FHIR Proof of Concept in Pharma: A Practical 90-Day Plan

A FHIR proof of concept in pharma should test one narrow, valuable use case, such as pulling structured clinical data from partner hospitals for a feasibility question, against success criteria agreed before any code is written. Ninety days is enough to prove whether Fast Healthcare Interoperability Resources (FHIR) can deliver usable research data, as long as the scope stays small, the data partners are committed and the governance model is settled on day one.
Why pharma teams are running FHIR pilots now
FHIR has moved from a hospital integration standard to a practical route for research data. Regulators and health systems in the US and Europe increasingly expect electronic health records to expose standard APIs, and the European Health Data Space (EHDS) is pushing toward common formats for secondary use of health data. For pharma, that creates a chance to reach clinical data faster than through bespoke extracts.
The catch is that FHIR was designed for exchanging records about individual patients in care, not for population research. A pilot is the cheapest way to find out where FHIR helps a given organization and where it needs to be paired with a research model such as OMOP. Teams that skip the pilot often commit to a platform before they know whether their partners can supply the resources they need.
Choosing the right use case
The single biggest predictor of a successful proof of concept is the use case. Good candidates share three traits: a clear business owner, data that partners already hold in structured form, and an outcome that can be measured in weeks.
Strong candidates
- Trial feasibility. Counting patients who meet inclusion criteria at partner sites using conditions, medications and lab results.
- Real-world data enrichment. Adding structured clinical detail to an existing observational study.
- Safety signal follow-up. Retrieving defined clinical facts for a cohort already under review.
Weak candidates for a first pilot
- Anything that depends on free-text notes or imaging.
- Use cases that need many partners to join before any result is possible.
- Regulatory submissions, which raise validation questions too early.
Who needs to be on the pilot team
A FHIR pilot touches clinical, technical, legal and commercial teams, so a small cross-functional group works better than a purely technical one. Keep it to the people who can make decisions.
- Business owner. Usually from clinical operations, epidemiology or real-world evidence. Owns the question and the final decision.
- Data engineer. Connects to endpoints, extracts resources and builds the mapping.
- Terminology lead. Handles code systems such as SNOMED CT, LOINC and RxNorm, and local codes that need mapping.
- Privacy and legal lead. Agrees access terms, output rules and data protection assessments with each partner.
- Partner contact. A named person at each site who can unblock endpoint access and answer data questions quickly.
Weekly 30-minute check-ins with this group catch blockers early, especially access delays, which are the most common reason pilots overrun.
The 90-day plan
Split the pilot into three phases of roughly a month each, with a go or no-go gate at the end of each one.
Days 1 to 30: scope, governance and access
Write a one-page charter: the question, the data partners, the FHIR resources required (often Patient, Condition, MedicationRequest, Observation and Encounter), the success criteria and the decision the pilot will inform. Settle governance in parallel. Agree who approves access, what leaves each site and how results are checked before release. Confirm that each partner’s FHIR endpoint supports the resources and search parameters you need, and whether bulk export is available for population-level pulls.
Days 31 to 60: build and map
Connect to the first partner, extract a test cohort and map the FHIR resources to your analysis model. Many research teams convert FHIR to the OMOP Common Data Model at this stage because OMOP is built for cohort analysis while FHIR is built for exchange. Track data quality as you go: missing codes, local terminologies and inconsistent units are the usual problems. Our explainer on FHIR, HL7 v2 and CDA covers why partners may still send some data in older formats.
Days 61 to 90: run, measure and decide
Run the use case end to end, compare results against a known baseline where possible, and score the pilot against the charter. Present findings as a decision: scale, extend the pilot or stop. Include the cost and effort per partner, because that number drives the business case more than any technical result.
Data quality checks to run during the pilot
FHIR guarantees a common structure, not common content. Run a short set of checks on every extract so problems surface in week five rather than week twelve.
- Completeness. What share of patients have each required resource, and which fields are empty most often?
- Coding. What share of codes use a standard terminology, and how many are local codes that need mapping?
- Units and values. Are lab results in consistent units, and are values in plausible ranges?
- Dates. Are event dates present, in order and inside the expected period?
- Duplicates. Do the same events appear more than once across encounters or feeds?
Record the results per partner. They become the baseline for deciding whether a partner is ready to join a scaled program.
Success criteria that hold up
Vague goals such as “show that FHIR works” produce vague outcomes. Agree measurable criteria before the build starts.
- Share of required data elements available through FHIR at each partner.
- Share of codes that map to standard terminologies without manual work.
- Time from approved request to analysis-ready data.
- Agreement between FHIR-derived counts and a reference count from the same site.
- Effort in person-days to onboard one additional partner.
Centralized pilot versus federated pilot
A key design choice is whether data is copied to a central environment or the analysis goes to each partner. The answer shapes governance, timelines and what partners will agree to.
| Dimension | Centralized FHIR pilot | Federated FHIR pilot |
|---|---|---|
| Where data sits | Copied to a sponsor or vendor environment | Stays at each partner; data never leaves the source |
| Partner approval | Data sharing agreements per partner, often slow | Approval for analysis and outputs, usually simpler |
| Time to first result | Fast once data arrives, slow to get data | Faster access; setup needed at each node |
| Privacy risk | Grows with every copy | Limited to approved aggregate results |
| Scaling to more partners | Each partner repeats the sharing process | Each partner adds a node; the analysis is reused |
| Fit with research models | Convert to OMOP centrally | Convert to OMOP at each site, run one query everywhere |
For pilots involving hospitals that are reluctant to export records, a federated design often removes the main blocker. A Federated Trusted Research Environment runs the analysis inside each partner’s environment and returns only results that pass an output check.
Common pitfalls
Most stalled pilots fail for organizational reasons rather than technical ones.
- Assuming every endpoint is equal. Partners may support different FHIR versions, profiles and search parameters.
- Ignoring terminology. FHIR carries codes but does not fix local coding habits. Budget time for mapping.
- Leaving governance to the end. Access approval can take longer than the build.
- No owner for the decision. Without a named business owner, a good pilot still ends with no next step.
- Treating FHIR as the analysis model. FHIR is excellent for exchange; cohort analysis is easier in a research model such as OMOP CDM.
What to do after the pilot
If the pilot meets its criteria, the next step is a repeatable pattern rather than a bigger pilot. Document the resource list, mapping rules, governance steps and output checks so each new partner follows the same path. Decide where conversion to a research model happens, and set a target for onboarding time per partner. If the pilot falls short, the scorecard tells you why: missing data, mapping effort or access delays each have different fixes.
A well-run FHIR proof of concept gives pharma leaders a clear answer about cost, speed and data quality, and a tested route to clinical data that respects each partner’s rules.
Frequently asked questions
How long should a FHIR proof of concept in pharma take?
About 90 days is typical for one use case and one or two partners: a month for scope and governance, a month to build and map, and a month to run and decide.
Which FHIR resources does a research pilot usually need?
Most start with Patient, Condition, MedicationRequest or MedicationStatement, Observation and Encounter. The use case decides the final list.
Should we convert FHIR data to OMOP?
Often, yes. FHIR is designed for exchanging records, while OMOP is designed for cohort analysis. Many teams use FHIR to access data and OMOP to analyze it.
Do partners have to send us their data?
Not in a federated design. The analysis runs inside each partner’s environment and only approved aggregate results come back.
What is the most common reason pilots stall?
Governance and access delays. Agreeing approvals, output rules and responsibilities at the start does more for the timeline than any technical choice.
How do we measure success?
Use criteria set before the build, such as data element coverage, mapping rate, time to analysis-ready data, agreement with a reference count and effort to add a partner.
