Where health data becomes knowledge.
Insights in days, not quarters — ask questions your team couldn't afford to ask before.
Medical AffairsPharmacovigilanceHEOR & Market AccessClinical DevelopmentDevice & Diagnostic RWENational Systems & RWD
Research programme
3 mo · 3 ppl
was ~12 mo · 10 ppl
Time-to-answer
72 hrs
was 6+ months · CRO
Studies per quarter
dozens
was 1–3
Global OMOP network
160EHR & provider networks
32Biobanks & research DBs
30Registries
21Claims & commercial
17Federal & coordinating
01 · Use cases
What becomes possible when a query replaces a project.
Real-world evidence across the lifecycle — medical affairs, safety, HEOR and access. The same platform, same query surface, different questions.
MEDICAL AFFAIRS
→ Characterisation · comparative effectiveness
External evidence requests from HCPs & payers
TODAYEvery MSL question routes to a small RWE team with a months-long queue. By the time a cohort characterisation comes back, the congress window has closed.
WITH UNISONMedical teams self-serve: author a protocol, run it federated across connected biobanks, cite the artefact in the response. Days, not quarters.
PHARMACOVIGILANCE
→ PASS · signal refinement · label support
Post-market safety & signal evaluation
TODAYA safety signal surfaces. Investigation means commissioning three CROs with three different CDMs, waiting 6–12 months, and reconciling the answers.
WITH UNISONRun the same protocol across every connected data source in parallel. Heterogeneity you can inspect — not a black box you have to trust.
HEOR & MARKET ACCESS
→ Payer dossiers · HTA submissions · indirect comparisons
Burden of illness, treatment patterns, real-world outcomes
TODAYEach payer wants the evidence cut a different way. Each cut is a new data-vendor SOW, a new analytic spec, a new 3–6 months.
WITH UNISONOne protocol, many slices. Re-parameterise the query per geography or subgroup; the federation returns results in a format an HTA body can audit.
CLINICAL DEVELOPMENT
→ Trial feasibility · synthetic-control arms
Feasibility, protocol design & external controls
TODAYFeasibility queries bounce between site lists, EHR vendors and disease registries. Inclusion/exclusion iterations take weeks each.
WITH UNISONTest inclusion/exclusion against the federation in an afternoon. External-control cohorts arrive with provenance and reproducibility built in.
DEVICE & DIAGNOSTIC RWE
→ Billing-data RWE · custom taxonomies
Post-market surveillance for products hospital systems can't find
TODAYDevices and diagnostics sit as free-text line items in hospital billing systems — invisible to every off-the-shelf evidence tool built around drug codes.
WITH UNISONCustom taxonomies ingested as first-class OMOP concepts alongside SNOMED. Products become queryable; post-market safety and utilisation become routine.
The common thread: every engagement above sits on the same platform. A protocol written for one biobank re-runs unchanged across the federation — so when a single query becomes cheap, the space of questions worth asking changes.
02 · Product
The Rosetta Stone for health data.
Source systems keep their language. Researchers work in one. Nothing moves. Nothing copies. Everything is auditable.
BATCH ETL · THE OLD WAY
Build a pipeline. Materialise a copy. It's stale when it lands.
12–18 months per site. Re-run the whole thing every time the source changes.
VIRTUAL CDM · UNISON
Mappings are a spec, not a pipeline. Execute at query time.
Refine a rule, re-query. Source updates, virtual CDM updates with it. We harmonise by describing, not by copying.
01 · LAYER
Structure mapping
How source tables correspond to OMOP. Which column becomes condition_start_date. Deterministic, schema-level.
02 · LAYER
Semantic mapping
What each source value means in the standard vocabulary. Which SNOMED concept resolves "catheter-related bloodstream infection." Requires clinical judgement — where the expertise moat lives.
AI for speed. Humans for defensibility. Platform for governance — every decision logged, attributable, auto-documented to FDA guidance standards (CFR 21 Part 11-ready).
03 · Data mapping
One vocabulary. Many source dialects.
Every dataset speaks its own language — SNOMED at one hospital, ICD at the next, Read codes at a third, free text beyond that. Unison doesn't force sources to change their language. It translates them — at query time — into whichever common data model your downstream work demands: OMOP for observational research, FHIR for clinical-care interop, CDISC for regulated trials, PCORnet or Sentinel where that's the submission pipeline — each backed by the standard vocabularies the community already curates.
01 · LAYER
Structure mapping
Schema-level · deterministic
How source tables correspond to the target CDM's entities — OMOP by default, FHIR / CDISC SDTM / PCORnet / Sentinel where the downstream pipeline demands it. Which column becomes condition_start_date. Which foreign key becomes person_id. Written as transformation rules in SQL — inspectable, version-controlled, no surprises.
02 · LAYER
Semantic mapping
Concept-level · expert-curated
What each source value means in the standard vocabulary. Which SNOMED concept resolves “catheter-related bloodstream infection.” Which RxNorm code covers the same molecule as the site's local formulary. Clinical judgement — where the expertise moat lives.
03 · LAYER
Value mapping
Row-by-row · logged
Every source value that flowed through to a standard concept is traceable. Every mapping decision is logged with the person, the date and the rationale. A reviewer walks backward from a number to a concept to a code to the row that produced it — in whichever CDM the downstream step consumes.
WORKED EXAMPLE · ONE CONCEPT, EVERY SOURCE
“Type 2 diabetes” looks different in every system. Your query shouldn't have to.
Source
Vocabulary
Raw code
Label at source
US claims
ICD-10
E11.9
Type 2 diabetes mellitus w/o complications
UK primary care
Read CTV3
C10F.
Type II diabetes mellitus
EU hospital
SNOMED CT
44054006
Diabetes mellitus type 2
Disease registry
Local
DM_T2_CONF
Diabetes type 2 · confirmed
→ RESOLVES TO · STANDARD CONCEPT
Type 2 diabetes mellitus
SNOMED 44054006 · OMOP 201826 · FHIR Condition.code · CDISC MH domain
One query using descendants_of("Type 2 diabetes mellitus") fans out unchanged across all four sources — and flows on into whichever CDM the downstream step consumes. Every row's path, source code → standard vocabulary → target concept, is logged and re-traceable from the signed UQL artefact down.
VOCABULARIES · WHAT WE MAP FROM, WHAT WE MAP TO
Clinical conditions
SNOMED CT · ICD-10 / ICD-11 · Read · CTV3 · KCD · J-codes
Drug exposures
RxNorm · ATC · NDC · DM+D · DCI
Lab results
LOINC · local-catalogue harmonisation
Procedures
CPT · HCPCS · OPCS-4 · ICD-10-PCS
Adverse events
MedDRA · WHO-ART · preferred-term harmonisation
Rare disease
Orphanet · OMIM · HPO · ICD-11 rare-disease chapter · ERN codes
Imaging
DICOM · RSNA RadLex · LOINC radiology · BI-/PI-/LI-RADS
Omics & genomics
HGNC · HGVS · Ensembl · UniProt · GA4GH VRS · G-CDM
Devices
UDI · GMDN · manufacturer SKUs · lot-level traces
Custom taxonomies
Customer-defined concept sets · biomarker panels · disease-activity scores
Units & timing
UCUM units · temporal windows · washout periods
Common data models
OMOP CDM · FHIR R4 · CDISC SDTM / ADaM · PCORnet · Sentinel · i2b2
AI for speed. Humans for defensibility. Unison proposes mappings with AI and compresses the concept-matching work that used to cost weeks per site. Every proposal is reviewed by a clinical or data-engineering lead before it lands; every decision is attributable to a person and a moment. The mapping is a spec you can read — not a model you have to trust.
04 · How it works
Source DB → Runner → Virtual Model → CDM Concepts → Query
01
Source DB
Lives where it always lived. No copy, no centralisation.
02
Runner
Docker container inside the custodian's environment. Stays resident. Data never leaves.
03
Virtual Model
Source tables bridged to OMOP entities. Extraction rules in SQL.
04
CDM Concepts
Source values mapped to SNOMED, LOINC, RxNorm, ICD-10.
05
Query · UQL
Author once. Fans out unchanged across public biobanks, hospital data, national registries and pharma EHRs.
UQL · UNISON QUERY LANGUAGE
One query. Every biobank. Aggregate-only by design.
Authored by a researcher (or an AI agent via MCP). Compiled to a deterministic, aggregate-only IR. Cannot express a patient-level query — exfiltration is structurally impossible, not merely prohibited.
Reproducible
citable, replayable artefact
Secure
aggregate-only IR
Interoperable
OMOP concepts, not columns
Scalable
map once, query federation
# unison-mcp · agent session
> "14-day adverse events after device X vs SNOMED comparator."
→ UQL compiled · federated across 2 biobanks
cohort_a 14,228
cohort_b 11,091
Δ rate 0.7% · 95% CI 0.4–1.1
# citable artefact · uql://query/e3f2a1
05 · Proof
Built on the standards the industry already audits against.
OMOP-native
Every query authored against the same common data model your biostatisticians already know. No bespoke schema, no lock-in.
Federated by default
Analyses travel to the data. Custodians keep control. Aggregate-only results return by design — not by policy.
Audit-ready
Every mapping decision, every query execution, every result is logged, attributable, and reproducible from the UQL artefact up.
Device manufacturer
National health data body
Health-data research org
Primary-care registry
Research university
Data analytics partner
Clinical research lab
EHR data partner
OMOP consultancy
EU data partner
Applied-research lab
Space-medicine programme
Accreditation:· Accredited platform provider· Cyber Essentials Plus· DSPT-aligned· CFR 21 Part 11-ready· EHDS-aligned
Scope a pilot in 2 weeks · first insights in 10 days