Concept Mapping / Aliases

I was wondering whether people had opinions on the contribution of supplementary data. I've started to try to analyse chartevents and it's a little frustrating to find out that PaO2 and PaCO2 have about 5 or 6 different item names each. It would be particularly helpful if we could come up with a way of aliasing them.

I would obviously very much prefer to not contaminate the mimiciii schema and keep that purely for the original MIMIC data, but perhaps we could add a contrib schema for data such as these? (Maybe contribiii to make it clear that it goes with mimiciii?)

contrib.aliases
alias_id
aliasname

contrib.d_items_aliases
row_id
alias_id (references contrib.aliases.alias_id)
itemid (references d_items.itemid)

This does introduce a whole host of potential issues, again relating to database versioning: I can hopefully do a sensible first-pass of various measures (PaO2, PaCO2, pH etc.), but there may come a point where another intensivist may turn around and say that actually "SpO2 XYZ" is actually not really the same as "SpO2 ABC" and "SpO2".

Would a possible solution in this case be to add some sort of version column to aliases? So that if someone decides to do their analysis with my incredibly flawed versions of aliases, then they can still replicate their data or (hopefully) choose the later version.

Another issue to consider with this sort of schema is whether there should be a unique mapping of items to aliases, i.e. could someone theoretically create a view that does SELECT * FROM chartevents and tacks on a column representing aliasname? I'm tempted to argue against this as it would potentially preclude versioning/other alias interpretations etc.

Thoughts?

Start by reviewing the proposed contrib.aliases and contrib.d_items_aliases tables and the discussion about alias versioning and item-to-alias mappings. Done means reaching agreement on the schema and semantics for supplementary aliases without altering the original MIMIC schema.