The IG prohibits string-based membership testing, but there may be cases where it is appropriate. For example, Address.state is represented as a string in FHIR and is often bound to value sets in regional IGs. Would we want to prohibit CQL such as Patient.address in NewEnglandStatesVS? If we do prohibit that, there is no alternative except to represent the ValueSet as a literal List<String> instead – which is not ideal.

The argument that it should be avoided because it does not consider the code system also does not apply to any FHIR element that is a code type (like most status elements), since they don't carry a system either.

For this reason, I think the prohibition should be relaxed to SHOULD NOT (rather than MUST NOT). I think the key point is that if an element is represented as a CodeableConcept or Coding in FHIR, membership testing must be applied to the full element rather than the primitive code part – although I'm not sure what the best way to say that is.