Description for Diagnosisstring in diagnosis table

[mit-lcp]

Description for Diagnosisstring in diagnosis table

[alistairewj]

Think of the diagnosis string as an "active problem". It is hierarchical and the pipes delimit the levels. Slashes are just regular characters here so the second level for the neurological example is "altered mental status / pain".

ICD codes are present for the diagnosis table where possible but often vague diagnoses do not have an ICD code (such as "diabetes" with no qualifier of type I or type II).

Finally, admitDxText is an admission diagnosis using a custom APACHE ontology (around 400 diagnoses or so). Likely that is your best option for determining admission diagnosis, when it's available.

[sachin1t]

Thanks for answering my questions.
The 'icd9code' from the diagnosis table appears to have multiple icd9 codes listed on each line separated by a comma. I initially thought the first code is icd9 and the next one is the equivalent icd10, but that doesn't appear to be the case. The tables appears to have some icd10 codes, even though the tables description says icd9 code.

I tried splitting the icd9 codes into separate columns, but I am not sure about its accuracy since the number of codes listed on each line varies. (I thought having the codes in distinct columns would make my queries simpler)
I used the following code assuming there are a maximum of 4 icd9 codes on each line.

SELECT patientunitstayid, diagnosisstring, diagnosispriority, icd9code,
split_part(icd9code, ',', 1) AS icd9_1 ,
split_part(icd9code, ',', 2) AS icd9_2,
split_part(icd9code, ',',3) AS icd9_3,
split_part(icd9code, ',',4) AS icd9_4
FROM eicu.diagnosis);

This approach however gives me a lot of missing values since a lot of rows have ditto signs instead of the actual icd codes.
Is copying the codes from the line above and then splitting the icd9code the only way to address this ?

If for instance I am interested in identifying admits related to suicide, I would have missed the below case if I searched based on icd9 code. Since the code for suicide (E950 & related ) are not listed.
Whereas if I search based on keywords I might get this hit.

I didn't see any notes section as with MIMIC-III data which might be used for keyword querying. Please correct me if wrong.

[jraffa]

My understanding is that there may be ICD-9 codes and/or an ICD-10 codes. Although the codes are nice, I would recommend using them with the diagnosisstring. I have found that the codes are hit and miss, and you should definitely attempt to validate your methods.

You may also want to look in pasthistory.

Unfortunately there are no notes for eICU-CRD.

[tompollard]

My understanding is that there may be ICD-9 codes and/or an ICD-10 codes.

This is correct, the column contains both ICD-9 and ICD-10 codes. We have talked about parsing the column into a view with separate columns for each, but haven't yet done this.