At 9:14 on August 21, a lab records a creatinine result of 1.4 mg/dL. The record contains more than the number. It carries the lab’s reference range, a high flag, the collection time, and the identity of the source that produced it.
Then the result travels. A receiving system accepts it. The value and collection time appear in the new chart. The reference range does not. Neither do the high flag or the source. Somewhere in the exchange, the transaction still earns a green check.
- Creatinine
- 1.4 mg/dL
- Collected
- Aug 21 · 09:14
- Local reference
- 0.70–1.30
- Interpretation
- High
- Author / source
- Lab A
- Creatinine
- 1.4 mg/dL
- Collected
- Aug 21 · 09:14
- Local reference
- Not carried
- Interpretation
- Not carried
- Author / source
- Not carried
Illustrative reduction only. This is not a conformance claim about a named implementation guide.
The number survived. Some of what made the number legible did not. This is not a real patient story, and it is not an accusation against a particular standard or product. It is a deliberately plain example of a problem that our usual definition of success can miss.
The distinction
Delivered is not the same as preserved.
A “successful exchange” often collapses three separate claims into one. The first istransport: did the connection hold, did the message arrive, and did the destination acknowledge it? The second is conformance: did the data satisfy the computable rules that apply to its structure, cardinality, terminology, profile, and implementation context?
Those checks are necessary. They are also bounded. HL7’s own FHIR guidance says validation methods can check computable aspects of conformance, but are incomplete. They cannot prove every narrative requirement, including whether all clinically important content is present; some business rules live outside the base specification altogether.1
The third claim is fidelity: did the meaning, completeness, and provenance of the source survive the trip? Fidelity is not a property of one message in isolation. It is a comparison between what left and what arrived.
Did the destination receive it?
Did it satisfy the applicable computable checks?
Did meaning, completeness, and provenance survive?
This does not require shrinking the formal meaning of interoperability. The 21st Century Cures Act sets a much higher goal than simple delivery: secure exchange and use without special effort, complete access, exchange, and use of electronically accessible health information, and no information blocking.2The gap appears when an operational green light measures something narrower than that goal and we stop asking questions.
Three tests
Fidelity has three parts.
Meaning. Does the destination still know what a fact means, not merely the characters used to represent it? At the base FHIR Observation level, interpretation and reference range are optional elements. Yet HL7 notes that interpretation may be necessary to understand some numerical results and that reference ranges help evaluate their significance. A profile or implementation guide can impose stronger expectations.3
The problem is not limited to missing fields. HL7 also notes that similar observation information can be represented in multiple ways, creating a challenge for computing semantic equivalence and interpreting results safely. That is one reason profiles and implementation guides matter: they narrow the room for two systems to mean different things with data that looks similar.4
Completeness. Did everything required for the intended use make the trip, or only the subset covered by the exchange? ONC describes USCDI as a baseline standardized set of data elements, updates it over time, and maintains USCDI+ for more specific program and use-case needs.5Completeness is therefore not “send every field in every context.” It is a test against the purpose the receiving record is expected to serve.
Provenance. Can someone tell where the data came from, who or what changed it, and which transformation produced the current version? FHIR’s Provenance resource was designed to record the entities and processes that create, revise, or influence a resource. Its guidance explicitly covers imports and middleware transformations because origin is part of how authenticity, trust, and reproducibility are assessed.6
Not a verdict on standards
The tools are already here.
FHIR, profiles, implementation guides, USCDI, and provenance structures are not the villains in this story. They are what make the fidelity question testable. They give an exchange a declared shape, a vocabulary, and a set of expectations that can be compared with what the destination actually stored and displayed.
Fidelity is not a replacement for interoperability. It is the missing verification step between “the interface worked” and “the receiving record can safely do the job we expect of it.” There may never be one universal fidelity score. Different uses require different information. But the questions are concrete: What arrived? What changed? What disappeared? Can the result still be traced to its source?
A record that arrives is not necessarily a record that still means what it meant at the source.