FHIR Conversion Assumptions

General

• UUIDs deterministically generated (mimic2fhir uses random UUIDs)
• No Location or Practitioner resources were generated since MIMIC-IV drops this information
• No Specimen resource was generated, the information didn't seem overly useful. MIMIC only has three columns on specimen which bleed into the microbio anyways
• Generated one organization, just had a generic MIMIC Hospital

Patient

• Birthdate was calculated use transfer.intime since the admission.admittime was not always present for a patient but at least one transfer.intime was
• Race/ethnicity are both mapped to the same thing, will need to add some logic for race
• Birthsex just pulls in gender
• Patient name is set to 'Patient_<subject_id>'

Condition

• ClinicalStatus all set to "active"
• Category all set to "encounter-diagnosis"
• Base identifier is just hadm_id, icd_code, and seq_num concatenated together

Encounter

• Status is fixed as 'finished' --> logic is anything in mimic is final now

Procedure

• Status fixed to 'completed Base identifier is just hadm_id, icd_code, and seq_num concatenated together.

Observation-labs

• Using charttime, not storetime (there was a note that chartime is significantly earlier than the result so may not be as useful as it seems)
• Storetime is currently being stored as 'issued' time when providers have access
• 'Status' set to 'final' for all observations
• 'Category' set to 'laboratory' for all observations

Observations-micro

• Structure: Observation_micro_test -> observation_micro_org -> observation_micro_susc
• Charttime used in micro-test
• Storetime used for micro-susc, since more a final time for susc results

Medication

• Current identifer/code based off of NDC, but some meds don't have NDC mapped.
• Created a table that maps these extra values to an identifier, that we can then set to its NDC later
• Status variable set to fixed 'completed'

MedicationAdministration

• Pulling in emar_detail information only for parient_field_ordinal=NULL (has information all in one row vs broken into multiple)
• Dosage pulled from MIMIC's dose_due vs dose_given
• Rate pulled from MIMIC's infusion_rate

MedicationAdministrationICU ( MIMIC inputevents)

• No real mapping to medications found in pharmacy/prescription

EncounterICU

• Made the encounters "partOf" the larger encounters found in admissions

Observation chartevents/datetimeevents/outputevents

• category variable set to d_items category (will need to be concept mapped later)
• code is just set to d_items itemid and label (again will need to be concept mapped later)