OHDSI/MIMIC

I have a question about wf_details in the waveform collecting stage

Kisooofficial opened this issue · 4 comments

Kisooofficial commented

Hello. I'm Korean student and I'm trying to convert MIMIC-IV to OMOP CDM.
I've already downloaded waveform data and storaged in local and google bigquery.
Also, I've already seen document about waveform data stored in waveforms directory.
My question is below.

  1. If directory is p100/p10014354/81739927/81739927n.csv, I understand like this.
    "case id = p100, subject id = p10014354, short_reference_id = 81739927, long_reference_id = google cloud storage address"
    Is my understanding correct?

  2. I'm having challenge getting wf_details table. First, I got wf_header table and schema is (case_id, subject_id, short_reference_id, long_reference_id). How could i get wf_details table?

Kisooofficial commented

I also have extra questions. In waveform examples, (csv.zip) there are segment name, date-time, src Name etc.. but, there aren't this columns in waveform data which extension is .csv. How can i get this columns from? Should I get it from a file with a .dat or .hea extension?

Kisooofficial commented
  • Why does waveform data needed for converting omop? I know that it didn't need at previous version.
manlik-brownsrdr commented

The simple answer is to take advantage of under-used objective data to better understand and detect response to treatments and possibly early indicators of improvement and/or degradation of the patient condition. Most waveform data isn't readily used for a host of reasons. Waveform data has existed in the MIMIC datasets - it has not been widely used. The clinical observations contained in the clinical side of MIMIC data is a course representation (low resolution) of the waveform data - what clinicians document from observations from bedside monitors, vitals taken, etc.

The clinical MIMIC data is easy to access, use, and mapped to an OMOP CDM.

My fundamental hypothesis is the moderate resolution data (CSV data) provide higher resolution and objective findings that can get you closer to physiological response to treatments and early indications of improving response or degrading response to the disease. Mapping the CSV data to OMOP CDM isn't that difficult, however it does require additional analytics to summarize episodes of interests. What I mean is that if the CSV reported heart rate is bouncing around 60 - 80 beats per minute for a few hours - that is a lot of data points that isn't too interesting and can simply be summarized as normal sinus rate for that period of time. If the heart rate jumps to 180 bpm - then it would be interesting to identify that and look at what else is happening from the clinical data and/or other physiological parameters. This gets us into annotations and event detection.

The same is true when considering using the waveform data that is also contained in the WFDB side of MIMIC.

Why it hasn't been used more widely is I suspect the need to create and apply signal processing and annotation algorithms to the data. I'm presently doing this sort of work now on the MIMIC-IV WFDB data - both for the CSV and raw waveform data. The later summarizing and annotating the data into 10 min representative waveform for measurements and morphology analysis (ECG and ABP).

pflugg commented