I want to post my response to email discussion on the forum.
The topic is that LOINC is too granular and measurements are not nicely ready for analysis across the network.
LOINC parts are important and having them in Athena is important (described here )
current parts structure imported into Athena comes from LOINC to SNOMED map and that is not optimal. Using raw LOINC file would be better.
LOINC groupings may be helpfull (this already is in Athena)
in ThemisConcept study - we did observe SNOMED CT being used for lab, but its use was not winning over LOINC.
some measurement reasoning may have to left to be tackled by analyst (and not during ETL) at study design time. E.g., the study (not CDM) will deal with the fact that analysis can take advantage of glucose in 12 hour urine somehow (if ideally I would want to have glucose in 24 hour urine)
-ThemisConcept study (and Anna’s ConceptPrevalence study) both try to produce network level insights into what the network is using as concepts in measurements tables
even if LOINC is too granular (but maybe for a good reason), the first step is to have same unit for that granular LOINC code. ETL help for folks to achieve that is here https://github.com/OHDSI/sandbox/tree/master/unitmorph (so directly convertable LOINC codes will be made beautiful this way)
if we want to group “equivalent LOINC” (not convertible but close enough), LOINC groups may be best. I know that LOINC admins would in fact welcome submissions of new grouping concepts (and I hope to submit some for HIV-related LOINC tests). That way the burden of maintenance is not on us but shifted to the SDO (Regenstrief in this case) (which is our mantra)