FHIR Chat · Observation.value non-specific value, how to represent · implementers

Stream: implementers

Topic: Observation.value non-specific value, how to represent


view this post on Zulip Samantha Wang (Jan 04 2019 at 22:37):

Question about how folks might be handling the creation of Observation FHIR resources for which the patient source data does not have the necessary information to specify an exact value, but we know one exists.

For example, a system may have a result document for a FIT-DNA test. How would someone construct an Observation FHIR resource that conveys that there is a result (i.e. the result is not null), but not specify the value, e.g. "positive/negative". This could happen for data that is unstructured.

One suggestion could be to leverage the "boolean" value type, and set it to "Y"; however, a concern would be mis-interpretation. Are there other suggestions?

Thanks for anyone's thoughts.

view this post on Zulip Mikael Rinnetmäki (Jan 05 2019 at 07:16):

Would you be able to use the valueString, to store either the unstructured data or a value explicitly stating that the value is unknown due to the data being unstructured?

view this post on Zulip Lloyd McKenzie (Jan 05 2019 at 23:59):

It's possible to have an Observation instance where the value is omitted - because it's unknown, because the use-case doesn't allow it to be shared, because it's still in the process of being determined or any other reason.

view this post on Zulip Samantha Wang (Jan 08 2019 at 14:49):

@Mikael Rinnetmäki I like the idea of valueString "unknown". For certain consumers of FHIR, Observation.value cannot be null, for example certain Quality Measure calculation that rely on QDM attributes.

view this post on Zulip Lloyd McKenzie (Jan 08 2019 at 16:27):

That's really not a preferred solution. It would be better to omit the string value and instead include a data-absent-reason extension or something else that flags it as unavailable. (It also means that the profile that mandates Observation.value being present is probably overly tight, so getting the profile relaxed is the desired target endstate.)

view this post on Zulip Eric Haas (Jan 08 2019 at 16:51):

There already is a Observation.dataAbsernReason element for this use case. That is the preferred solution.

view this post on Zulip Lloyd McKenzie (Jan 08 2019 at 17:52):

Right - but that wouldn't help if Observation.value were mandatory - which is why you'd need to get the profile loosened first


Last updated: Apr 12 2022 at 19:14 UTC