Inselspital, University Hospital Bern

Procedure classifications like the Swiss procedure catalogue CHOP or the German procedure catalogue OPS often contain similar content, but differ on granular level and in hierarchical structure and format. Moreover, they have to be translated to English to be processed by standard mapping tools to interoperable terminologies. With data interoperability gaining importance in the provision of health care and in international research, supporting tools are essential. The Observational Health Data Sciences and Informatics (OHDSI) team offers the Usagi tool to help in (auto-) mapping codes into standard terminologies (in this study SNOMED CT). As a preparation in this study using a transformer-based LLM, the CHOP and the OPS medical procedure codes were translated into English (accuracy of 90%), contextually augmented and loaded into Usagi. We then tested the Usagi term similarity approach as a validation for both datasets of translated and mapped procedure catalogues from the different coding systems. The agreement of the SNOMED CT identifier pairs (semantic match) was compared and rated (narrower, broader, equivalent, exact). The resulting validation by mapping to SNOMED CT concepts showed the following results: 52 of 494 SNOMED CT concept pairs (10.5%) were rated as equal (identical SNOMED CT concept identifier) and 20 of 494 (4.1%) as semantically equivalent (different SNOMED CT concept identifier). 273 (55.3%) concept pairs in total were rated as equal, equivalent, narrower or wider. In summary, the Usagi tool with SNOMED CT as standard terminology supports content and workflow for automapping and efficient validation.