MIMIC-IV is a credentialed dataset and is only available to approved users.

1. Request access to MIMIC-IV by following the instructions at the bottom of the official PhysioNet page:

2. Once approved, return to the same PhysioNet page and complete the step “Request access using Google BigQuery.” RelBench retrieves MIMIC-IV data through BigQuery rather than local file downloads.

After you have been granted BigQuery access, complete the following steps on the machine where your Python code will run.

1. Install the Google Cloud SDK by following the official instructions:
   https://docs.cloud.google.com/sdk/docs/install-sdk

2. Create Application Default Credentials (ADC). Run the following command once in a terminal:

   gcloud auth application-default login
   

   This step is required for Python client libraries such as google-cloud-bigquery. Running gcloud auth login alone is not sufficient.

3. Set a billing or quota project. BigQuery requires a project with billing enabled to execute queries:

   gcloud auth application-default set-quota-project YOUR_PROJECT_ID
   

   Replace YOUR_PROJECT_ID with a Google Cloud project that has billing enabled.

4. Locate the ADC credentials file. After login, gcloud prints the path where credentials are stored, for example:

   ~/.config/gcloud/application_default_credentials.json
   

   This is the file that Python must be able to access.

5. If the ADC credentials file is not located at the default path, set the environment variable:

   export GOOGLE_APPLICATION_CREDENTIALS="/path/to/application_default_credentials.json"
   

6. Save the project ID environment variable used by your code. For example:

   export PROJECT_ID="your_project_id"
   

While the pre-prepared RelBench MIMIC-IV uses only a subset of the full dataset, you can customize the dataset to meet your needs by loading the dataset directly and specifying parameters such as patients_limit, tables_limit, drop_columns_per_table, min_age, etc.

For example:

drop_columns_per_table = {
    "admissions": [...],
    "chartevents": [...],
    ...
}
tables_limit = ["patients", "admissions", "icustays", "chartevents", "procedureevents", "d_items"]


dataset = MimicDataset(
    patients_limit=20000,
    out_path='/data',
    cache_dir='/cache',
    tables_limit=tables_limit,
    db_params=db_params,
    drop_columns_per_table=drop_columns_per_table
)
db = dataset.make_db()