WAS_A - Discussion

This association type is the "association of last resort": it should be only used whenever no active concept(s) exist to precisely cover the EXACT same semantics as the inactivated concept.

The purpose of this association type has been to record the complete set of direct parent supertype(s) that were assigned to the inactivated concept in the last release in which it was still active.

A WAS_A association thus enables the inactivated concept to continue to be recognised as a "kind of" both its historical direct parents and all their ancestors recursively, subject to the proviso that the ancestors of an inactivated concept's former parents may change over time across successive releases.

The WAS_A association can not guarantee that the inactivated concept will be returned as a subclass of all semantically appropriate supertype classes that may exist in later releases. If, prior to its inactivation, the concept did not already possess all semantically valid parents and ancestors, many of these "missed classifications" will necessarily persist forever.

Finally, WAS_A associations do not enable the determination of what concepts  in the current release (whether now active, or also inactive) either used to be subtypes of the inactivated concept prior to its inactivation, or might now reasonably be judged to be their semantic subtype even if they were not classified as subtypes prior to the inactivation.

Proposed changes:

The original requirement for explicit declarations of WAS_A relationships was born out of limitations inherent to SNOMED's RF1 encoding of its release materials. Under the newer RF2 encoding, these limitations no longer apply and so it has been argued that explicitly stating the last known active parents by means of WAS_A associations is now entirely redundant, as this information can be trivially retrieved programmatically directly from the FULL distribution of the release data.

This is true, in principle.

However, if the set of historical parents can now be retrieved entirely automatically, an opportunity arises for a revised semantics of manually asserted WAS_A associations: a much smaller volume of explicitly and manually stated WAS_A statements might be maintained to either enhance or to entirely override those that are derivable automatically. In particular, where an inactive concept was not optimally classified prior to its inactivation and therefore was already missing significant semantically appropriate parents and ancestors, this "original sin" could be corrected even after inactivtion by manually asserting WAS_A relationships to ancestors that are semantically true but that are not otherwise derivable from the published data.

For example, the concept 196311008 Excessive tooth attrition NOS (disorder) was made inactive in January 2010. In the immediately preceding release, it had one direct parent 234977009 Tooth surface loss (disorder) and, thereby, the following ancestors all of which remain current active concepts today (August 2020)

At the time of its inactivation, therefore, a solitary WAS_A association was declared and remains present today (August 2020): 

However, both the original pre-inactivation taxonomy and the limited inferences still enabled today by the WAS_A association fail to include a clinically more useful relationship between inactive 196311008 Excessive tooth attrition NOS (disorder) and the still active concept 53963006 Excessive attrition of teeth (disorder). Back in 2009, the two codes were siblings.

To remedy this taxonomic omission and the false negative reporting that will always flow from it, a new manual WAS_A could be considered:

The huge majority of similar examples involve the inactivation of Classification list closure concepts, but they are plentiful. They include:

Combinatorial Logic

Whenever an already stated WAS_A target itself also become inactive - whether at the same release or later - the combinatorial logic of associations should be:

A WAS_A (B AND C)  and  B SAME_AS D implies A WAS_A (C AND D)
A WAS_A (B AND C)  and  B REPLACED_BY D implies A WAS_A (C AND D)
A WAS_A (B AND C)  and  B MOVED_TO D implies A WAS_A (C)
A WAS_A (B AND C)  and  E MOVED_FROM B implies A WAS_A (C AND E)
A WAS_A (B AND C)  and  B POSSIBLY_EQUIVALENT_TO (D OR E) implies A WAS_A (C), A WAS_A (D OR E)
A WAS_A (B AND C)  and  B WAS_A (D AND E) implies A WAS_A (C AND D AND E)