CDISC for AI · Worked example · early, exploratory work (October 2026)
One terminology change, four places that carry part of its meaning, and no link between them. A worked example of what flat standards leave implicit, and what an explicit graph would state.
Between SDTM Controlled Terminology 2026-03-27 and 2026-09-25 four instrument codelists changed class. Three Clinical Global Impression variants moved from Questionnaire to Clinical Classification, and the Lansky Play-Performance Status Scale moved the other way. Where NCIt states who rates the instrument, each move follows the rater. Yet the rater is not recorded anywhere as data.
| Codelist | Instrument | Move | Who rates, per NCIt |
|---|---|---|---|
LPPSS1TC | Lansky Play-Performance Status Scale | Clinical Classification → Questionnaire | parent (C38144) |
ACGC01TC | ADCS Clinical Global Impression of Change | Questionnaire → Clinical Classification | clinician (C103515) |
CGI02TC | Clinical Global Impression, generic modification | Questionnaire → Clinical Classification | clinician (C135738) |
GCGI01TC | General Clinical Global Impressions | Questionnaire → Clinical Classification | not stated |
The CDISC definition of the Clinical Classification domain (C228234) says these instruments are "based on a trained healthcare professional's observation and clinical judgement". A parent-rated scale does not fit; clinician-rated ones do.
Each concept is correct on its own terms. The rater, the class rule and the leftover names from the earlier class sit in text, in different places.
The rater and the class become statements that can be queried and checked. A container still named after the old class, or a parent-rated instrument in Clinical Classification, shows up as a failed check, not as something a reader has to notice.
Each move renamed the codelists. The question container concept and an old synonym still carry the earlier class.
| Codelist | Class now | Synonym still carried | Container concept, as named | Instrument concept |
|---|---|---|---|---|
LPPSS1TC | Questionnaire | CC-LPSS TESTCD | C138263 LPSS Clinical Classification Question | C38144, not linked to the codelist |
ACGC01TC | Clinical Classification | QS-ADCS-CGIC TESTCD | C103490 ADCS-CGIC Questionnaire Question | C103515, renamed, under C118969 |
GCGI01TC | Clinical Classification | QS-GCGI TESTCD | C105145 GCGI Questionnaire Question | none found |
CGI02TC | Clinical Classification | QS-CGI Generic Modification Version TESTCD | C135705 CGI Generic Modification Version Questionnaire Question | C135738, renamed, under C118969 |
C118969 is Clinical or Research Assessment Classification. For Lansky, the instrument concept with the rater sits under Performance Status (C20641).
An illustration in Turtle. The NCIt identifiers are real; the ex: predicates and classes are placeholders. Choosing them, ideally from established ontologies, is the open work.
@prefix ncit: <http://ncicb.nci.nih.gov/xml/owl/EVS/Thesaurus.owl#> .
@prefix ex: <https://example.org/illustrative#> .
ncit:C138215 ex:codelistOf ncit:C38144 . # LPPSS1TC → Lansky Play-Performance Status
ncit:C38144 ex:ratedBy ex:Parent ;
ex:classifiedAs ex:Questionnaire . # in SDTM CT 2026-09-25
ncit:C228234 ex:requiresRater ex:HealthcareProfessional . # Clinical Classification domain
ncit:C138263 ex:containerFor ncit:C138215 . # still labelled "Clinical Classification Question"
With statements like these, two questions become queries: which instruments are clinician-rated? and does any instrument's class contradict its rater, or its container? A change note can then say what a reclassification means — the rater did not fit the class — and not only which labels changed.
Sources: NCI EVS SDTM Terminology 2026-09-25 and 2026-03-27; NCI Thesaurus as used in the repository. Instrument–codelist matches from SDTM_Instrument_Identity.xlsx.