search

MIT-LCP / mimic-code

MIMIC Code Repository: Code shared by the research community for the MIMIC family of databases
https://mimic.mit.edu
MIT License
2.41k stars 1.5k forks source link

missing icustay records #1598

Open hlblade opened 11 months ago

hlblade commented 11 months ago
Hi, I just found there were some icustay records lost in the mimiciv_icu module. For example, patient with hadm_id 20238386. When I search the transfer records for this patient, I found: subject_id hadm_id transfer_id eventtype careunit intime outtime
18385068 20238386 34322766 admit 2114-08-29 00:21:11 2114-08-29 09:46:01
18385068 20238386 39952860 transfer Cardiac Vascular Intensive Care Unit (CVICU) 2114-08-29 09:46:01 2114-09-01 17:10:45
18385068 20238386 34786921 transfer Cardiac Surgery 2114-09-01 17:10:45 2114-09-06 14:59:15
18385068 20238386 39820107 discharge 2114-09-06 14:59:15

The transfer records showed this patient was admitted to CVICU for 3 days, but no record could be found in icustays table, can some body help to explain this?

Thx!

tsteffek commented 10 months ago

I just stumbled over the same issue, coming from a completely different angle. Maybe I can help to give a ballpark estimate of the issue.

During my experiments to identify all the MIMIC-III data in MIMIC-IV, I've identified about 3k patients that look like they're from MIMIC-III (so therefore should be ICU data), but are not part of the icu module. The one hlblade discovered, 20238386, is among them.

For even more info, I'm currently detecting MIMIC-III data by matching all diagnoses of an admission, as well as hours, minutes and seconds of 'admittime', 'dischtime', 'deathtime', 'edregtime', 'edouttime' with MIMIC-IV admissions.

alistairewj commented 10 months ago

I just stumbled over the same issue, coming from a completely different angle. Maybe I can help to give a ballpark estimate of the issue.

During my experiments to identify all the MIMIC-III data in MIMIC-IV

Worth looking at https://physionet.org/content/mimic3-carevue/1.4/ if you haven't yet

I've identified about 3k patients that look like they're from MIMIC-III (so therefore should be ICU data), but are not part of the icu module. The one hlblade discovered, 20238386, is among them.

For even more info, I'm currently detecting MIMIC-III data by matching all diagnoses of an admission, as well as hours, minutes and seconds of 'admittime', 'dischtime', 'deathtime', 'edregtime', 'edouttime' with MIMIC-IV admissions.

We removed a small percentage of subjects in MIMIC-IV (randomly), and we didn't do this in MIMIC-III. That might be the reason. But it wouldn't explain it if you can find their hospitalization, but not their ICU stay, as the original post seems to imply.