Where should results-based ICD-10 / CPT II codes be represented in a FHIR Observation?

Question

My understanding is that, in general, for standard EHR use-cases, FHIR encodes ICD-10 diagnosis codes in Condition resources on a Patient, which may be linked to an Encounter that established the Condition. That Encounter may refer to one or more Observations that were used to diagnose the condition. This is of course helpful for long-term tracking of the patient's condition and care plan.

In my company's case specific case, however, we perform transactional diagnostic services on behalf of other payers/health systems (who are responsible for the actual care of the patient), and so use a more encounter/observation-oriented model rather than patient-oriented model for our data systems (databases, APIs, and data exports), much as a reference lab or 3rd party imaging provider might. This however leaves us with a question of how best to model CPT II procedure codes and ICD-10 diagnosis codes within the Observation resource.

For example, for a patient with Type 2 diabetes, we might perform a diabetic retina exam on both eyes and receive the following diagnoses from our ophthalmology vendor:

• Left eye: Type 2 diabetes mellitus with mild nonproliferative diabetic retinopathy without macular edema (ICD-10: E11.3292; CPT II: 2022F)
• Right eye: No diabetic retinopathy found. Glaucoma: Optic nerve cupping (ICD-10: E11.9, H40.011; CPT II: 2023F)

Based on https://chat.openai.com/c/95831f6d-befc-4eaa-aa7c-948772308753 we're led to believe that the Observation.code field could be used to support both the CPT II code and the ICD-10 code. But it's unclear whether the Observation.code field is really appropriate for results of the observation. We could instead use Observation.value[x]. valueCodeableConcept field, which is more suggestive of a "result" than an observation type, but we're not sure if it is closer to an "interpretation" than a result.

So, in cases like ours:

• Are ICD-10 diagnose codes generally encoded in the code, valueCodeableConcept, or interpretation fields?
• Are CPT II codes (which, unlike CPT I codes, include the results of the procedure) generally encoded in the code, valueCodeableConcept, or interpretation fields?

Answer 1

Either code or value are possible. The guidance that exists is here: https://build.fhir.org/observation.html#code-interop.

There are diverse perspectives (often use-case driven) about whether it's better to have code="diagnosis", value="diabetes" or code="diabetes", value="present".

Interpretation would not be an appropriate place to capture a diagnosis.

Answer 2

If you are aiming to stick with Observation, then valueCodeableConcept would be the best choice. code is better suited to identify the type of Observation (e.g., a LOINC code to represent the eye exam) and interpretation is better suited for flagging whether the value falls in or out of some accepted range/normal state (e.g., abnormally high).

That said, it seems like you have options beyond just Observation. One would be ClinicalImpression; this resource is less mature than Observation or Condition, but it would allow you to convey the fact that you are performing some kind of assessment and that you made some kind of finding, which could result in a formal diagnosis down the line.

Or, you could use the Condition resource. If the expectation is that the entity who is truly responsible for the patient's care needs to confirm your findings, the Condition could be created with a verificationStatus of provisional or differential. You can also link evidence (via a Reference to any resource) to the Condition.