MedVertical
Open menu

BlogFHIR operations

Who Decides Your FHIR Data Is Good Enough?

A supplier can run FHIR data-quality checks. The team accepting the delivery needs agreed criteria and evidence it can inspect.

Conceptual checking and acceptance workflow: two available data sources and one missing source lead to checks, a report with passed checks and a warning, and a responsible person with decision options. A dashed feedback loop links evidence, re-checking and analysis back to the data source.
5 min read
fhirdata-qualityquality-measuresacceptanceevidence

A data supplier delivers a FHIR dataset and a green quality report. The team responsible for using the data decides whether it is good enough for reporting, research or another exchange. That decision needs agreed requirements and findings the team can inspect.

The supplier can execute the checks. The receiving team makes the acceptance decision using agreed criteria and needs access to the findings. That responsibility matters whether the checks run inside the FHIR server, in a separate service or in the recipient's own tooling.

A 100 percent result can leave a delivery incomplete

Consider a hypothetical delivery from three expected laboratories. Two deliver 200 Observation resources between them. Every received resource contains the required code field.

CheckResult
Required code field present (200 resources)200/200 · 100%
Source deliveries (three expected)2/3

Both results matter: the received resources pass the code-field check, while one expected source is missing. Your team must define the expected sources and delivery windows and reconcile them with what arrived. The resources alone cannot reveal a missing delivery. These numbers are illustrative, not a measured Records result.

This is where Records helps: it checks selected FHIR data and shows which rules were applied, which issues were found and what could not be evaluated. After a correction, the team can run the checks again. These findings provide evidence for its decision. Your team defines the expected deliveries and decides which remaining deviations are acceptable.

An acceptance report should distinguish the expected delivery, the received data and the evaluated data. Put excluded and not-evaluable cases beside the score. The report needs to show which requirement was checked and which data the result covers.

Running the checks and accepting the data are separate responsibilities

A current example comes from the United States. In the program for its Health Innovation Summit on October 6, 2026, the National Committee for Quality Assurance (NCQA) describes the shift from manual, point-in-time verification to continuous, automated data-quality checks closer to the source.

Digital quality measures describe care quality. Both the measure calculation and the underlying data need their own checks.

The prevalidation program within NCQA Data Quality Solutions assesses whether data-quality rules are implemented correctly in software. It validates neither an organization's data nor its results. For your acceptance decision, the question remains whether the particular delivery meets the agreed requirements.

Independence starts with control over the criteria

A supplier knows its platform and can investigate findings close to their source. Agree the acceptance basis with that supplier before delivery: expected sources, periods, data selection, profile and terminology versions, quality rules and thresholds.

The supplier can execute the checks. The receiving team needs to inspect the rules, exclusions and findings, and request a re-check. The person responsible for acceptance decides which remaining deviations are acceptable for the intended use, taking applicable reporting or audit obligations into account.

Record approval for changes to scope, thresholds or the treatment of findings. Otherwise, the score can stay green while the meaning of the result changes.

Control over criteria and inspectable results make acceptance reviewable. They do not, by themselves, establish an independent review. Checks inside a FHIR server can supply this evidence. Independence also depends on who reviews the results and what organizational or technical dependencies affect that review. Depending on the intended use and consequences of a wrong decision, additional review by another team, another rule implementation or verification against the sources may be needed. That check also needs a clear scope and visible limits.

Five questions to ask before signing off

Use these questions in the next delivery review or supplier discussion:

  1. What did we expect, and what arrived? Reconcile agreed sources, periods and populations. Record any sampling or access restrictions.
  2. What exactly produced this score? Name the rules, versions and thresholds. Show how many resources passed, how many were evaluated, and what was excluded or not evaluable.
  3. Can we investigate the findings? Retain appropriate links to affected resources and fields, rule context and the responsible team. A summary percentage should lead to an explanation.
  4. What would demonstrate a correction? Re-check the corrected data under comparable conditions and record the result. A smaller data selection or disabled rule does not demonstrate a correction at the source.
  5. Who can accept the remaining deviations? Name the person responsible for acceptance and retain the justification. A passed check cannot settle every question about intended use.

These questions apply across platforms. Supplier comparison also needs compatible data scope and evaluation conditions; using the same rule names alone does not make two scores comparable.

What Records can contribute to the review

Records supports three practical steps in that review:

  1. Check the data. Validate selected FHIR resources and evaluate data quality using versioned indicators. The defined scope, rules, thresholds and evaluated population establish what the results cover.
  2. Investigate the findings. Use individual findings and recorded run context to understand where checks failed and which cases could not be evaluated. This gives the team specific issues to discuss with the supplier.
  3. Re-check corrected data. Run the relevant checks again with comparable scope, rule versions and dependencies. Review the new findings to see whether the identified issues remain.

These steps support your team's acceptance decision. Expected deliveries and acceptable deviations still need to be agreed. A repeated check needs the corresponding data and dependencies that influence the result. Clinical correctness requires its own evidence beyond the selected checks.

The three questions every FHIR team should answer offer a related framework. After a mapping or profile change, a comparable check in operation helps establish what the new delivery shows.

For the next exchange, start with one receiving requirement and decide what evidence would satisfy it. Ask the supplier to show how its result connects to that requirement, including what remains unknown. This gives your team an acceptance decision it can justify and revisit after a change.

Written with AI assistance.

André Sheydin

About the author

André Sheydin

André is the founder of MedVertical and a product and design lead based in Cologne. He has spent more than 25 years shaping digital products, platforms, and design systems across complex domains, including healthcare, pharma, automotive, and SaaS. His work focuses on turning technical requirements into product structures that teams can actually build and operate.

Your next step

What evidence is missing from your next delivery?

Describe one data delivery and an acceptance question. Agree permitted test data, a bounded scope and the relevant profile versions with our team.