Data Sharing Throughout the Research Lifecycle

A single data asset may be used in multiple ways (towards journal articles, posters, or conference presentations) and often undergoes several transformations. Tracking data provenance (origin, version history, and chain of transformations) ensures data transformations and outputs are traceable and verifiable.

Lessons learned: Without clear provenance it becomes difficult to determine which data version supports a specific finding. This limits reproducibility and may lead to misinterpretation or unnecessary raw data re-processing. Maintaining strong provenance builds trust, ensures transparency, and protects against misuse.

What to do:

• Establish version control strategies to track changes to data, code, and other applicable research processes. Some data repositories automatically version-control data deposits, including archiving older versions to support reproducibility.
• Document provenance, maintaining records of the origin, transformations, and handling of datasets.

Additional resource:

• HEAL Data Sustainability How To Guide

Data cleaning enhances data quality and integrity, ensuring findings are valid and datasets are suitable for reuse.

Lessons learned: Inconsistent or incomplete cleaning can compromise analysis and results, inhibit reproducibility, and distort downstream analyses and insights..

What to do:

• Automate cleaning with code, removing or correcting erroneous or duplicative data points to ensure datasets are valid, accurate, and fit for purpose.
• Adopt quality control (QC) processes, including regular quality reviews or automated QC checks, to report on data quality measures during data collection and processing.
• Document calculations, aggregations, derivations, and missing data, using code / scripts, SOPs, etc., to provide transparency into the data provenance and allow secondary users to assess dataset validity.
• Transform, normalize, and aggregate data to ensure structural and semantic consistency. For example, convert categorical values into standard formats or align date/time formats across sources.

Additional resources:

• Article: Normal Workflow and Key Strategies for Data Cleaning Toward Real-World Data: Viewpoint
• ICPSR Guide to Social Science Data Preparation and Archiving

Protecting sensitive information is essential for ethical research and compliance with regulations like HIPAA. De-identification and anonymization transform or eliminate direct (e.g., names, email addresses) and indirect (e.g. geographic information) identifiers, but anonymization may require more specialized skills, such as statistical disclosure control.

Lessons learned: Unclear, inconsistent, or late de-identification or anonymization can lead to privacy breaches, legal and ethical noncompliance, rework, or unsharable data. On the other hand, excessive anonymization may strip away important context, reducing data’s reusability.

What to do:

• Determine when de-identification should occur by considering factors such as data access permissions, storage environments, and the audience. Aim for early de-identification to reduce deductive disclosure risk.
• Tailor de-identification to the data type. For example, redact or pseudonymize direct and indirect identifiers in qualitative data.
• Apply techniques proportional to privacy requirements. For HEAL-funded research, de-identification and access controls are often sufficient for data privacy requirements.
• Document de-identification techniques and tools. Describe any review or validation process performed and the results of the review.
• Use a human validator to validate de-identification results. Institutions and data repositories may provide honest brokers or data curators for support.

Additional resources:

• HEAL Stewards Sensitive Data Decision Tool
• Types of Sensitive Data used in HEAL Research
• GREI - Best Practices for Data Submission in Generalist Repositories: A Checklist

Metadata is essential for making research data findable, interpretable, and reusable. In the HEAL Data Ecosystem study-level and variable-level metadata (SLMD and VLMD) support HEAL Data Platform findability and enable HEAL Semantic Search interoperability. Tabular data VLMD is often stored in a data dictionary or codebook. Protocols, standard operating procedures (SOPs), and README files typically contain VLMD. Non-tabular VLMD may be stored in files such as interview scripts or coding schemas (qualitative data), instance files or annotations (imaging), or sequence files (genomics data).

Lessons learned: Insufficient metadata makes data hard to find, interpret, integrate, or reuse, even by the original study, over time. Missing metadata reduces visibility and increases the risk of misinterpretation or misuse.

What to do:

• Complete the HEAL SLMD form to share study-level metadata.
• Submit VLMD, ideally in a data dictionary/codebook or appropriate non-tabular format.
• If needed, consult a data curator from your organization’s libraries or from a data repository.

Additional resources:

• HEAL Metadata Resources
• HEAL Data Dictionary Preparation Guidance
• Generate a HEAL-compliant Data Dictionary (HEAL Platform)
• OSF’s How to create a data dictionary
• The Data Curation Network’s Data Curation Primers