What does ‘sharing’ mean?

When discussing ‘sharing’ in the context of academic investigations, we mean making relevant materials, key to understanding the conduct and analysis of a study, available to interested parties to the extent ethically and legally possible (box 1). Fundamental to sharing of study materials is that they are made available as part of the timely and complete dissemination of results. While we focus on sharing as it relates to data and materials being made available either before or alongside results, biases in the availability of results will undermine any efforts for more open and transparent science. For many, sharing in the context of research means making their raw data available. Sharing data allows for external review and reuse to support future research.25 In 2018, the International Committee of Medical Journal Editors began requiring that articles reporting clinical trials in member journals must contain a data-sharing statement.26 Notably, this is not a requirement to share data, but rather an attempt to be transparent about the extent or conditions around data sharing. However, there is no guarantee these statements will be implemented, honoured or to lead to increased sharing.27 28

While data sharing is important, it is only one aspect of the study methods that can be made available.29 A wide array of documentation, data collection sheets and instruments, analysis plans and code, and various other forms of information can be shared to provide insights into how a study was conducted (box 2). The Transparency and Openness Promotion Guidelines grade journal policies across eight standards, including the sharing of data, analytic code, materials and documentation (ie, protocols and analysis plans), and each area is assessed across four levels ranging from ‘not implemented’ to having an active requirement and verification of sharing.30

Why should study materials be shared?

It is common practice in manuscripts with data-sharing statements to make even public, non-sensitive data, or other study materials, available only by ‘reasonable requests’ to an author which can pose barriers between potential users and study materials.31 32 Email requests for access can go to defunct addresses, get lost in busy inboxes, or simply be ignored or rejected.27 33 Moreso, decisions to share are reliant on an individual’s subjective evaluation of what is ‘reasonable’ unless clear criteria for requests are articulated and consistently applied.34 If data can legally and safely be shared, it should be the default to allow for proactive transparency and ease of use.

The first article in this series covered pre-registration which can be considered a specific type of sharing of analysis plans and protocols prior to study commencement.35 Making these available allows for assessments of potential biases and provides details for future researchers. Making the analytic code used to clean, process and analyse study data in programs like R, STATA or Python available can help catch errors, facilitate future research and provide insights into how a study protocol was practically implemented.3 Observational health researchers working with large datasets can share the codelists (eg, ICD-10 codes) that are used to extract analysis populations and define variables of interest in order to avoid duplicating efforts and ensure consistency in analyses. Those involved in qualitative studies can make study materials, like interview guides, available to provide insights into how data were collected. Repositories, such as the National Heart, Lung and Blood Institute Biologic Specimen and Data Repository, facilitate access to biospecimens and other clinical and epidemiologic data to benefit other researchers.

These are just some common examples of how sharing may occur and aid in promoting reproducibility, secondary analyses and increased utility of publicly funded research outputs. There is even evidence that more open sharing practices leads to higher citation rates of a given work.36 For any study, we believe it is valuable to reflect on what materials were key to its conduct and whether or not it can, and should, be shared. It may also be prudent to consider sharing to the extent possible prior to publication, rather than after publication, to ensure that reviewers, editors and other interested parties can review study details for issues, errors or biases before they enter the literature.37

How can materials be shared?

Proponents of better data sharing have developed the FAIR standards for “Findability, Accessibility, Interoperability, and Reuse.”38 The details of FAIR are specific to making data available and usable, and have recently been expanded to research software,39 but provide a useful framework for thinking about what makes sharing successful more generally:

• Findability and accessibility: Sharing in obscure or unindexed locations impacts whether others can discover and use the information. In addition, one-off sharing on personal, or other non-persistent, repositories does not guarantee interested parties will be able to access the information into the future. Where relevant, ensuring appropriate metadata is attached to your information will also allow for more efficient discovery by others.

• Interoperability and reuse: Posting information in a repository without any documentation or instructions can mean these materials are indecipherable to others. While sharing a protocol and statistical analysis plan that explains the methods and analysis in detail is a minimum standard, further clarifications—possibly including a data dictionary, commented and documented code, or shareing instructions or guides about data extraction—can help the inspection and reuse of data. Some journals offer researchers the opportunity to publish formal descriptions of available data sets to increase their visibility and document their contents and use (eg, Nature Scientific Data). Properly presenting and documenting data or code will also make their future use more effective and efficient.40–42 It can also be useful to share information in standard formats that are easily usable across a variety of platforms. For instance, data in a .CSV file is more accessible than a program-specific data file (eg, a STATA .DTA file). Some industries or research areas will have niche standards for interoperable, open data formats (eg, Clinical Data Interchange Standards Consortium standards) that may be relevant to certain researchers. When sharing in non-standard or proprietary formats is the best option, it should be clearly documented how and what is necessary to access and use the information.

Specific methods for sharing will also depend on what is being shared. The simplest form of sharing would involve making any study materials available as supplements to the manuscript. This has the clear benefit of being immediately available to readers; however, this practice may have other limitations. Links to files can break in journal website redesigns and there may be restrictions as to how information can be shared, accessed and reused. For instance, sharing a large data set as a PDF would compromise its usability by others. This method also makes sharing contingent on journal publication.

Alternatively, information can be shared on services and repositories created to persistently house academic outputs. Preprint servers can host additional study materials, with few limitations, alongside an open-access version of the manuscript. The Open Science Framework, Figshare and Zenodo are some examples of more general-purpose repositories that offer generous free storage of nearly any type of file, structured metadata, version control and digital identifiers (eg, DOIs) for easy citation and linkage and often integrate with one another (table 1). More specialised repositories, like GitHub for code, or ClinicalTrials.gov for trial registrations, offer similar functionality and can be valuable tools for sharing. Using these repositories makes sharing as easy as putting a persistent URL or DOI into a manuscript that points to all available materials. Some types of studies may require more sophisticated sharing methods that ensure reproducible computational environments.43

Some platforms are specifically designed for securely warehousing sensitive data, particularly relevant in health and medical research. Vivli and YODA are two examples of repositories that help securely store and manage access to individual patient data from clinical research.44 There is also increased investment in Trusted Research Environments in which data providers offer approved researchers firewalled access to conduct analysis in which the raw data would never leave the environment.45 By sharing code designed for analysis within a given Trusted Research Environment, any researcher with access should be able to replicate, verify or expand on a finding. It may also be worthwhile to check the requirements and recommended resources or repositories from the researcher’s institution or funders for depositing and managing access to sensitive information.

Barriers and precautions

While increased sharing may be ideal, numerous barriers exist, especially for health and medical research.46 The biggest concern with sharing information, especially data, from health and medical studies is the potential to expose sensitive or confidential information. Legal regimes, such as the General Data Protection Regulation in Europe, must also be considered when sharing sensitive data derived from research. A key component of sharing is that it must be done thoughtfully and responsibly.47 As discussed previously, there are specialised repositories which can help manage access to sensitive data.

For some studies, it may be prudent to make data sharing part of the initial participant consent process. Using this model, study participants would understand exactly what data they are providing and the ways in which these could be used or shared. If the risks of re-identification are well described, understood and fully consented to, this may allow for more effective sharing of data.48 Even when sharing data in accepted or responsible ways, care must be taken to ensure information is not overly disclosive or the anonymity of the data is overemphasised. Pseudonymisation is a common practice in which some identifiable data are removed; however, this does not mean that the data are not disclosive.49 Furthermore, thought should be given to whether existing datasets could be combined with your open data to de-anonymise participants.50 Data leakage and vulnerabilities may even exist in unexpected places. For instance, SVG files, a common image format requested by journals, embed the data underlying the image directly in the file and could be disclosive.

Researchers should also consider whether they have ownership or permission to legally share data. For instance, one might do a study using the UK Clinical Practice Research Datalink dataset which is built from primary care electronic health record data, or with proprietary data purchased from a commercial data supplier such as those detailing insurance claims or drug sales. If sharing is not possible, be transparent about the reasons why, state who owns and manages the data, and detail how the dataset was used to the extent possible. Additional materials (eg, the protocol and analytic code) can often still be shared. When these considerations are not relevant, information can be shared under appropriately permissive licensing like Creative Commons or The MIT Licence for software.

Lastly, properly sharing data requires both human and financial resources to build data-sharing capacity, access platforms, respond to inquiries, and meet data security specifications. This creates a need for investment in skills and resources alongside systems that promote and reward openness and transparency.51