🕵️ MEDS-Inspect

MEDS (Medical Event Data Standard) is "the simplest possible standard for health AI" (https://medical-event-data-standard.github.io/).

But after building your own MEDS ETL you might be wondering:

• Is my ETL missing data?
• What codes are contained in my dataset?
• How does my data compare to other MEDS datasets?
• What preprocessing steps are still needed in order to train models?

.. and many more questions related to data exploration.

MEDS-Inspect is an interactive data visualization app that supports you in your data quest.

Getting started

You can use any of the ETLs available in the MEDS ecosystem to create your own MEDS dataset, see here for an updated list.

pip install MEDS-Inspect

Then start a server with the following:

MEDS_Inspect port=8052 +initial_path="path/to/your/meds/dataset"

This will start a local web app that you can access in your browser. Running this command without a file path will default to the MIMIC-IV Demo data in MEDS

You should also be able to enter an arbitrary filepath from the GUI.

On HPC systems you might need to forward the port, for example with SSH tunneling:

ssh -N -f -L localhost:8090:localhost:8090 remote_user@remote_host

Then access the app at localhost:8090 in your browser. For any problems, please refer to your system administrator.

Getting started (development)

Clone repository:

git clone https://github.com/rvandewater/MEDS-Inspect.git
cd MEDS-Inspect

Create environment:

conda create -n "meds-inspect" python=3.12
conda activate meds-inspect

Install requirements:

pip install -r requirements.txt

Launch app:

python src/MEDS_Inspect/__main__.py

This should start a locally hosted web app.

Functionality

You can start the caching directly from the command line. Caching creates the folder .meds-inspect-cache

python MEDS_Inspect_cache path/to/your/favorite/meds/dataset

Note

you need to input the directory with your /data and /metadata folder, for example: /sicdb/MEDS_cohort\

Impression: