Incorrect LOINC codes : SPEP and CSF electrophoresis / lack of LOINC validation

Prerequisites

• [ X] Put an X between the brackets on this line if you have done all of the following:
  • Checked the online documentation: https://mimic.physionet.org/about/mimic/
  • Checked that your issue isn't already addressed: https://github.com/MIT-LCP/mimic-code/issues?utf8=%E2%9C%93&q=

Description

I believe the LOINC codes in d_labitems for the following itemid values are incorrect:
50975 (listed LOINC code: 24351-9)
51017 (listed LOINC code: 24352-7)

These are LOINC codes for protein electrophoresis order panels (serum and CSF), not observations. I determined this by running every LOINC code in d_labitems though a validation procedure that looked at the LOINC 'Order vs Observation' value. Items in d_labitems are all observations, so any 'Order vs Observation' value of 'Order' (ie order only) is invalid. The aforementioned itemid values refer to protein electrophoresis panels (serum and CSF).

It would appear (on very rough look) that item ID 50975 should have LOINC code "12851-2" (protein fractions interpretation for SPEP). Item ID 51017 likely has LOINC code of "12782-9" (oligoclonal band interpretation for CSF electrophoresis). (Again, these were very quick searches on my part, with results based a lot on the popularity values in LOINC; I probably did not reach the 'due diligence' standard).

I think an overall LOINC code validation algorithm for the d_labitems values should include looking up the order vs observation status for the LOINC code (e.g. via UMLS), and flagging any 'order only' values.

It also might be useful to rank the d_labitems itemid values based on the number of rows that reference each value in the labevents table. This could be compared to the UMLS "Common US Lab Results Rank". Obviously, a single hospital system is not going to totally parallel the multiple hospital-systems that LOINC used in determining rank, but it might point out significant discrepancies (eg if a lab type ranked very highly in MIMIC database, but did not register in the most popular LOINC observations, this might need to be looked at more closely).

Start by checking the d_labitems entries for itemids 50975 and 51017 and compare their LOINC Order versus Observation status using the suggested UMLS validation approach. Confirm the appropriate observation mappings, then inspect labevents references if implementing the proposed usage ranking; done means the affected codes are validated and any broader validation output clearly flags order-only values.