rapid desk rejections could also allow quick resubmissions of paper mill manuscripts to other journals. These factors highlight the importance of considered desk rejections, where the timeliness of editorial decisions may be less important (Wykes and Parkinson 2023).

Encouraging confidential comments to editors could also provide opportunities for reviewers to make vexatious claims to disadvantage competitors (Swanson 2023). While this represents a possibility, it is important to recognise that rival reviewers have other opportunities to disadvantage competitors through their comments to authors. Some journals allow authors to exclude reviewers from peer review to reduce the possibility of vexatious claims or delays. From our experience as both editors and peer reviewers, single reviewer descriptions of integrity concerns seem unlikely to determine editorial decisions. However, peer review processes that allow manuscripts to be (i) declined for peer review due to possible integrity concerns and (ii) receive peer review comments on similar issues could be less susceptible to vexatious claims.

Many journal editors have described rising numbers of problematic submissions that are increasingly overwhelming peer review resources and leading to rising numbers of low-value publications. The Swiss Cheese Model allows peer review to be reconsidered from an error detection and accident prevention perspective, where each peer review stage can be considered a new opportunity to detect problematic submissions. The changes that we have suggested could allow journal editors and peer reviewers to more effectively flag problematic submissions, producing a more resilient peer review system and contributing towards a more evidence-based understanding of peer review. More widespread use of best practices and a willingness to constantly update peer review processes to emerging threats (Abalkina et al. 2025; Netzer 2024) could allow better use of valuable peer review resources and reduce unethical publications from paper mills.

Jennifer A. Byrne: conceptualization, writing – original draft, writing – review and editing. Anna Abalkina: writing – review and editing, writing – original draft. Jana Christopher: writing – original draft, writing – review and editing. Marie F. Soulière: writing – original draft, writing – review and editing.

All opinions expressed in this commentary represent the views of the authors and do not represent the views of any listed employer or institution.

M.F.S. is employed by Frontiers Media SA. All opinions expressed in this commentary represent the views of the authors and do not represent the views of any listed employer or institution. The other authors declare no conflicts of interest.