Pooled SUPPQUAL must be split into per-domain NS-- datasets

Rule

NS-002

Category

Structural

Basis

Confirmed Traceable to the published specification text cited below.

Detectability

Define-XML conclusive The Define-XML settles this one way or the other. No further evidence is needed to know whether it applies.

Finding type

Impact

Applies to

SDTMIG 3.2, 3.3, 3.4

What changes in v4.0

The v4.0 NS-- structure is per-parent-domain by construction: the domain abbreviation is used "as a value in the RDOMAIN variable in ... the Non-Standard Variables for [dataset name] (NS--) relationship datasets", and Section 8.2 states that "there is a one-to-one correspondence between a domain dataset and its NS-- dataset". A pooled SUPPQUAL dataset spanning several RDOMAIN values has no direct equivalent: it must become several datasets, one per parent domain, each split again if its parent domain is split.

What to do about it

Split SUPPQUAL by RDOMAIN into one NS-- dataset per parent domain, then carry out the full SUPP-to-NS conversion on each — which is a content review, not a transpose. Everything NS-001 lists applies to each resulting dataset: deciding which QNAM values are non-standard variables at all, resolving each parent record to the parent domain's --SEQ, the IDVARVAL-to-IDVARVLN change from character to numeric, records that duplicate a QNAM against one parent, restoring type, length and codelist metadata lost to the character QVAL, and finding homes for QLABEL, QORIG and QEVAL. A pooled dataset makes parent resolution harder rather than easier, because one IDVAR value has to be interpreted against whichever RDOMAIN the row belongs to.

What the scan can and cannot see

A dataset named SUPPQUAL is the pooled supplemental structure, and Section 8.2's one-to-one correspondence between a domain dataset and its NS-- dataset settles that it must be split. How many parent domains it actually spans, and therefore how many NS-- datasets fall out of it, is record-level and is not claimed.