tx.fhir.org SNOMED ECL bugs: memberOf sentinel leakage and cardinality counting

[jmandel]

Summary

While testing SNOMED ECL expansion against https://tx.fhir.org/r4 (FHIRsmith 0.9.6 from /metadata), I found several ECL behaviors that appear incorrect or incomplete.

Checklist

• Bare memberOf expressions such as ^700043003 return fabricated inactive code 4294967295.
• Bare memberOf over a non-reference-set concept such as ^404684003 also returns 4294967295 instead of rejecting or returning empty.
• Wrapped memberOf operands such as ^(<<900000000000455006) are rejected as unsupported.
• Attribute cardinality appears to count raw relationship rows rather than non-redundant matching attributes.
• Add regression tests for the above cases.

References:

• SNOMED ECL memberOf: memberOf returns concepts from referencedComponentId for active reference set members, for reference sets whose referenced components are concepts: https://docs.snomed.org/snomed-ct-specifications/snomed-ct-expression-constraint-language/behaviour-specification-with-examples/6.1-simple-expression-constraints
• SNOMED ECL cardinality: attribute cardinality is evaluated over non-redundant matching attributes, not raw relationship rows: https://docs.snomed.org/snomed-ct-specifications/snomed-ct-expression-constraint-language/behaviour-specification-with-examples/6.3-cardinality

Bug 1: memberOf returns fabricated sentinel code 4294967295

Bare memberOf expressions return a one-code expansion containing inactive code 4294967295. This happens both for apparent reference-set concepts and for concepts that are plainly not reference sets.

Reproducer:

for ECL in '^700043003' '^900000000000497000' '^404684003'; do
  printf '%s\t' "$ECL"
  jq -nc --arg ecl "$ECL" '{
    resourceType:"ValueSet",
    compose:{include:[{
      system:"http://snomed.info/sct",
      filter:[{property:"constraint",op:"=",value:$ecl}]
    }]}
  }' |
  curl -sS -X POST 'https://tx.fhir.org/r4/ValueSet/$expand' \
    -H 'Content-Type: application/fhir+json' \
    --data-binary @- |
  jq -r '[.expansion.total, (.expansion.contains[0].code // "no-code")] | @tsv'
done

Observed:

^700043003           1   4294967295
^900000000000497000 1   4294967295
^404684003           1   4294967295

Expected:

• For a valid concept-based reference set, ^refset should return the active referenced component concepts from that reference set.
• For 404684003 |Clinical finding|, which is not a reference set, the server should reject the expression or return no members.
• The expansion should never contain fabricated SCTID 4294967295.

Impact: clients using ECL memberOf expansion can receive a non-SNOMED sentinel as if it were a real inactive SNOMED concept.

Bug 2: wrapped memberOf operands are rejected as unsupported

Reproducer:

for ECL in '^(<<900000000000455006)' '^(404684003)' '^(<<404684003)'; do
  printf '\n%s\n' "$ECL"
  jq -nc --arg ecl "$ECL" '{
    resourceType:"ValueSet",
    compose:{include:[{
      system:"http://snomed.info/sct",
      filter:[{property:"constraint",op:"=",value:$ecl}]
    }]}
  }' |
  curl -sS -X POST 'https://tx.fhir.org/r4/ValueSet/$expand' \
    -H 'Content-Type: application/fhir+json' \
    --data-binary @- |
  jq -r '.issue[0].details.text'
done

Observed:

^(<<900000000000455006)
The ECL expression is not supported: '^(<<900000000000455006)': (ECL ^ (member-of) with a non-concept-reference refset is not yet supported)

^(404684003)
The ECL expression is not supported: '^(404684003)': (ECL ^ (member-of) with a non-concept-reference refset is not yet supported)

^(<<404684003)
The ECL expression is not supported: '^(<<404684003)': (ECL ^ (member-of) with a non-concept-reference refset is not yet supported)

Expected: ECL permits the memberOf operand to be a refset expression. For example, ^(<<900000000000455006) means members of all reference sets in the descendants-or-self closure of 900000000000455006 |Reference set|. The server should resolve the operand expression to reference-set concepts and return the union of their referenced component concepts.

Impact: clients cannot use standard ECL patterns that compute the set of reference sets dynamically.

Bug 3: cardinality appears to count raw relationship rows instead of non-redundant attributes

The witness concept is 903008 |Chorioretinal infarction|.

First, tx.fhir.org lookup reports one 116676008 |Associated morphology| value for the concept:

curl -sS 'https://tx.fhir.org/r4/CodeSystem/$lookup?system=http://snomed.info/sct&code=903008' |
jq -r '[.parameter[]
  | select(.name=="property")
  | select((.part[]? | select(.name=="code").valueCode) == "116676008")
  | (.part[] | select(.name=="value").valueCode)] | @json'

Observed:

["55641003"]

Now compare [1..1] and [2..*] cardinality for that same attribute/value constraint:

for ECL in \
  '<<404684003:[1..1]116676008=<<55641003' \
  '<<404684003:[2..*]116676008=<<55641003'
do
  printf '%s\t' "$ECL"
  jq -nc --arg ecl "$ECL" '{
    resourceType:"ValueSet",
    compose:{include:[{
      system:"http://snomed.info/sct",
      filter:[{property:"constraint",op:"=",value:$ecl}]
    }]}
  }' |
  curl -sS -X POST 'https://tx.fhir.org/r4/ValueSet/$expand' \
    -H 'Content-Type: application/fhir+json' \
    --data-binary @- |
  jq -r '[.expansion.total, ((.expansion.contains // []) | map(.code) | index("903008") != null)] | @tsv'
done

Observed:

<<404684003:[1..1]116676008=<<55641003  257  false
<<404684003:[2..*]116676008=<<55641003  11   true

Expected: if 903008 has one non-redundant associated morphology matching <<55641003 |Infarct|, it should satisfy [1..1] and should not satisfy [2..*].

Observed behavior is the opposite: 903008 is excluded from [1..1] and included in [2..*]. This is consistent with counting raw relationship instances instead of the non-redundant attribute values required by ECL.

Impact: cardinality-constrained ECL queries can include false positives and exclude true positives.

Sanity check

Ordinary attribute refinements are not simply ignored:

ECL='<<404684003:116676008=<<55641003'
jq -nc --arg ecl "$ECL" '{resourceType:"ValueSet",compose:{include:[{system:"http://snomed.info/sct",filter:[{property:"constraint",op:"=",value:$ecl}]}]}}' |
curl -sS -X POST 'https://tx.fhir.org/r4/ValueSet/$expand' -H 'Content-Type: application/fhir+json' --data-binary @- |
jq '{total: .expansion.total, first: [.expansion.contains[:5][] | {code, display}]}'

Observed total: 268, with infarction-related first results. The cardinality issue appears more specific than all refinements being ignored.