# MIMIC-IV Demo in the Medical Event Data Standard (MEDS)

## Facts

- Provider: van de Water, Steinberg, Wornow, Rockenschaub and McDermott (MEDS working group / PhysioNet)
- License: ODbL-1.0 (https://physionet.org/content/mimic-iv-demo-meds/0.0.1/)
- Format: Parquet (in sampled files)
- Stored at: Amazon S3, s3://physionet-open/mimic-iv-demo-meds (us-east-1)
- Access: public, readable without an account
- DOI: https://doi.org/10.13026/t2y8-ea41
- Category: Health & Biomedical
- Source: https://doi.org/10.13026/t2y8-ea41
- Page: https://scigantic.com/archive/60a4885b-dceb-594d-b7d3-28b3ca5bb356
- Keywords: Health & Biomedical, Parquet, AWS Open Data

## About

An automated ETL conversion of the MIMIC-IV Clinical Database Demo (v2.2) into the Medical Event Data Standard (MEDS), a minimal schema for streams of medical events from electronic health records or claims, intended to make EHR machine learning more reproducible across datasets and model architectures. It covers routinely collected critical care EHR data for 100 patients (subjects), split into data/train (80 subjects), data/tuning (10, validation) and data/held_out (10, test). Each split holds a Parquet file of event rows with the required MEDS columns subject_id, time and code and the optional numeric_value and text_value. metadata/codes.parquet describes each code with a description and parent_codes, metadata/dataset.json records the ETL and version, and metadata/subject_splits.parquet assigns subjects to splits. Produced with the MIMIC_IV_MEDS package v0.0.6 built on MEDS-Transforms, as a single shard. Measured 2026-11-13 (version 0.0.1): data/train/0.parquet is 4,860,946 bytes and metadata/codes.parquet 390,097 bytes; a full zip (4,923,223 bytes) also sits at the prefix root. Read with pandas, polars or pyarrow, or with the meds Python package. Released on PhysioNet under the Open Data Commons Open Database License v1.0; cite van de Water et al. (2025), doi 10.13026/t2y8-ea41.

## Load the data

```python
import boto3
from botocore import UNSIGNED
from botocore.client import Config
s3 = boto3.client('s3', region_name='us-east-1', config=Config(signature_version=UNSIGNED))
s3.list_objects_v2(Bucket='physionet-open', Prefix='mimic-iv-demo-meds/', Delimiter='/')
```

Lists the files with anonymous access. Runs on any machine, with no cloud account.

## Data structure

File types in sampled files: Parquet

Example file: `mimic-iv-demo-meds/0.0.1/data/held_out/0.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/60a4885b-dceb-594d-b7d3-28b3ca5bb356
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/60a4885b-dceb-594d-b7d3-28b3ca5bb356
- Over MCP: `get_schema_card` with id `60a4885b-dceb-594d-b7d3-28b3ca5bb356` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Health & Biomedical datasets](https://scigantic.com/topics/health-and-biomedical)
