Data Clean Rooms vs Federated TREs: Key Differences

A data clean room is a neutral computing environment — originating in advertising technology — where two or more parties match and analyse their combined datasets without exposing raw records to each other. A federated Trusted Research Environment (TRE) is a governed research platform in which analysis is dispatched to the data where it lives, so the data never leaves the source. The core difference is directional: a clean room still centralises copies of contributed data into one neutral venue, while a federated TRE moves the computation instead of the data — and adds the vetting of people, projects, and outputs that regulated health research demands.

Why pharma teams are asking this question now

Clean rooms became mainstream in advertising after third-party cookies collapsed, and the vocabulary has migrated into life sciences: sponsors now hear “clean room” pitched for real-world evidence collaborations, trial-recruitment feasibility, and data partnerships with health systems. At the same time, the May 2026 UK Biobank incident reset expectations of what “secure” means. Approved researchers exported participant-level data through a centralised platform’s normal, policy-compliant workflow — no hack, no rule broken, just an architecture that permitted bulk egress. For any organisation evaluating collaboration infrastructure, the incident made one question unavoidable: does the platform make large-scale data exfiltration procedurally forbidden, or architecturally impossible?

Regulation is moving in the same direction. The European Health Data Space (EHDS) regulation, in force since March 2025, requires secondary use of health data to happen inside secure processing environments with controlled outputs — a description that fits a Trusted Research Environment far more naturally than an adtech-style clean room. Sponsors choosing infrastructure today are choosing what they will have to defend to regulators, ethics committees, and data custodians for the next decade.

What each technology actually is

Data clean rooms: neutral ground for matched analysis

A data clean room lets two parties — classically an advertiser and a publisher, increasingly a pharma company and a data partner — bring datasets to a neutral environment, match them on hashed or tokenised identifiers, and run approved queries over the intersection. Privacy controls typically include aggregation thresholds (no result below a minimum cohort size), query restriction, and noise injection; the Interactive Advertising Bureau (IAB) Tech Lab has published guidance and interoperability standards for these mechanisms. Clean rooms are genuinely useful for what they were built for: measuring overlap and computing aggregate statistics between organisations that do not trust each other with raw data. Their trust model, however, is thin by health-research standards — the parties are vetted commercially rather than through governance review, the venue holds copies of both datasets for the duration, and output rules are configured by the parties themselves rather than reviewed by an independent custodian.

Federated TREs: governed research where the data lives

A Trusted Research Environment is a secure analytics environment purpose-built for sensitive health data, governed by the Five Safes framework developed at the UK Office for National Statistics: safe people (accredited researchers), safe projects (approved purposes), safe settings (controlled infrastructure), safe data (de-identified or pseudonymised), and safe outputs (disclosure control on everything that leaves). A federated TRE extends this by removing the central copy altogether: each data custodian — a biobank, hospital network, or national programme — keeps its data inside its own boundary, and the platform sends approved queries, pipelines, or model-training jobs to each site, returning only aggregate, disclosure-checked results through an automated airlock. The federated Trusted Research Environment is the architectural answer to the UK Biobank failure mode: there is no central store to exfiltrate from, and record-level egress is not a permission that can be misconfigured — it is a capability that does not exist.

Data clean room vs federated TRE: the comparison

The federated TRE angle: why direction of travel decides everything

Most differences in the table reduce to one design decision. A clean room mitigates the risks of pooling data; a federated Trusted Research Environment eliminates the pool. That distinction cascades through every governance conversation a sponsor will have. Data custodians — the hospitals, biobanks, and registries whose participation determines whether a real-world evidence programme succeeds — are far more willing to join a network in which their data remains inside their firewall, under their access policies, in their jurisdiction, than one that asks them to ship a copy elsewhere, however “neutral” the destination. Federation is not only a security property; it is a recruitment strategy for data partnerships.

It also changes what analysis is possible. Clean rooms restrict researchers to a narrow query grammar because the venue cannot risk arbitrary code against pooled raw data. A federated TRE inverts this: because each site controls execution locally and only aggregates leave, the platform can safely support genuinely deep workloads — genome-wide association studies, survival analysis, and federated machine learning in which models train across sites without any site’s records moving. The distinction between federated analytics and federated learning matters here: clean rooms offer neither in full generality, while a federated Trusted Research Environment offers both under a single governance framework.

Real-world example: national-scale federation

Genomics England demonstrates the model at national scale. Its research environment gives thousands of approved researchers analytical access to pseudonymised whole genomes linked to longitudinal National Health Service (NHS) records — a dataset that could never lawfully or ethically be uploaded to a commercial clean-room venue. Researchers run full bioinformatics and statistical workloads inside the environment, and only disclosure-checked results leave. In Southeast Asia, Lifebit’s work with Singapore’s public-health ecosystem applies the same pattern across institutional boundaries: data holders retain custody, compute is dispatched to each source, and national sovereignty requirements are satisfied by architecture rather than by contract. These are workloads and trust relationships that clean rooms — built for matching marketing datasets — were never designed to carry.

Choosing between them: a practical framework

The technologies are not strict competitors; they answer different questions. A clean room is a reasonable fit when two commercial parties need aggregate overlap analysis on tokenised data, the analysis fits a constrained query model, and no custodian, ethics, or sovereignty approval is in play. A federated Trusted Research Environment is the right choice when the data is patient-level health or genomic data; when custodians will not or cannot release copies; when the science requires real pipelines or model training rather than counts; when outputs need independent disclosure control; or when GDPR, EHDS, or national data-residency rules apply. A useful test for sponsors: write down the sentence you would need a regulator or a data custodian to accept, then choose the architecture that makes that sentence true by construction. “The venue’s aggregation thresholds were correctly configured” is a weaker sentence than “the data never left the custodian and no record-level export path exists”.

Common pitfalls when evaluating the two

Three misjudgements recur in procurement. The first is vocabulary drift: vendors increasingly describe clean rooms as “TREs” and vice versa, so evaluate the architecture, not the label — ask where copies of the data reside during processing, who vets the people and projects, and who reviews the outputs. The second is underestimating governance debt: a clean room adopted for a quick pilot with tokenised data becomes the default venue for progressively richer datasets, until patient-level records sit in an environment whose trust model was designed for advertising measurement. The third is assuming aggregation thresholds equal disclosure control: a minimum cohort size blocks the crudest re-identification but not differencing attacks, repeated overlapping queries, or rare-condition leakage — risks that formal statistical disclosure control in a Trusted Research Environment is designed to catch. Each pitfall is avoided the same way: match the infrastructure’s design centre to the sensitivity of the data it will eventually hold, not the data in the first pilot.

What to do next

If you are evaluating collaboration infrastructure for real-world evidence, multi-site research, or data partnerships, start by classifying your use cases along three axes: data sensitivity (tokenised commercial data versus patient-level health data), analytical depth (counts and overlap versus pipelines and models), and custodian posture (willing to share copies versus custody-retaining). Anything landing in the second column of each axis points to a federated Trusted Research Environment. Then pressure-test any vendor — clean room or TRE — against the UK Biobank question: what, architecturally rather than contractually, prevents an approved user from walking record-level data out the front door? Platforms with a good answer will show you an airlock and a federation topology. Platforms without one will show you a policy document.

Frequently asked questions

What is the main difference between a data clean room and a federated TRE?

A clean room centralises copies of each party’s data in a neutral venue and restricts queries over the pooled set. A federated Trusted Research Environment sends the computation to each data source instead, so the data never leaves its custodian and only aggregate, disclosure-checked results are returned.

Are data clean rooms suitable for patient-level health data?

Generally not. Clean rooms were designed for tokenised marketing data and thin commercial trust models. Patient-level health data typically requires researcher accreditation, project approval, custodian oversight, and independent output control — the Five Safes governance that TREs implement and clean rooms do not.

Do clean rooms and federated TREs both avoid sharing raw data?

Both avoid exposing raw records between parties, but differently. A clean room still requires each party to upload data to the venue, which holds copies during processing. In a federated TRE no copy is contributed anywhere; each dataset stays inside its custodian’s infrastructure permanently.

Can a federated TRE do the overlap analysis a clean room does?

Yes. Federated cohort counting and overlap assessment are standard TRE workloads, often using privacy-preserving record linkage on tokenised identifiers. The reverse is not true — clean rooms cannot generally run genomics pipelines or train models across sites.

Which model fits the EHDS requirements for secondary use of health data?

The European Health Data Space requires secondary use to occur in secure processing environments with controlled outputs and prohibits record-level extraction. A federated TRE matches this design directly; an adtech-style clean room would need substantial re-engineering and governance additions to qualify.

Why did the UK Biobank incident change this evaluation?

Because the data left through an approved, policy-compliant workflow on a centralised platform, it showed that permission-based controls fail silently. Evaluators now ask whether bulk egress is architecturally impossible — the property federation provides — rather than merely forbidden by configuration.

Is a clean room ever the right choice for a pharma sponsor?

Yes — for commercial analytics such as measuring audience overlap with a media partner on tokenised, non-clinical data. For research on patient-level or genomic data, or any collaboration involving health-data custodians and regulators, a federated Trusted Research Environment is the appropriate infrastructure.