Hi Roger! This challenge is near and dear to my heart. This is indeed a known limitation of the vocabulary, and impacts not just the use case you describe but also standard OMOP CDM research queries.
Traditional research teams often work with ICD-based codelists, so as an analyst I’m faced with the need to translate these into OMOP concept sets. To do so I follow the exact process you’ve described; I use OMOP to discover missing codes and validate the original ICD-based list in the process of building my concept set. Doing so is incredibly time consuming and requires clinical input when discrepancies arise (which is often).
From my perspective, there are 2 separate issues/challenges. The first is OK and part of the job when we’re trying to standardize. The second is not OK and should be fixed.
- As you’ve pointed out, sometimes SNOMED’s method for organizing codes is just different from ICD’s or individual researchers’ methods. We can handle this by curating our concept set expression with the right mix of SNOMED codes to match the researcher’s intention.
- Even if we do #1 correctly we still might end up with an invalid concept set. This is due to erroneous and/or “uphill” mappings of ICD codes to SNOMED codes in the OMOP vocabulary. Errors can be reported to the vocab team and fixed. Uphill mappings, if not implemented in error, don’t currently have a solution. These occur when there is not a SNOMED code corresponding to a given ICD code, and then the ICD gets mapped to a more-general SNOMED that carries with it other ICDs and source codes that don’t match the clinical description. Currently the only way to solve these is to build a custom source concept set.
@Dymshyts led a discussion about this at last year’s Symposium, and maybe he has an update/thoughts on the path forward? 
Vocabulathon 2025: Precise mapping