Standardizations by Source

• eICU (carePlanGeneral):
  • "Comfort measures only" → CMO
  • "Do not resuscitate" → DNR
  • "Full therapy" → full code
  • "No CPR" → DNCPR
  • "No intubation" → DNI
• MIMIC-III (chartevents):
  • "Full Code", "Full code" → full code
  • "Comfort Measures", "Comfort measures only" → CMO
  • "Do Not Resuscita", "DNR (do not resuscitate)" → DNR
  • "Do Not Intubate", "DNI (do not intubate)" → DNI
  • "DNR / DNI" → DNR / DNI
  • "CPR Not Indicate" → DNCPR
• MIMIC-IV (chartevents):
  • "Full code" → full code
  • "Comfort measures only" → CMO
  • "DNI (do not intubate)" → DNI
  • "DNR (do not resuscitate)" → DNR
  • "DNR / DNI" → DNR / DNI
• UMCdb (numericitems):
  • "I" → full code
  • "II" → DNCPR
  • "III" → CMO

Pseudocode Steps

1. Initialize paths and dataset references.
2. Extract code status data from each dataset as per the table above:
   • Load the relevant source table(s).
   • Apply the specified filter/condition.
   • Select the item and time columns.
   • Standardize the item column values as described above.
   • Convert time columns to seconds relative to admission where needed.
   • Add a global ICU stay ID.
3. Combine all datasets:
   • Concatenate all code status records.
   • Select relevant columns: global ICU stay ID, time relative to admission, code status.
   • Cast code status to a categorical type.
   • Group by global ICU stay ID and time, keeping the first code status per timepoint.
   • Remove consecutive duplicate code status entries per stay.
   • Sort by global ICU stay ID and time.
4. Return the final DataFrame with standardized code status per ICU stay and timepoint.
   • Concatenate all code status records.
   • Select relevant columns: global ICU stay ID, time relative to admission, code status.
   • Cast code status to a categorical type.
   • Group by global ICU stay ID and time, keeping the first code status per timepoint.
   • Remove consecutive duplicate code status entries per stay.
   • Sort by global ICU stay ID and time.