Terminology Versions

[ChristophB]

Background

Terminologies are sometimes updated - some periodically, some only occasionally.

There are typically the following changes carried out in an update:

• ADD new code
• REMOVE code
• MODIFY code

There can be cases where domain experts need to decide manually how to apply a version update. For example, a code is removed and one or more new codes are added as a replacement.

It is unclear if there is a common practice in providing migrations or change sets for version updates. For instance, BfArM provides update lists for ICD-10-GM in CSV format (https://www.bfarm.de/DE/Kodiersysteme/Services/Downloads/_node.html)

Implications for the TOP Framework

• The terminology version used should be saved in TOP, so that it is possible to see whether codes used are potentially outdated. (repository level vs. entity level)
• Problem: Codes from different versions in a model must be avoided!
• A terminology update could lead to changes in the codes in a model. This means that phenotypes must be updated. The old codes would still be included in the version history.
• Most changes between versions can probably be covered automatically with migrations or mappings (see previous section).

[RalphBln]

A quick investigation revealed a certain diversity in how creators of terminologies publish changes

• ICD-10 ✅
  • two CSV tables containing codes and code labels of the current and the preceding version, respectively. Changes can easily be identified via diff.
  • A CSV file with code transitions: old code, new code, and hints about about in which direction(s) automated transitions can be applied or not.
• OPS ✅
  • like ICD-10, with an additional flag indicating whether a new code requires a supplementary character (for laterality or diagnosis confidence)
• PZN ❌
  • no version-to-version mapping info
• ASK ❌
  • no version-to-version mapping info
• ATC ✅
  • an XML file containing a structural diff between ontology versions (additons and deletions of classes and (subClass) axioms)
  • not trivial to identify renamings
• LOINC ✅
  • A CSV file with mappings of deprecated codes to their new replacements.
• SNOMED
  • Several csv files with mapping information - needs further investigation

[RalphBln]

The imported terminologies already contain version information, which can also be retrieved via the OLS API. Adding it to the TOP data model should be straightforward.