The extension for valuePeriod with additional constraints failed the validation check that requires the end date to be greater than start date

[bruce4000www]

I have a CarePlan resource with extension artifact-effectivePeriod. The extension is modified to set period.start cardinality to 1..1.
When submitting a CarePlan resource with period.start > period.end in the extension, there is no error generated in the validation result, which is unexpected.

CarePlan resource:

{
  "resourceType": "CarePlan",
  "status": "active",
  "intent": "plan",
  "subject": {
    "reference": "Patient/example",
    "display": "Peter James Chalmers"
  },
  "extension": [
    {
      "url": "http://hl7.org/fhir/StructureDefinition/artifact-effectivePeriod",
      "valuePeriod": {
        "start": "2017-06-02",
        "end": "2017-06-01"
      }
    }
  ]
}

Extension structure definition:
Download the artifact-effectivePeriod from simplifer.net here.
Modify the extension to restrict period.start cardinality to 1.

{
  "differential": {
    "element": [
      // Add following rule will make the checking end > start fail
      {
        "id": "Extension.value[x].start",
        "path": "Extension.value[x].start",
        "min": 1
      }
    ]
  }
}

The validation result:

{
  "resourceType": "OperationOutcome"
}

[mmsmits]

R4 expression of PER-1: start.empty() or end.empty() or (start <= end)",
R5 expression of PER-1: "start.hasValue().not() or end.hasValue().not() or (start.lowBoundary() <= end.highBoundary())"

Let's check if the R5 fix of the expression resolves this. And if so, fix the expression in the the StructureDefinitionCorrectionResolver.

Otherwise: debug.

[andrzejskowronski]

The correct element rule would be the following, as value[x] is not a valid element to have children like start

{
   "id": "Extension.valuePeriod.start",
  "path": "Extension.valuePeriod.start",
  "min": 1
}

This is an user error, but it highlights the need for us to complain about invalid profile definitions when we generate snapshot of them. I've added a ticket for the SDK snapshot generator and referred to this issue ;)

[bruce4000www]

@andrzejskowronski thanks for the solution, it works 👍. I am using Forge to configure profile, the generated the profile is value[x] instead of valuePeriod, it will be great if the Forge tool can be enhanced to cater for this :)

[andrzejskowronski]

Thanks for letting me know, I've notified Forge team about it then :)

[alexzautke]

We re-discussed the issue. The snapshot is correct, see https://fire.ly/blog/type-slicing-in-fhir-r4/ for reference. We'll investigate what's going on in the validator.

[ewoutkramer]

As @andrzejskowronski mentioned, it's an exceptional circumstance where the author is walking into an element, but the value[x] is not renamed in the diff to valuePeriod. However, the typeRef is reduced to only contain Period. The snapshot generator copies some/most of the children into the value[x] element, without creating a typeslice, which I think is correct, since there's only one type left, so we don't need a slice. However, this constraint, which is on the root of the Period type, is not copied.

In the validator, we then do the following check:

• If there is > 1 type -> check based on TypeRef
• If not, and there are children -> check based on children
• Check based on TypeRef.

So, in this case, where there are at least some children and 1 type, we're not running the referenced Period type, but are dependent on the children. These don't contain the necessary invariant however.

I think the bug is in the snapgen, but that's hard to fix at this moment (we should still do it), and so this PR will contain a fix in the schema generator.