Amending published articles: time to rethink retractions and corrections?

A fundamental underlying problem

To date, we have observed a bewildering variety of notices on articles posted after publication. We have captured nine most commonly used and point to examples in Table 1. Furthermore, many of these are implemented differently by different publishers. No standard taxonomy of updates exists for publishers to adopt. This leads to inconsistencies from journal to journal and potential confusion for the reader.

Moreover, a lack of willingness to engage in proper post-publication correction and amendment of the literature is further exacerbated when any type of post-publication ‘updates’ are misconstrued as punishments or admissions of guilt. This is particularly the case with retraction, a term which many feel has come to be loaded with blame and recrimination. It is fair to say that no one who has been involved with the retraction of an article – either as an editor, publisher, reviewer or author - has ever walked away from the process feeling wholeheartedly good about the experience. This is the case even if a retraction is done for the best of reasons – a genuine, no fault mistake^11,12. As a result, we believe a ‘retraction’ will never be fully embraced as a positive outcome by researchers.

There is a fundamental misconception that retractions are ‘bad’ without pausing to ask why the retraction took place. A fourth principle is, therefore, that we need neutral terminology for the method of correcting published work that implies no fault on any party. In order to provide this, it is important to distinguish the correction of the published record from any investigation or description of misconduct that has occurred. If misconduct or fraud has occurred, this should be reported on, but such reporting should be considered as distinct from the process of correcting the literature. Such a separation is especially important as those who are likely to be responsible for an investigation into misconduct will be different from those issuing any correction of the published record, and the two may need to happen on quite different time frames. During this time, we feel it is important to alert readers to the possible issues with the published work, and to update the literature without awaiting the final outcome of a lengthy investigation¹³. Once any investigation is complete, however, its outcome should be recorded on any it has considered.

Although corrections may not be as universally disliked as retractions, once an article has more than one or two corrections, in the current publishing system it is difficult to track what has happened. Tracking corrections is even more challenging when they are documented outside of the publication itself. For such ‘external corrections’ tracking is arguably even more crucial, as articles (as well as references to them) increasingly propagate across the internet and often do not link back to one version of record. However, developments such as Crossmark¹⁴ are beginning to address this issue. Furthermore, there are no universal guidelines for corrections, and editors and publishers often act on a case-by-case basis. Many editors and publishers struggle with the need for a correction notice for a very minor amendment to an article (such as a typographical error that has no effect on meaning), while many readers feel that all amendments post-publication need a clear audit trail. Our fifth key principle is that all changes made post-publication should be traceable and all versions of an article should be preserved. The circumstances in which an article should be removed (for ethical, legal or safety reasons) are extremely rare, and even in these cases a metadata record should remain, alongside an explanation.

Now is the time for change

There are better approaches for amending scholarly communications post-publication. Examples from newspapers and blogs [e.g. 15] take a simple approach to corrections which are effective, speedy and user-friendly. However, these types of outlet rarely require any cross-referencing to external articles and hence the process is much simpler than for academic research articles.

While we reaffirm the importance of preserving the integrity of the published literature, we equally strongly affirm that research outputs are now dynamic objects online. Our sixth key principle is that the idea of the journal article as a monolithic object that will stand for all time unless formally retracted has gone. Rather we are seeing calls for articles to be viewed as organic publications or “living articles”¹⁶. If this idea is to be accepted, it is critical to ensure that updates to the scholarly record of a publication are appropriately made and that they properly link the latest update to the original record. Readers need a complete history of changes to an article. We now have the technical tools at hand to do this, and indeed a number of publishers now publish successive versions. These versions can be elegantly handled for readers online by making it clear which version is being displayed and by employing links to help readers navigate between versions (e.g. F1000Research¹⁷). Crossmark also provides a vital insurance mechanism, as it displays the publication history consistently across publishers¹⁸. Our seventh key principle is that both human and machine readers should be able to trace the full history of an article with ease, and should by default be taken to the most recent version of any article.

We now have the technical tools to ensure there is clear notation of version history in the citation record. The digital object identifier (DOI) is central to this as a unique alphanumeric string that identifies the content and provides a persistent link to the location of each resource on the Internet. It is an actionable identifier as it resolves to (i.e., takes the user to) the corresponding resource online. It is also descriptive as it binds the DOI to specific metadata about the digital object.

To support versioning, publishers could assign a new DOI (or, more logically a DOI with a suffix), to each version, and link to the previous version in the metadata record. Crossref ensures that all versions are linked through the relationship metadata included in the record. Once deposited with Crossref¹⁹, all versions could be threaded together²⁰ through their DOIs and made available to systems across the research ecosystem. This practice provides specificity and precision to the citation record that has not been possible before. Researchers can thus cite a specific version, rather than, unintentionally, the original one. Essentially, we can now, technologically, think beyond the article of record to a number of versions, whether they are preprints or postprints or institutional copies. The primary challenge, however, is facilitating publisher (and perhaps institutional) adoption and implementation of practice to deposit and update metadata records as changes occur.

If a fundamental change in how we amend published articles is to be successful, we need both the technology to make it happen (as outlined above) and the will and support from the community to embrace the change.

The amendment model

In order to facilitate a more reliable and useful scholarly literature, we propose the following components of handling changes to published articles. The changes can be implemented by the same journal (or other publishing venue) that published the original research and be linked to the original article. We propose removing the terms correction and retraction altogether and instead using the term “amendment” to describe all forms of post-publication change to an article.

The term “amendment” carries a neutral tone and is generic enough to apply to a wide array of cases, including the smallest instances such as a typographical error all the way to wholesale withdrawal of an article. By employing a uniform term, we hope to remove any associated stigma in the context of scholarly literature. When readers encounter each amendment, they can read the notice for details on each change, and can judge the article and its revisions on their own terms.

We considered alternative names such as “update” but we felt they imply progress or addition. In particular, we strongly feel that retaining the word “retraction” even for the most egregious instances of scientific malpractice would further perpetuate the problem of stigma and is thus not desirable.

We propose an amendment model made up of three different types: minor, major, and complete (Figure 3). These are distinguished by the scale of change entailed. Minor amendments cover small changes such as typographical errors, or other minor amendments to the content or metadata that has no effect on the substance of the article. Issues currently worded as corrections belong as Major amendments, but might also include clarifications and addenda (which are not currently easy to make under most current publishing workflows). Amendments of this type would make changes to one or two small parts of the article but not its whole message. Examples are changes in authorship, correction of one figure or method, or additional data or discussion. Lastly, Complete amendments covers situations where the article as a whole is considered unreliable in its current form. There may be elements that remain ‘correct’ but large proportions are not. Instead of “retract and replace” as currently practiced by some publishers, we would recommend “retract and republish” with a new DOI that resolves on the newer version and makes plain the chain of events. In cases where authors and/or journal may wish to dissociate themselves from the original article completely, this can be noted with a complete description in the associated narrative and no attempt to insert new text or other content. We would emphasise the importance of using the single neutral term ‘complete amendment’ to cover all types of retraction including those for honest errors retractions and misconduct. The narrative should include explanation of why the amendment is being made, and can report on any misconduct allegations once they are known and proven. However, an amendment could nevertheless be made initially without awaiting the outcome of any misconduct allegation or investigation.

Figure 3. Three amendment types.

The amendment notice

The notice (for all amendment types) is comprised of a declaration followed by key details. The declaration is posted at the forefront of the document stating that: “The authors and/or the journal wish to make the following amendment to the published article [article full reference].” We envisage this declaration being authored most often by the authors of the original article, in consultation with the journal or publisher, but the format allows for the possibility that a journal or publisher can amend an article without an author’s consent if that proves necessary. In either case the source of the amendment will be clearly visible to readers.

Every amendment notice would then include the following:

The process of the amendment within the publication lifecycle is straightforward and consistent for all types of amendment. Whether the incident at hand merits a minor, major, or complete amendment, the publisher can issue the notice, assign a new DOI to it, register it with Crossref and link to the target publication. Moreover, the same process is also consistent and streamlined to apply at a higher level to include all the various versions of the publication from the original publication to the posting of amendments (Figure 4).

Figure 4. The amendment process.

Publishers register all amendments and versions of the paper as they currently do now with Crossref. Each amendment is assigned a DOI, and its metadata should contain a link to its associated article. This links the amendment notice and specific version of the paper in both directions. When a subsequent version of a paper is published, the publisher registers it with Crossref with a new DOI so that it can contain its own, independent set of metadata. The metadata should contain a link to the previous version, thereby creating a sequential chain across multiple updates.

This version of scholarly communications supports the dynamic nature of the research process, as researchers continue to refine or extend the work. They can publish updates along the way, sharing their latest findings, analysis, and conclusions. For each version of an article an amendment can be issued. In each case, the publisher assigns a new DOI and deposits the metadata with Crossref so that researchers can cite with clarity and specificity.

Amendment display and linking

While the proposed amendment model simplifies the process by which published results are shared and updated, it also increases the potential number of components that might be published from a single set of research results. As such, linking amendments to their associated articles and across individual versions needs to be carefully implemented online so readers can easily navigate between versions.

Since every publisher employs their own specific design approaches to content delivery, we recommend the following linking and display strategies to ensure that amendment display fully supports editorial intent (Figure 5). We also illustrate the proposed amendment model, to a generalised case in Supplementary file 1 that contains elements of a number of COPE cases²¹.

Figure 5. Amendments display and linking.

Linking. Link from each amendment to the respective article it amends and vice versa, so that the reader can easily navigate back and forth between the notice and the research itself. Also link general amendments to their respective article version to ensure that the amendment notice links back to the specific article version to which it amends. And in the event of a correction to a figure legend, the link should direct users to the latest version of the article, complete with the correct figure legend. Readers can then go back and look at it with the wrong legend as they wish (i.e. be transparent about the change). This is akin to the journalistic model and is much cleaner.

Article versioning. Each article version has its own DOI and URL, which persists even with the publication of subsequent versions. Where the reader is on an outdated version, clearly communicate its date on the article and provide a prominent link to the most up-to-date version. Provide a clean and simple way for readers to navigate to and from each version. In cases where the publisher is linking to the article in general (e.g. from a journal home page, etc.), directing users to the latest version is recommended.

DOI handling. Although any two DOIs can be linked via Crossref to explain the relationship between versions of an article, or between an article and an amendment, we suggest that the relationship between these would be more obvious to human readers if the original article’s DOI were given a suffix (e.g. https://doi.org/10.1136/bmj.j1072.1; https://doi.org/10.1136/bmj.j1072.2, etc). Furthermore, The DOI should direct users directly to the specific document at hand, which corresponds to the precise version at hand. (This goes hand in hand with the recommendation to assign a new DOI to each version.)

Amendment metadata and propagation

All amendments must be evident to readers and machine harvesters of the literature. They must be inextricably and permanently linked to the original article and also propagated to systems that index articles. Some instances of complete amendments would mean publication of a new article, with a new DOI linked to the original article.

For publication updates to reach all the places online where articles are read, indexed, shared, discussed, recommended, etc. publishers need to make the metadata available in a central store where the data is freely available for systems and applications to consume. Crossref currently provides this facility and their metadata framework fully supports full disclosure of amendments. Once metadata updates are available from publishers, systems need to apply the latest information wherever the publications affected may appear online. All metadata are openly licensed for reuse, propagated through a variety of interfaces and formats via Crossref APIs²².

Anyone can search the published literature (humans and machines) to find the latest updates for any publication regardless of origin in the Crossref corpus (85+ million publications at time of writing). Publishers can flag this not only in their own content (online and PDF versions) via Crossmark, but also in the references of papers they publish. They can propagate these notices through other delivery channels offered such as email alerts, RSS feeds, recommendations, and so on. Non-publisher platforms such as indexers, reference managers, recommendation systems, social bookmarking tools, researcher profile systems, and others, can apply the update information and bibliographic metadata to the content they display as well. This information is also potentially useful for research information systems used by funders and research institutions, which also track scholarly outputs.