Genomic Real-World Data: Linking DNA to Clinical Outcomes
Genomic real-world data is genetic information, such as whole genome sequences, exome data or panel test results, that is linked to clinical data collected during routine care rather than in a controlled trial. It lets researchers connect what is written in a person’s DNA to what actually happened to them: diagnoses, prescriptions, lab results and outcomes over years.
This guide covers where genomic real-world data comes from, why it is harder to work with than either genomic or clinical data alone, and how research teams can analyze it at scale without moving sensitive records around.
Why Genomic Real-World Data Matters Now
For most of the history of genetics, studies were built one cohort at a time: recruit participants, sequence them, follow them up. That produced deep but narrow datasets. Today, sequencing is part of routine care in areas such as oncology, rare disease and pharmacogenomics, and national programs have sequenced large populations with consent for research. The result is a growing body of genomic data that sits next to years of electronic health records.
Linking the two answers questions that neither can answer alone. Which variants predict a poor response to a common drug? How do outcomes differ between patients with the same diagnosis but different tumor mutations? Which rare variants show up in people whose symptoms took years to diagnose? These are real-world questions, and they need real-world data.
The same linkage raises the stakes on privacy. A genome is the most identifying data a person has, and it says something about their relatives too. The May 2026 UK Biobank incident, where approved researchers moved data out through a normal workflow, is a reminder that access models designed for less sensitive data do not hold up for genomic records.
Where Genomic Real-World Data Comes From
Clinical Sequencing
Hospitals and reference labs sequence patients to guide diagnosis and treatment. Tumor panels in oncology, exome and genome tests for rare disease, and pharmacogenomic tests before prescribing are the most common examples. These results are generated for care, so they come with rich clinical context but variable formats.
National and Population Programs
Public programs such as UK Biobank and All of Us combine sequencing with consent for research and linkage to health records. They provide scale and consistent processing, and they are often the reference datasets other studies compare against.
Biobanks and Disease Registries
Hospital and academic biobanks hold samples with consent for research, some already sequenced. Disease registries add longitudinal follow-up for specific conditions.
Pharma and Trial Archives
Clinical trials increasingly collect samples for genomic analysis. When those participants are linked to post-trial health data, the archive becomes a real-world resource. For more on the broader categories of real-world data, see these real-world data examples.
Why It Is Hard to Work With
Scale and Format
A single whole genome can occupy tens of gigabytes in raw form. Variant files are smaller but still large, and different pipelines produce different outputs. Clinical data, meanwhile, arrives as coded records, free text and lab values in many local formats.
Linkage
Genomic and clinical records are often held by different teams, in different systems, under different identifiers. Linking them reliably, without exposing direct identifiers, is a project in its own right.
Harmonization
To combine data across sites, both halves need common standards. On the clinical side, that usually means mapping to the OMOP CDM; the OHDSI community has also worked on representing genomic findings alongside it. On the genomic side, Global Alliance for Genomics and Health (GA4GH) standards such as VCF for variants and Phenopackets for phenotypes are widely used. Without consistent data harmonization, cross-site analysis produces noise.
Governance and Consent
Consent terms differ between clinical testing and research programs. Laws differ between countries. Many custodians cannot legally send genomic data across borders at all, which rules out pooling everything in one place.
Centralized Pooling Versus Federated Analysis
There are two broad ways to analyze genomic real-world data from several sources.
| Dimension | Centralized pooling | Federated analysis in a TRE |
|---|---|---|
| Where the data sits | Copied into one central store | Stays with each custodian |
| Data transfer | Large genomic files moved and duplicated | Analysis code moves; data does not |
| Cross-border rules | Often blocks participation | Each site stays within its own jurisdiction |
| Consent and withdrawal | Must be tracked across copies | Managed once, at the source |
| Privacy risk | Grows with every copy | Record-level data is never exposed outside the source |
| Who can join | Only custodians able to share data | Any custodian able to host analysis |
For genomic data, the federated route is often the only one that lets the most valuable sources take part at all.
How Federated Genomic Analysis Works
In a federated model, each custodian runs a Trusted Research Environment (TRE) where its genomic and clinical data is stored and harmonized. An approved researcher writes one analysis, for example a genome-wide association study or a survival model stratified by variant, and sends it to each participating site. The analysis runs locally, and only aggregate results come back for combination.
Three things make this practical:
- Common data models at every site, so the same code runs everywhere without rewriting.
- Workflow tools built for genomics, so pipelines for alignment, variant calling and annotation run reproducibly inside each environment.
- Output controls, so every result is checked for disclosure risk before it leaves, using an automated airlock with human review where needed.
The principle is simple: data never leaves the source, but the science still happens across all of it. A Federated Trusted Research Environment extends this to many custodians at once, each keeping control of its own records.
What Research-Ready Genomic Real-World Data Looks Like
Not every linked dataset is ready for analysis. Research-ready genomic real-world data tends to share a handful of qualities, and checking for them early saves months later.
- Consistent processing: variants called with a documented pipeline and reference genome version, so results from different batches can be compared.
- Clear provenance: each genomic file traceable to the sample, the sequencing run and the participant pseudonym it came from.
- Coded clinical data: diagnoses, medications and lab results mapped to standard vocabularies rather than held as local codes or free text.
- Known follow-up: a record of how long each participant has been observed, so missing outcomes are not mistaken for healthy ones.
- Documented consent: the scope of research use recorded for each participant in a form software can filter.
- Quality metrics: coverage, call rates and sample-level checks stored alongside the data.
Datasets that meet these tests can join a federated study quickly. Datasets that do not will need work first, and it is better to find that out before a study timeline depends on them.
Common Use Cases
- Pharmacogenomics: linking variants to drug response and adverse events recorded in routine care.
- Oncology: comparing treatment outcomes by tumor mutation profile across hospitals.
- Rare disease: finding enough patients with the same variant across many sites to study a condition at all.
- Target discovery: using human genetic evidence linked to clinical outcomes to prioritize drug targets.
- Population health: understanding how genetic risk interacts with real-world factors over time.
Privacy and Regulation
Genomic data is treated as special category data under GDPR and is protected under several US state laws and sector rules. Pseudonymization reduces risk but does not remove it, because a genome can be re-identified by comparison with other genetic data. That is why access models matter as much as de-identification. For a broader view of the rules, see this guide to genomic data privacy regulations.
Practical Next Steps
- Inventory what you hold. List genomic datasets, their formats, the clinical data they can link to and the consent that covers them.
- Agree standards early. Choose a common clinical data model and genomic formats before the first multi-site study, not after.
- Solve linkage once. Set up a pseudonymized linkage service so each new project does not rebuild it.
- Choose an access model that fits genomic risk. Favor analysis in place over data release.
- Start with one question. Pick a pharmacogenomic or oncology question with clear value and run it across two sites to prove the approach.
Frequently asked questions
What is genomic real-world data?
It is genetic data such as genome, exome or panel results linked to clinical data collected in routine care, including diagnoses, prescriptions, lab results and outcomes.
How is it different from genomic trial data?
Trial data is collected under a fixed protocol for a specific question. Genomic real-world data comes from routine care and population programs, so it is broader and longer term but less uniform.
Why is genomic real-world data hard to share?
Genomes are highly identifying, files are large, consent terms vary and many jurisdictions restrict transfer. Together these make pooling difficult or impossible.
Which standards help harmonize genomic and clinical data?
The OMOP Common Data Model is widely used for clinical data, and GA4GH standards such as VCF and Phenopackets are widely used for genomic and phenotype data.
Can genomic real-world data be analyzed without moving it?
Yes. In federated analysis, approved code runs inside each custodian’s Trusted Research Environment and only aggregate, checked results are returned.
What are the main use cases?
Pharmacogenomics, oncology outcomes research, rare disease studies, target discovery and population health research.