Ethics is a key component of the way humans interact with each other and with their environments, including in research contexts. Research ethics provides a framework and guidance for making and evaluating decisions touching upon intellectual, social, legal, practical, cross-cultural and other dimensions of research and the context in which it is situated (e.g. Bonde et al. 2015).

In some research fields - particularly those involving human subjects, animal experimentation, biodiversity or cultural heritage - the formalization of ethical norms and expectations has many decades of history (e.g. see Nelson 1967). This has led to detailed policies and guidelines that provide a framework for handling ethical issues and assisting compliance with applicable regulations (e.g. Yip et al. 2016, Childress and Thomas 2018). However, the norms may not be sufficiently standardized (e.g. Satalkar and Shaw 2013) in that they might lack clear practical implications like consistent incorporation into applicable workflows and cultural contexts, or simply uptake or proper communication (e.g. Murphy et al. 2015, Chiumento et al. 2020).

As formalization progresses, it tends to raise attention to ethical matters related to individual steps of research workflows, ranging from requesting ethical approval to documenting informed consent and providing ethics statements in funding applications or publications (e.g. Borovecki et al. 2018) .

Much of the process behind ethical review of research remains hidden (e.g. Vardigans et al. 2019) - for instance, it is rare to find public documentation of ethical approval (for an example, see Rothschild (2021)). This hampers the establishment of common standards and makes it difficult to teach or otherwise share practical aspects of research ethics across institutions, let alone jurisdictions. Furthermore, there is no mechanism through which ethics information can be looked up - let alone in a standardized fashion - for a given set of parameters, e.g. approval numbers (cf. Vardigans et al. 2019).

As illustrated in Fig. 1, ethics statements might contain information on a number of related matters (see JATS4R 2020 for best practice recommendations).

Figure 1.  

Ethics Statement from Cui et al. (2021) with putative markup of some key elements. Colors indicate the legal basis (pink), some boilerplate language pertaining to ethical review, approval and permissions (purple), oversight body (yellow) and approval number (green) as well as the aspect of the research that triggered the need for ethical oversight (grey).

These frequently include

• the legal or policy basis for handling these issues on an international level (e.g. the Declaration of Helsinki) and/ or within a given jurisdiction or institution;

• the procedures followed to conform with these legal requirements, along with information about the role of key stakeholders in the process (e.g. approval by an ethics committee, or informed consent by donors and participants, or protocols for anonymization, or (parts of) organizations where the research was performed);

• the aspects of the research - if any - that pose ethical issues (e.g. acquisition of personally identifiable information, or animal experimentation or involvement of minors or prisoners).

This kind of information may assist others in engaging with the research that was performed, with the underlying methodology or the resulting data, with research projects of a similar nature or with education about matters related to said research.

While the majority of ethics statements refer directly to the research described in the respective publication, some such statements - particularly in certain types of reviews - refer to ethical aspects of cited publications, often summarizing the information for several of them using more generic phrases than in individual-article ethics statements. An example is given in Fig. 2 .

Figure 2.  

Ethics statement from Yang et al. (2022), a meta-analysis, with putative markup of some key elements. Color code as above. The language is more generic overall than in Fig. 1.

To ensure that ethics statements are present in publications when required by applicable policy or legislation, it is important that ethics-related information is available in a structured format to both humans and machine. This aim is in line with the FAIR principles (Wilkinson et al. 2016), whose application to ethical contexts (Mietchen et al. 2019) would imply that ethics-related information is

• F indable by everyone involved in the publishing process - authors and their co-authors as well as editors, reviewers, publishers and readers, along with any tooling that assists them in matching features of the reported research to relevant policy elements;

• A ccessible to the above stakeholders and their tool chains;

• I nteroperable across studies, institutions, journals, funders and others involved in research ethics workflows;

• R eusable in another context (e.g. a different clinical, geographic or demographic focus).

At present, FAIR information about ethics is an exception rather than a rule, and we argue that this should change if ethical aspects of research are to receive proper attention.

Once the ethics statements are present where they should be, another set of considerations revolves around standardization of these statements: are all necessary pieces of information present, and are they expressed in a way that allows them to be compared, aggregated, assessed for compliance with applicable policy or otherwise used across studies?

Here, several factors come into play, for instance

• Policy elements - what information is required by what part of which policy that is applicable to what aspect of the respective research;

• Checklists with standardized “boilerplate” language for each policy element;

• Machine actionability of these policy elements and their corresponding textual representations in the checklists.

In order to assist in the standardization of research ethics and associated documentation, we propose to do the following:

1. mine ethics statements from full-text articles using dictionaries (cf. Fig. 3) of seed terms and phrases as starting points and Europe PMC (EPMC) as an example repository;

2. extract associated entities (e.g. subject areas, policies, authorities or research facilities) and vocabulary (e.g. terms and phrases related to handling informed consent or incidental findings);

3. assess the degree to which the language or other aspects of these statements - e.g. their location within a publication - are already standardized;

4. reconcile the extracted entities and vocabulary terms with Wikidata items and lexemes;

5. prototype and facilitate the creation of open infrastructure and automated workflows that allow to look up and query information about the research ethics landscape in general as well as ethics approvals in particular, along with the corresponding processes, standards, entities and vocabulary.

Figure 3.  

An example dictionary for text mining, containing various seed terms in a structured format that can be easily expanded. Each entry consists of three parts:

• (left): a string found or to be found in the mined texts (this part is mandatory)
• (center): a human-readable name for the string (this is optional)
• (right): an identifier for the string (still optional, but highly recommended)

Below, we will outline some use cases and practical steps towards implementing these ideas.

Ethics statements contain information about ethical aspects of the research reported in the respective manuscript. Having straightforward access to such information may assist readers in engaging with said research or with research projects of a similar nature. Possible scenarios here include researchers wanting to pool their own data with that of the reported study, or wishing to repeat the study under slightly different conditions (e.g. involving a different demographic, location, time of the year or medical procedure). Other scenarios include patients or members of their social circles trying to find out about clinical trials to potentially enrol in, funders or institutions that wish to monitor compliance with their policies (e.g. as per Rasberry and Mietchen 2021), or students or instructors engaged in education about research ethics.

If the relevant information in the ethics statements were available in a standardized fashion, this would allow for it to assist discovery in such scenarios. For instance, the terms used there or the relationships between them could be reused for parameterizing searches or for filtering search results. To achieve such standardization, communal language and ontologies or other forms of structured terminologies need to be created, and the process of creating them in turn assists in forming, strengthening or otherwise engaging such communities.

To demonstrate the feasibility of implementing the core ideas presented here, this section provides some methodological background, focusing on workflows that we used for prototyping.

Databases

Software pipeline

Software for accessing Europe PMC and similar repositories exists in several programming languages. We chose here to develop a Python-based pipeline that builds on a software suite originally implemented in Java a few years back and currently being developed as a tool called docanalysis (Hegde et al. 2022) to mine the literature for specific content like ethics statements. Fig. 4 provides an overview of the pipeline.

Figure 4.  

Ethics Statements mining pipeline. Works identified through a search query are being retrieved in full text, the text is then searched for key terms from the ethics dictionary to identify ethics-related article sections, which are then partitioned into sentences that are parsed to try to identify named entities. The results of the mining can be compared to entities and terms known from Wikidata and/ or the dictionaries, which can be continuously improved in an iterative process that can lead to a controlled vocabulary and eventually an ontology for ethics statements, ethics committees and related concepts.

We will discuss this process on the basis of the example use case of extracting information about ethics committees. However, the approach can be generalized to extracting other information, be that related to ethics - e.g. approval numbers, consent types, applicable policies and guidelines - or beyond, e.g. data availability (cf. Colavizza et al. 2020) or conflict of interest statements.

Creating iterative feedback loops between the mining, curation and annotation of ethics statements

An ontology of ethics committees and Institutional Review Boards (IRBs) can be created via Wikidata and used via the Wikidata SPARQL service. This ever-updating resource can then be used to aggregate and visualize ethics committee information extracted from the wider scientific literature. For instance, one could ask questions like which ethics committees have approved a particular study, or studies on particular subjects, involving specific demographics, using particular interventions or funding sources.

Large search engines are usually optimised for terms and synonyms, not higher levels of concepts like “ethics”, and they often rely largely or even solely on metadata, which might well contain no information about the ethics process. In order to find statements about ethical aspects of a publication, it is hence necessary to analyze its full text.

In subsequent rounds of mining, information from Wikidata can be used to finetune the entity recognition, e.g. by providing terms to be included in the dictionaries used for mining, or by providing context for entity disambiguation. For instance, geoinformation can be used to distinguish between Calvin University in South Korea and Calvin University in the United States. Further synonyms can frequently be resolved in a straightforward fashion: “X University” often maps to “University of X”, though for a small group of X (Wikidata knows 7 examples), both might exist as separate entities, either in close proximity (as is the case for Hyogo or Shizuoka), at different places within the same country (e.g. Rochester, Jinan, Miami), in neighbouring countries (Ottawa) or continents apart (York).

For common words, we may need stemming (“approved” => “approv~”) or more generally lexemes (“X is grateful” or “we are grateful”) => “X <be> grateful”. Modern NLP tools can now identify such phrases from their context with high confidence. Wikimedia has an active lexeme project which can resolve lexical forms and map them to concepts, e.g. the English terms “ethics committee” and “informed consent form” are represented by the Wikidata lexemes L497553 and L497589, respectively. These lexeme entries in turn link information about these English nouns, their grammar and meaning to information about the underlying concepts (e.g. Q59057226 for “ethics committee” as a subclass of committee) as well as equivalent terms in other languages, which can also occasionally be found in ethics statements.

For instance, Fig. 5 shows that the German noun for ethics committee, Ethikkommission, (known to Wikidata as L562403) is used in ethics statements, both in articles written in German - e.g. Nuessle et al. (2021), Gugatschka et al. (2021) - as well as in English - e.g. Krajka et al. (2021), Leuenberger et al. (2021).

Figure 5.  

The German term for ethics committee - Ethikkommission - used in the context of documents otherwise written in English (top, from Krajka et al. 2021) or German (middle, from Nuessle et al. 2021, and bottom, from Gugatschka et al. 2021).

Complementing these mono- and bilingual examples, Fig. 6 gives an example in which several ethics committees are listed using both a name in their original languages and an English-language equivalent.

Figure 6.  

Examples for entities in English text (highlighted in peach), with text elements in multiple languages (underlined, darker shade): French, Portuguese, German.

Taking such cross-linguistic information into account can thus facilitate entity recognition in ethics statements even in English texts and help expand the methodology to mining articles in other languages as well, e.g. to identify or distill boilerplate phrases in a given language or cultural differences across languages in terms of how ethics-related information is handled. For any language with information about such boilerplate phrases, a score could be computed that could represent the similarity between boilerplate text and phrasing from a given article. Such scores could be used, for instance, to guide community curation efforts - high similarity to known boilerplate means high potential for automation and less need for human oversight, while low similarity indicates a need for community review.

Ethics statements as a less explored use case for testing text mining approaches

Standardizing the ethics statements

In this work, we outlined a set of core ideas for mining the literature, extracting ethics-related entities and relationships, reconciling them with a controlled vocabulary, making the information queryable and creating a positive feedback loop between the structured information and the mining workflows by iteratively using one to improve the other.

Much like in other areas of data mining, initial challenges for the mining of ethics statements include handling inconsistent approaches to the naming of relevant entities (e.g. institutions, ethics committees, laws and other relevant policy frameworks). This is compounded by inconsistency as to where in a document the ethics statements are located (e.g. in a dedicated section, or as part of the Methods or in an Annex).

If these challenges can be addressed, the mining of ethics statements can provide significant value in terms of elucidating the research ethics landscape (highlighting relevant organizations, along with policies, guidelines and other standardization efforts) as well as documenting, improving, teaching and standardizing current practices in research ethics. A systematic analysis of the ethics statements will also highlight institutional, disciplinary and other contexts in which such statements are common or well-developed, uncommon or underdeveloped, or anywhere in between.

This can form the basis for studying ethical aspects of the research process - as well as ethics review - under specific conditions and for addressing ethical aspects of research both in practice as well as in teaching. For instance, key elements of contexts in which well-developed ethics statements are common - such as a clear policy, readily actionable community guidelines or scalable workflows - could serve as a starting point for exploring best practices or synthesizing recommendations, while other contexts could be explored in terms of their potential for improvements.

Another point to consider is that access to the ethics-related information contained in a publication currently requires access to the full text. However, the basic ethics data - such as whether the research reported in the publication received ethical approval, what the approving bodies were and what the relevant approval numbers are - should be considered metadata and in the public domain. Ideally, they would be incorporated into the filtering mechanisms provided by individual databases or scholarly search engines and visualization tools more generally. Some databases like stem cell registries (cf. Kurtz et al. 2022) already provide ethics-related information.

We plan to work towards implementing the core ideas presented here, and we very much welcome collaborations in this regard.

In particular, we plan to extract information and phrasing pertaining to ethics committees and other entities commonly found in ethics statements (e.g. policies and guidelines) and to make this information available via suitably annotated Wikidata items and lexemes that can in turn be used by mining pipelines. Once the data models in this area have stabilized, it would be possible to scale up these workflows by increasing their automation and expanding the mining to auxiliary materials like approval letters, which are currently shared only very rarely , or to annotating ethical aspects of things other than formal publications, e.g. clinical trials or their consent forms that are now increasingly being made public too.

Further, we plan to work on visualizations that present this structured information and that can be incorporated into suitable parts of the open knowledge ecosystem, particularly through Wikimedia platforms and associated visualization services like Scholia (Nielsen et al. 2017, Lemus-Rojas et al. 2022).

Beyond ethics statements, we plan to apply the ideas outlined here also to other non-traditional parts of research manuscripts, e.g. data availability or conflicts of interest. We also aim to explore how these approaches can assist with the enrichment of mining efforts targeted at less-mined aspects of manuscripts, e.g. the citation of data, software and material resources. In doing so, we will focus on resources that are openly available.

Neither the work proposed nor the work presented here has so far received funding.

The authors declare that they have no conflicts of interest pertaining to the research described here.