Last verified: August 16, 2026. A trusted research environment (TRE) is a secure, controlled computing environment that lets approved researchers analyse sensitive data — usually de-identified health, administrative, or other personal data — without the underlying data ever leaving the environment. Only aggregated, disclosure-checked results go out. For the core definition, see the Trusted research environment dictionary entry. This guide covers the part a definition can’t: how the Five Safes framework is actually assessed, what the accreditation and access procedure looks like, and how “TRE,” “data safe haven,” and “secure data environment” relate to one another.
The Five Safes framework: what each “Safe” is actually assessed on
The Five Safes framework was developed by Tanvi Desai, Felix Ritchie, and Richard Welpton (Bristol Centre for Economics and Finance working paper, 2016) as a way to decompose “is it safe to give this researcher access to this data” into five independently manageable controls. It now underpins UK government statistical microdata access (the Office for National Statistics’ Secure Research Service), the NHS England Secure Data Environment network, and most TREs and safe havens internationally.
| Safe | Question it answers | What is actually assessed |
|---|---|---|
| Safe People | Are the researchers trustworthy and competent? | Identity verification, relevant training (e.g. completed data-protection/disclosure-control training), and a track record or institutional affiliation that supports accountability. |
| Safe Projects | Is this specific use lawful, ethical, and in the public interest? | A defined research question, legal basis for processing, ethics/IRB approval where applicable, and a public-benefit or statutory-purpose test tied to the specific dataset requested. |
| Safe Settings | Is the analysis happening somewhere the data can’t leak? | Technical controls on the compute environment itself — no local downloads, no internet egress from the analysis environment, audited/logged access, virtual desktop or similar architecture. |
| Safe Data | Has the data been reduced to only what’s needed? | De-identification or pseudonymisation appropriate to risk, minimisation to the variables the approved project actually needs, and sometimes synthetic or dummy data for pre-analysis code development. |
| Safe Outputs | Could anything leaving the environment re-identify someone? | Statistical disclosure control review of every output (tables, charts, model coefficients) before release — checking for small cell counts, residual disclosure risk, or outputs that could be combined with external data to re-identify an individual. |
A TRE operator’s job is to translate each of these five into concrete, auditable controls; a researcher applying for access is being assessed against all five at once, not just asked to sign a data-sharing agreement.
How TRE access actually works: the procedure
The exact application form, portal, and turnaround time differ by TRE operator — this is a general procedure common to Five-Safes-based TREs, not a single named application process. Always check the specific TRE’s own current documentation for forms and timelines before applying.
- Researcher accreditation. You (the individual researcher) apply to be recognised as a trustworthy user — typically identity verification, institutional affiliation, and evidence of relevant training. In the UK, the Digital Economy Act 2017 created a formal legal gateway for “accredited researcher” status to access de-identified personal data held by public bodies for research; several TRE operators run an equivalent accreditation step even where the Act itself doesn’t apply.
- Organisational/processor accreditation. The institution or team you’re working through (university, agency, NHS trust) may itself need to be accredited or under an existing data-sharing agreement with the data controller before any of its researchers can be approved.
- Project approval (Safe Projects). You submit a specific research proposal — the question, the legal/ethical basis, the exact variables and cohort needed, and (for identifiable or sensitive data) evidence of ethics/IRB approval. Access is granted to a project, not to a dataset in general; a new project typically means a new approval.
- Data access agreement. A data sharing agreement or equivalent contract is signed, setting out the data controller and data processor obligations, retention limits, and permitted uses.
- Provisioning into the Safe Setting. You’re given credentials to the controlled environment itself — a remote virtual desktop or similar architecture with no local download path and no unrestricted internet access, with all activity logged.
- Analysis inside the environment. All coding, cleaning, and analysis happens inside the TRE against the de-identified Safe Data; nothing is copied out at this stage.
- Output disclosure review (Safe Outputs). Before any table, chart, or result leaves the environment, TRE staff (or an automated statistical disclosure control check) review it for re-identification risk — small cell sizes, unique combinations of characteristics, or anything that could be cross-referenced against public data.
Trusted research environment, data safe haven, secure data environment: one idea, three names
These terms describe essentially the same “bring the analysis to the data” model, and in current usage they overlap almost completely — the differences are mostly historical and terminological, not functional.
| Term | Where it comes from | Current usage |
|---|---|---|
| Data safe haven | NHS England and Scottish government information-governance literature, late 2000s onward (predates “TRE” as the dominant term) | Still used by name by some operators (e.g. the eDRIS Safe Haven in Scotland); functionally a TRE. |
| Trusted research environment (TRE) | Popularised across UK health data policy through the 2010s-2020s, especially the Five Safes-aligned NHS/ONS infrastructure | The most common umbrella term today for any Five-Safes-based secure analysis environment. |
| Secure data environment (SDE) | The specific term adopted by NHS England for its national and sub-national network of TREs, following the 2022 Goldacre Review | Effectively NHS England’s branded name for its own TRE network; used interchangeably with “TRE” in NHS contexts. |
If you’re searching for any of these terms, you’re generally looking for the same underlying access model — the Five Safes framework, applied through an accreditation-and-approval procedure like the one above.
TRE infrastructure researchers actually encounter
Named, real examples of live Five-Safes-based TRE infrastructure include:
- The Office for National Statistics’ Secure Research Service — the original UK government implementation the Five Safes model was built around.
- NHS England’s Secure Data Environment (SDE) network, including sub-national SDEs, established following the 2022 Goldacre Review recommendations.
- SAIL Databank (Wales) and the eDRIS Safe Haven (Scotland), both self-described as safe havens operating under Five Safes principles.
- The Health Informatics Centre (HIC) at the University of Dundee.
- OpenSAFELY, a platform-based approach that runs analysis code against NHS primary care data without researchers ever directly accessing patient-level records.
Could not independently verify at draft time: specific current application turnaround times, fees, and technical requirements vary by operator and change over time — check the named service’s own current documentation rather than relying on this page for those specifics.
Standards and accreditation landscape
There is no single global “TRE certification” a researcher or environment gets stamped with; instead, several UK initiatives are working toward common standards:
- DARE UK (Data and Analytics Research Environments UK) coordinates national work on TRE standards, including efforts toward a common technical architecture across TRE operators.
- The UK Health Data Research Alliance published a Green Paper on TREs setting out shared principles for trustworthy operation.
- The 2022 Goldacre Review (“Better, broader, safer: using health data for research and analysis,” UK Department of Health and Social Care) recommended consolidating NHS data access around a smaller number of well-governed SDEs rather than large numbers of data extracts and copies — the policy basis for NHS England’s current SDE network.
- Individual TRE operators typically also hold general information-security certifications (e.g. ISO/IEC 27001) alongside their Five Safes accreditation processes, though the specific certifications held vary by operator.
Frequently asked questions
What is the Five Safes framework?
A model for assessing whether it is safe to give a researcher access to sensitive data, broken into five independent controls: Safe People, Safe Projects, Safe Settings, Safe Data, and Safe Outputs. See the table above for what each one actually checks.
Is a data safe haven the same thing as a trusted research environment?
In current usage, yes — “data safe haven” is largely the same concept under an older name, still used by some UK operators (e.g. in Scotland), while “trusted research environment” and NHS England’s “secure data environment” are the more common current terms for a Five-Safes-based controlled analysis environment.
How do I actually get access to a TRE?
Broadly: get accredited as a researcher, have your project approved on Safe Projects grounds (including ethics approval where relevant), sign a data access agreement, get provisioned into the controlled Safe Setting, do the analysis inside it, and have your outputs pass disclosure control review before they leave the environment. Exact steps and forms vary by operator — check the specific TRE’s own documentation.
Do I need ethics/IRB approval before I can apply for TRE access?
Usually yes for projects involving identifiable or sensitive personal data — ethics/IRB approval is typically part of what’s assessed under Safe Projects, alongside the legal basis for processing and a defined public-benefit or research purpose.
Can I download or copy data out of a TRE?
No — that is the defining feature of the model. Only aggregated, disclosure-checked outputs (tables, statistics, model results) are allowed to leave, after review under Safe Outputs; the underlying record-level data stays inside the Safe Setting.
