FHIR and human readability

Question

I am a bit confused about human readability with regards to FHIR. Here, FHIR claims that human readability is required for each and every resource. However, here, it says that any resource that is domain resource may include human-readable narrative. It also seems that it is possible to have a valid (valid in the sense that the FHIR jar validator validates it) resource that does not include narrative. So, which is which?

I am asking this because I am making a bit of comparison between CDA and FHIR. In CDA, each section must contain text for human readability. In FHIR, I guess the equivalent is composition.section.text but even here, text is not a must as it can contain just an entry, which can just reference another resource that may not contain any narratives.

Answer 1

The assertion in the v2 comparison is too strong. This was once true, but we've since softened the rules. (If you can submit a change request for us to fix that, it would be much appreciated.)

The general gist is that narrative is strongly encouraged because there's almost always a chance that some consuming system downstream at some distant time in the future might not know how to parse all of the discrete information and will need the narrative to fall back to for human interpretation. However, there are use-cases such as ultra-low bandwidth exchange, passing data to a decisions support engine that won't persist the data, etc. where narrative doesn't make sense. As well, two resources (Binary and Bundle) don't support narrative at all.

In a FHIR document, I believe it's technically legal (though probably not useful) to have sections where there's no narrative. The CDA on FHIR profile will almost certainly prohibit this for "clinical" documents. Unfortunately, we can't easily do this with an invariant in the base spec because invariants can't reach across resources (to check that the resource referenced by a section without text points to a resource with text).