§ 01
The situation
Western health systems are not short of data. They are short of answers.
Federated analytics has been the stated direction of Western health-data policy for over a decade. The data exists, in hospitals, registries and biobanks. The governance exists, in TRE frameworks. What does not yet exist, at scale and in production, is the capability to put one question across the federation and receive a defensible answer the same quarter.
FIG · 01·01
12+ years
Federated analytics has been policy direction across NHS, HDR UK, OHDSI and EHDS for over a decade.
FIG · 01·02
~215 TREs
Across Europe, trusted research environments hold the data. No single query surface reaches more than a fraction of them.
FIG · 01·03
18 months
Typical elapsed time from approved research question to a first cross-TRE answer in today's data-access model.
FIG · 01·04
0 sovereign LLMs
AI models used against national health data today overwhelmingly run on foreign cloud. Prompts and embeddings leave the country by default.
§ 02
The questions that matter
Three questions a national system needs to answer first.
Ranked by impact: the questions where a federated, sovereign-AI answer directly changes how the system spends, regulates, or plans. Every Western health system we speak to recognises these three — and none of them can yet answer one.
Q·01
HIGHEST IMPACT
Chronic disease control
Which patients are actually under control — and where aren't they?
Hypertension, diabetes and cholesterol drive the single largest share of Western disease burden. Every system has the records. Almost none can produce a national control map — by region, by practice, by cohort. The primary-care reinvestment case lives inside that map.
Q·02
FASTEST ROI
Medicines & devices · RWE
Are the things we pay for actually working in the real population?
National procurement, HTA bodies and payers commit billions on evidence from narrow trial populations. A federated engine lets the system re-ask the efficacy question on its own population, its own subgroups, its own comorbidities — annually, not once a decade.
Q·03
PREPAREDNESS
Population resilience
When the next shock comes, can we see it in hours — not quarters?
Pandemic response, drug-shortage response, winter-pressure response all depend on the same capability: put one question across every hospital and every primary-care system, receive a signed answer the same day. Today this is rebuilt from scratch each crisis. It should be a standing national capability.
§ 03
The outcome on week 13
Four capabilities the nation acquires — once, and keeps.
At the end of a first pilot, the programme has a federated, AI-queryable view of its own population. It is not a report. It is a standing capability — reused for every question that follows.
CAPABILITY · 01
See the map
A region-by-region view of control, outcomes and care gaps across the whole system. Where variation is largest. Where existing care is working.
“Where is outcome worst — and why?”
CAPABILITY · 02
Target the spend
Direct procurement, screening, and primary-care investment to the places where the evidence — not the lobbying — shows the need is greatest.
“Where does the next £100m of NHS or payer spend go?”
CAPABILITY · 03
Defend the system
A cited, replayable national picture is what unlocks Treasury business cases, EHDS participation, and public trust. Evidence with a signature on it.
“What is our case — with receipts?”
CAPABILITY · 04
Ask the next question
Once the engine is installed, the next national question takes weeks, not years. Medicines safety. Device post-market. Workforce. Cancer pathways.
“What do we want to know next?”
§ 04
How it works
Three steps. No new systems in hospitals. No data movement.
1
Weeks 1–2
CONNECT
Runner is deployed inside each TRE, hospital or registry. No schema changes. No data movement. No new infrastructure for the custodian to operate.
2
Weeks 3–6
TRANSLATE
Each source is mapped to OMOP CDM. A diabetes patient in Manchester and a diabetes patient in Rotterdam are the same shape to the federation.
3
Weeks 7–12
ASK
Researchers — and AI agents — put questions to the federation in plain language. The system compiles to signed UQL, dispatches, and returns aggregates. Never patient records.
FIG · 04·01 — FEDERATION TOPOLOGY
researcher / AI agent → question in natural language
compiled to signed UQL →
┄┄┄ NATIONAL BOUNDARY · DATA · MODEL · INFERENCE REMAIN INSIDE ┄┄┄
NODE · 01
Region A · TRE
hospital · OMOP
RUNNER + MODEL · IN-SITU
n = 2.4M
NODE · 02
Region B · TRE
primary care
RUNNER + MODEL · IN-SITU
n = 6.1M
NODE · 03
Region C · TRE
registry
RUNNER + MODEL · IN-SITU
n = 420k
NODE · 04
National biobank
consented
RUNNER + MODEL · IN-SITU
n = 500k
┄┄┄ EGRESS · AGGREGATE ONLY · NO PATIENT RECORDS · NO PROMPTS · NO EMBEDDINGS ┄┄┄
federation → pooled aggregate · signed · replayable · cited back to source
← uql://query/82c
§ 05
Sovereignty
Architecture, not policy. Enforced by the system, not promised by the vendor.
Unison is built so that a single patient record, a single prompt, and a single embedding cannot leave the national boundary — even if an operator tries. Sovereignty is structural, observable, and auditable by the programme.
TIER · 01
Data sovereignty
Runner executes inside each TRE. Patient-level records remain behind the custodian's firewall. The federation is a logical overlay, not a data lake.
TIER · 02
Model sovereignty
Inference runs inside the national boundary. Bring-your-own-model. On-prem LLM. National GPUs. No prompts, embeddings or patient context cross the border.
TIER · 03
Agent sovereignty
Every AI agent action is mediated by signed UQL. Governance approves what the model is allowed to ask — not only what data exists. Compliance becomes observable in the loop.
STANDARDS & FIT· OMOP CDM-native· OHDSI-aligned· FHIR-aware· MCP-native· BYO-model· On-prem LLM· Cyber Essentials Plus· DSPT-aligned· EHDS-aligned· 5 Safes-compatible
§ 06
Where we already work
Embedded in the infrastructure Western health systems are already building.
Programme names, not logo-wall. Each row states the relationship and the contribution — strictly.
HDR UK
Co-lead · federated analytics workstream
Working with Health Data Research UK and national NHS datasets to make queries executable across multiple TREs without moving patient-level records.
CONTRIBUTION
UQL execution across federated TREsDARE UK
Grant partner
Grant-funded partner on DARE UK programmes developing next-generation TRE interoperability. Contributing UQL and Runner as reference components.
CONTRIBUTION
Reference implementation · interoperability demonstratorsUKRI · TREvolution / TRExt
Academic partnership
UKRI-funded partnership with a UK medical school on a bidirectional OMOP ↔ CDISC demonstrator. Clinical-trial and real-world evidence speak the same vocabulary.
CONTRIBUTION
OMOP ↔ CDISC demonstrator · published methodsNHS England
Contractor · federated hospital programme
Contracted to deliver federated query execution across regional hospital data environments. Runner deployed inside each region; governance preserved.
CONTRIBUTION
Federated execution · regional sovereignty preservedEuropean Health Data Space
Aligned
Deployments aligned with the emerging EHDS framework: aggregate-only federated queries, audit trails, and cross-border governance without data movement.
CONTRIBUTION
EHDS-aligned deployment blueprintNone of the above are customer logos. Relationships are stated as they are.
§ 07
Next steps
Scope the pilot in two weeks. Answer on week 13.
The data is there. The policy direction is there. The missing piece is the engine that makes one question travel across the federation and return a signed answer. That is what a 12-week pilot installs.
01
Pick the question
Confirm the first federated national question — chronic-disease control, medicines RWE, or preparedness.
02
Pick the sites
Three to five sentinel TREs, trusts or registries to form the pilot federation.
03
Pick the date
Kick off within thirty days of scope agreement. Week-13 outcome on the programme director's desk.
Scope a 12-week pilot →
Read the sovereignty model →