Toward the Development of Data Governance Standards for Using Clinical Free-Text Data in Health Research: Position Paper

Kerina Jones, Elizabeth Ford, Nathan Lea, Lucy Griffiths, Lamiece Hassan, Sharon Heys, Emma Squires, Goran NenadićJournal of Medical Internet Research 2020 · DOI

Abstract

By drawing together the findings of a combination of activities, we present a position paper to contribute to the development of data governance standards for the reuse of clinical free-text data for secondary purposes. While working in accordance with existing data governance frameworks, there is a need for further work to take forward the recommendations we have proposed, with commitment and investment, to assure and expand the safe reuse of clinical free-text data for public benefit.

Introduction: The Untapped Potential of Clinical Narratives

Electronic health records (EHRs) are a cornerstone of modern health research. While structured data—such as laboratory results, medication codes, and demographic information—is relatively easy to standardize and de-identify, a vast wealth of clinical knowledge remains locked in "free-text" formats. These unstructured narratives, including doctor’s notes, discharge summaries, and psychiatric evaluations, provide critical context regarding disease progression and social determinants of health that structured fields often omit.

However, the use of free-text data presents significant privacy challenges. Unlike structured data, where identifiers are confined to specific columns, personal information in free-text can appear anywhere (e.g., "Patient lives next door to the Green Grocer on High Street"). This creates a "utility-privacy" trade-off: aggressive de-identification can strip the text of its clinical meaning, while insufficient scrubbing risks patient re-identification. Currently, there is a lack of established, authoritative guidelines for governing the reuse of this sensitive data format.

The TexGov Study: Methodology

To address these gaps, the "TexGov" study employed a multi-pronged qualitative approach focused on the UK healthcare landscape. The study consisted of four primary components:

  1. Regulatory Review: An analysis of how the GDPR and the UK Data Protection Act (2018) apply specifically to unstructured data.
  2. Rapid Literature Review: A survey of 51 studies to identify how researchers currently access and govern free-text data.
  3. Infrastructure Case Studies: In-depth interviews with managers of four major UK data infrastructures—THIN, CRIS, CPRD, and SAIL—to understand their specific de-identification and access models.
  4. Stakeholder Engagement: Workshops with text-mining researchers and public engagement events, including a "citizens' jury" style panel, to explore perceived barriers and social acceptability.

Analysis of Existing Governance Models

The study highlighted several distinct approaches to managing free-text data across major UK repositories:

  • THIN (The Health Improvement Network): Utilizes a "blacklist" approach to mask known identifiers.
  • CRIS (Clinical Records Interactive Search): Focuses on mental health records, operating behind a secure firewall to maintain privacy.
  • CPRD (Clinical Practice Research Datalink): A large primary care database that previously provided free-text but discontinued the service due to privacy complexities.
  • SAIL (Secure Anonymised Information Linkage): Employs a "whitelist" approach, extracting only known clinical terms to ensure no identifiers leave the source.

Key Insights: The "Potentially Identifiable" Paradigm

A central finding of the TexGov study is that even "de-identified" free-text should be treated as potentially identifiable. Because the risk of residual identifiers can never be entirely eliminated, the authors argue that this data should not be released as open data. Instead, access should be restricted to accredited "Data Safe Havens"—secure environments where researcher activity can be monitored.

Furthermore, the study proposed the concept of a "Text Bank." Much like an organ donor registry, patients could donate their raw clinical notes to create a gold-standard training set for the community to improve Natural Language Processing (NLP) and de-identification algorithms.

Recommendations for National Standards

The paper culminates in six core recommendations to establish a national data governance standard for clinical free-text:

  1. Clear Regulatory Guidance: Authorities must provide specific rules for free-text, distinguishing between blacklisted, whitelisted, and raw formats.
  2. Transparency: Patients should be better informed about how their unstructured text data flows through the research ecosystem.
  3. Publicity of Benefits: Case studies of successful free-text research should be published to demonstrate value to the public.
  4. Continuous Improvement: A commitment to a "national challenge" to benchmark and improve de-identification algorithms.
  5. Safe Haven Access: Access to blacklisted text should be restricted to accredited safe havens unless a thorough manual review proves it is truly anonymous.
  6. Standardized Reporting: Journals should require a "data governance statement" for any study using clinical free-text.

Conclusion and Future Directions

While the public generally supports the use of free-text for the "public good," this support is conditional on transparency and security. The TexGov study provides a roadmap for balancing these needs. However, limitations remain: the study was UK-centric and did not directly engage the clinicians who create the data. Future research must investigate whether the secondary use of notes leads to "defensive writing," where clinicians omit sensitive details to protect privacy, potentially undermining the data's utility for research.

Summary updated

Abstract

By drawing together the findings of a combination of activities, we present a position paper to contribute to the development of data governance standards for the reuse of clinical free-text data for secondary purposes. While working in accordance with existing data governance frameworks, there is a need for further work to take forward the recommendations we have proposed, with commitment and investment, to assure and expand the safe reuse of clinical free-text data for public benefit.
BibTeX
@article{jones2020toward,
  author = {Kerina Jones and Elizabeth Ford and Nathan Lea and Lucy Griffiths and Lamiece Hassan and Sharon Heys and Emma Squires and Goran Nenadić},
  title = {{Toward the Development of Data Governance Standards for Using Clinical Free-Text Data in Health Research: Position Paper}},
  journal = {Journal of Medical Internet Research},
  year = {2020},
  doi = {10.2196/16760},
}
Figure