FORTRESS explores the opportunities and challenges in accessing synthetic free-text data within Trusted Research Environments (TREs): from data generation, via validation to deployment and access.
By using synthetic free-text data that does not include information about real people, FORTRESS (Federated, Open and Reliable TREs for Synthetic Textual Healthcare Records) aim to enhance privacy of TREs and use such data as a risk minimisation approach for a data type that has not been widely available within TREs
Free-text data (such as hospital letters, financial reports, interview transcripts, etc.) are a useful source of information for research, but it often contains sensitive information and therefore is not widely available within TREs.
Automated, large-scale de-identification methods can remove some pieces of sensitive information, but the overall risk when such data is accessed (e.g. for model development) is unclear. Synthetic text generation is emerging as an alternative solution, but it requires clear protocols for generation, validation and deployment within TREs.
If the process for synthetic free-text data generation can provide good quality data, it can be used to generate large scale, purpose-specific datasets (e.g. cardiology) that can be used for a variety of tasks (e.g. model development and diagnosis) with significant reduction in privacy risks. This will increase public benefit and trust in responsible use of sensitive free-text data within TREs.
Led by the University of Manchester in collaboration with Imperial College, the project is delivered in partnership with our public advisory group, TREs (Secure eResearch Platform), regulatory bodies (e.g. Medicines and Healthcare products Regulatory Agency), industry (Umbizo) and the wider DARE UK family (TREvolution, STAR-TRE, TRExt, SAFETEXT, etc.).
What FORTRESS aims to achieve
By the end of the project, the FORTRESS team aims to better understand the challenges in generating, accessing and using synthetic free-text data, using healthcare as a case study. Their goals include:
- Involving the public – understand public views on the use, purpose and challenges of synthetic free-text data.
- Develop a prototype – create a model for text generation approach, and demonstrate its capabilities in the healthcare domain.
- Validate prototype – develop and demonstrate a validation toolkit to assess the quality of generated free-text data.
- Infrastructure testing – explore the barriers in the deployment of synthetic text data within TREs, including the environment needed for the execution, tuning and validation of models and datasets.
- Integrate state-of-the-art principals – develop an approach to make synthetic text data FAIR (findable, accessible, interoperable and reusable) by documenting the generation process and outcomes.
Progress so far
Within the first six months of delivery, the FORTRESS team has:
- Established a Public Advisory Board and conducted three focus group discussion events with the members of the public (January-March 2026). The outcomes of these discussions have been used to shape our research approach and are being prepared as a report/publication.
- Developed and tested a prototype version of text generation approach (Term2Note), which generates healthcare notes based on publicly available real-world dataset (MIMIC III) using SNOMED CT coding and differential privacy.
- Packaged the prototype version of Term2Note and prepared it for deployment to a TRE. A testing protocol has been developed.
- Developed a first version of the validation toolkit that includes both automated and manual checks of the authenticity of the generated healthcare notes.
- Organised and presented the FORTRESS project at several community events and conferences.
In the remainder of the project, we will engage with the wider TRE community (including the DARE UK working groups on synthetic data, information governance, statistical disclosure control) to seek feedback on the current developments. We will complete the validation toolkit and run the deployment trial with partner TREs, including expanding the RO-CRATE meta-data model to describe both the text generation and outcomes (synthetic text dataset) to support FAIRification.
Relevance to DARE UK vision
The FORTRESS project will provide evidence and improved understanding of the opportunities and challenges in reducing privacy risks by providing access to synthetic text data to support research and innovation. This goal will contribute to DARE UK vision for responsible and secure use of diverse sensitive data. The outcomes could transform the free-text data availability within TREs for particular purposes – for example, as development data for application development, model training or staff education.
About the DARE UK Next-Gen Catalysts
The DARE UK Next-Gen Catalysts are early-stage, exploratory projects developing and testing the next generation of capabilities for Trusted Research Environments (TREs) across the UK. Bringing together researchers, technology specialists, public contributors and other partners, these projects explore bold new ideas that could expand how sensitive data is accessed, analysed and governed while maintaining privacy, security and public trust. Their role is to generate evidence, prototypes and practical learning that can help shape the future of secure, trustworthy data research for public benefit.
