Skip to main content

Which tool do I use for what?

Going from raw EHR data to a shared benchmark result, each step of the MEDS workflow has tools built for it. Click a tool in the flowchart to jump to its card, with its GitHub stars and links to its code, documentation and paper. To search the tools by name or topic, browse all tools.

Start: your raw EHR data

  1. Convert your data to MEDS

    Run a ready-made ETL for a public dataset or another data standard (such as OMOP), or build your own from a template.

    ETL Templates/Tools

    • ETL Template
    • MEDS-ETL
    • MEDS Extract
    • EHR2Trace
    • WeavEHR
    • EHR2MEDS
    • MESSY Synth

    Public Datasets

    Alternate Data Standards

  2. Explore and inspect

    See what is in a MEDS dataset with summary statistics, plots and interactive dashboards.

  3. Pre-process and load

    Filter, normalize and tokenize MEDS data, and load it efficiently for model training.

  4. Define a prediction task

    Describe a cohort and its labels once, then extract them from any MEDS dataset.

  5. Build models

    Train tabular baselines or neural and foundation models, with a modeling library or a published, reproducible recipe.

    Modeling

    Published Models

  6. Evaluate

    Score a model's predictions with standardized metrics.

  7. Benchmark and share

    Compare against published results, and share your tasks, models and results with the community.

Across the workflow

The MEDS schema itself, and utilities that are useful at any step.

MEDS Core

Testing

Semantic Web

Tools by step

Convert your data to MEDS

ETL Templates/Tools

  • ETL Template

    A Copier template for creating a config-only MEDS-Extract ETL package that converts a new dataset into MEDS.

    ⭐ 7 stars

    📦 0.1.0

    GitHub
  • MEDS-ETL

    A collection of ETLs that convert MIMIC-IV 2.2 and OMOP CDM 5.3 and 5.4 data into MEDS 0.3.3.

    _Use for the MIMIC dataset is deprecated_!

    ⭐ 44 stars

    📦 0.3.11

    GitHub
  • MEDS Extract

    Converts raw EHR data into MEDS using ETL pipelines that are declared in a configuration file and built on MEDS-Transforms.

    ⭐ 9 stars

    📦 0.8.1

    GitHubPyPI
  • EHR2Trace

    Converts hospital EHR exports into OMOP CDM 5.4 and MEDS 0.4, linking every output row to its source rows.

    ⭐ 4 stars

    📦 v0.7.0

    GitHubPaper
  • WeavEHR

    Extracts and harmonizes ICU data from sources such as MIMIC-IV, eICU and OMOP exports into MEDS, using declarative YAML configurations.

    ⭐ 2 stars

    📦 0.1.2

    GitHubDocsPyPI
  • EHR2MEDS

    Formats raw EHR data dumps and converts them to MEDS, via an intermediate format called preMEDS.

    ⭐ 4 stars

    GitHub
  • MESSY Synth

    Generates synthetic raw data matching a MEDS-Extract config, so an ETL can be demoed or smoke-tested without the real dataset.

    ⭐ 0 stars

    📦 0.0.1

    GitHub

Public Datasets

  • MIMIC-IV

    Converts MIMIC-IV, a dataset of emergency department and ICU patients at Beth Israel Deaconess Medical Center in Boston, into MEDS.

    🚀 Demo Available

    ⭐ 27 stars

    📦 0.2.0

    GitHubPyPIData repositoryMEDS-DEV
  • MIMIC-IV Demo

    The MIMIC-IV Clinical Database Demo, with 100 critical care patients, already converted into MEDS and openly downloadable.

    Data repository
  • AUMCdb

    Converts AmsterdamUMCdb, a de-identified ICU database from Amsterdam University Medical Centers, into MEDS.

    ⭐ 3 stars

    📦 0.1.0

    GitHubPyPIData repositoryMEDS-DEV
  • NWICU

    Converts NWICU, the Northwestern ICU database from a network of hospitals in Chicago, into MEDS.

    ⭐ 4 stars

    📦 0.0.13

    GitHubPyPIData repositoryMEDS-DEV
  • SICdb

    Converts SICdb, the Salzburg Intensive Care database of admissions to four ICUs at University Hospital Salzburg, into MEDS.

    ⭐ 6 stars

    📦 0.1.0

    GitHubPyPIData repository
  • INSPIRE

    Converts INSPIRE, a perioperative dataset of about 130,000 surgical cases from an academic institution in South Korea, into MEDS.

    ⭐ 3 stars

    📦 v0.1.0

    GitHubPyPIData repositoryMEDS-DEV
  • eICU

    Converts the eICU Collaborative Research Database, a multi-center database of ICU admissions across the United States, into MEDS.

    🚀 Demo Available

    ⭐ 7 stars

    📦 0.2.0

    GitHubPyPIData repositoryMEDS-DEV
  • HiRID

    Converts HiRID, a high time-resolution ICU dataset from Bern University Hospital in Switzerland, into MEDS.

    ⭐ 3 stars

    📦 0.1.0

    GitHubPyPIData repository
  • CDSL

    Converts the tabular data of CDSL, on patients hospitalized with confirmed or suspected COVID-19 at Spain's HM Hospitales, into MEDS.

    ⭐ 0 stars

    📦 0.0.1

    GitHubPyPIData repository
  • ALOTT

    Converts the EMR tables of ALOTT, a database pairing bedside telemetry with EMR data from two hospitals, into MEDS.

    ⭐ 0 stars

    📦 0.0.1

    GitHubPyPIData repository
  • PIC

    Converts the Paediatric Intensive Care (PIC) database, from the Children's Hospital of Zhejiang University School of Medicine, into MEDS.

    ⭐ 0 stars

    📦 0.0.3

    GitHubPyPIData repository
  • Zigong

    Converts a database of ICU patients with infection at Zigong Fourth People's Hospital in Sichuan, China, into MEDS.

    ⭐ 0 stars

    📦 0.3.1

    GitHubPyPIData repository
  • MC-MED

    Converts MC-MED, clinical and monitoring data from adult emergency department visits at Stanford Health Care, into MEDS.

    ⭐ 0 stars

    📦 0.1.0

    GitHubPyPIData repository
  • K-MIMIC (synthetic)

    Converts Synthetic K-MIMIC (SYN-ICU), synthetic ICU data generated from Seoul National University Hospital records, into MEDS.

    ⭐ 0 stars

    GitHubData repository
  • EHRSHOT

    A benchmark for few-shot evaluation of EHR foundation models, with the full records of 6,739 Stanford Medicine patients and 15 tasks.

    _Distributed natively in MEDS (`EHRSHOT_MEDS.zip` on Redivis); no ETL needed._

    ⭐ 234 stars

    GitHubDocsPaperData repository

Alternate Data Standards

  • OMOP (via MEDS-ETL)

    Converts OMOP CDM 5.3 and 5.4 datasets into MEDS 0.3.3 with the OMOP ETL of MEDS-ETL.

    ⭐ 44 stars

    📦 0.3.11

    GitHubPyPI
  • OMOP (via dedicated OMOP_MEDS)

    Converts OMOP CDM 5.3, 5.4 and 5.5 datasets into MEDS with a pipeline built on MEDS-Extract.

    ⭐ 5 stars

    📦 v0.4.0

    GitHubPyPI
  • FHIR

    Converts HL7 FHIR R4 and R5 resources, from bulk-export NDJSON files or patient bundles, into MEDS.

    ⭐ 3 stars

    📦 0.2.0

    GitHubPyPI

Explore and inspect

  • MEDS-Inspect

    An interactive web app for visually exploring the codes and contents of a MEDS dataset, for example to check an ETL.

    ⭐ 7 stars

    📦 0.0.11

    GitHubPyPI
  • suMEDS

    Summarizes a MEDS dataset into a compact code catalog with descriptions, related codes and counts, masking or removing rare codes.

    ⭐ 3 stars

    📦 0.6.3

    GitHubDocsPyPI
  • MEDS Summary Stats

    Computes privacy-safe summary statistics of MEDS datasets and compares them, to validate ETLs and catch regressions between versions.

    ⭐ 2 stars

    📦 0.0.4

    GitHubPyPI
  • MEDS Visualizations

    Plots data extracted from MEDS datasets, such as code frequencies, in Jupyter notebooks.

    ⭐ 1 stars

    📦 0.0.5

    GitHub

Pre-process and load

  • MEDS-Reader

    Converts a MEDS dataset into a fast database and reads its subjects as Python objects, to process with plain Python.

    ⭐ 22 stars

    📦 0.2.0

    GitHubDocsPyPI
  • MEDS Transforms

    Builds and runs pre-processing pipelines over MEDS datasets from reusable stages, such as filtering, normalization and outlier detection.

    ⭐ 41 stars

    📦 0.7.1

    GitHubDocsPyPI
  • MEDS TorchData

    Pre-processes a MEDS dataset into tensors and loads it as a PyTorch dataset for training models.

    ⭐ 5 stars

    📦 0.9.0

    GitHubDocsPyPI

Define a prediction task

  • ACES

    Extracts task-specific cohorts from MEDS datasets, using a configuration file that defines the task's predicates and criteria.

    ⭐ 50 stars

    📦 0.7.4

    GitHubDocsPaperPyPI
  • MEDS Random Task Sampler

    Generates reproducible, model-independent task rows from MEDS datasets by pairing codes and time horizons with sampled patient prediction times.

    ⭐ 0 stars

    📦 v0.2.0

    GitHubPyPI
  • ehrlint

    Audits a prediction task on a MEDS or OMOP dataset for leakage, such as post-index features or patient overlap across splits.

    ⭐ 0 stars

    GitHub

Build models

Modeling

  • MEDS-Torch

    Trains deep learning models on MEDS data, either supervised or pretrained and fine-tuned, with several ways to tokenize events.

    ⭐ 30 stars

    📦 0.0.8

    GitHubDocsPaperPyPI
  • FEMR

    Supports machine learning on MEDS data with ontology tools, labeling, tabular featurization and training of MOTOR foundation models.

    ⭐ 185 stars

    📦 0.2.4

    GitHubPyPI
  • MEDS to CEHR

    Converts MEDS datasets into CEHR-GPT's visit-based patient sequences, without depending on the CEHR-GPT model itself.

    ⭐ 0 stars

    GitHub
  • meds2text

    Transforms MEDS data extracted from OMOP CDM sources into textual representations of patient records.

    ⭐ 3 stars

    GitHubPaper
  • TwinWeaver

    Turns patient histories, including MEDS data, into text prompts so that LLMs can forecast biomarkers and predict clinical events.

    ⭐ 24 stars

    GitHubDocsPaperPyPI

Published Models

  • CEHR-BERT

    Pretrains a BERT-style model on MEDS or OMOP data, using temporal information, then fine-tunes it for clinical prediction tasks.

    ⭐ 50 stars

    📦 v1.5.0

    GitHubPaperPyPIMEDS-DEV
  • MEDS-Tab

    Tabularizes MEDS data into sparse features aggregated over time windows and trains baseline models, mainly XGBoost, on them.

    ⭐ 21 stars

    📦 0.3.0

    GitHubPaperPyPIMEDS-DEV
  • GenHPF

    A general healthcare prediction model that embeds EHR data from any system, including MEDS, without code mapping or feature selection.

    ⭐ 25 stars

    📦 v1.0.12

    GitHubPaperPyPIMEDS-DEV
  • EHR-RAGp

    A retrieval-augmented clinical prediction model that retrieves relevant segments of a patient's own EHR history, using MEDS-based preprocessing.

    ⭐ 2 stars

    GitHubPaper
  • CEHR-GPT

    A GPT-style foundation model for structured EHR data, including MEDS, for patient representations, zero-shot prediction and synthetic data generation.

    ⭐ 20 stars

    📦 v0.1.10

    GitHubPaperPyPI
  • MOTOR

    A time-to-event foundation model, pretrained on Stanford Medicine EHRs, that produces patient representations for predicting when clinical events occur.

    _Reads MEDS v0.3; pretrained weights are gated (CC-BY-NC 4.0)._

    ⭐ 185 stars

    📦 0.2.4

    GitHubDocsPaperPyPI
  • CLMBR-T-base

    A 141-million-parameter autoregressive foundation model, pretrained on Stanford Medicine EHRs, that turns MEDS patient records into representations for prediction.

    _Pretrained weights are gated (CC-BY-NC 4.0)._

    ⭐ 234 stars

    GitHubDocsPaper
  • Context Clues

    Pretrained long-context EHR foundation models (GPT, Llama, Mamba and Hyena), with code for training Hugging Face models on MEDS data.

    _Pretrained weights are gated (CC-BY-NC 4.0)._

    ⭐ 55 stars

    📦 v0.1.2

    GitHubDocsPaperPyPI
  • ETHOS

    A GPT-2-based EHR foundation model that tokenizes MEDS data and makes zero-shot predictions by generating future patient health trajectories.

    ⭐ 68 stars

    GitHubPaper
  • MEDS-EIC-AR

    An "everything-is-code" autoregressive generative model for MEDS data, capable of zero-shot inference.

    ⭐ 12 stars

    📦 0.5.1

    GitHubPyPI
  • EveryQuery

    A promptable foundation model trained on MEDS data that answers whether a code will occur in a future time window.

    ⭐ 2 stars

    📦 v0.7.0

    GitHubPaperPyPI
  • BONSAI

    Turns MEDS records into tokenized patient histories, pretrains a GPT- or BERT-style transformer on them and fine-tunes it for outcome prediction.

    ⭐ 14 stars

    📦 v2.0.1

    GitHubPaper

Evaluate

  • meds-evaluation

    Evaluates the predictions of models built in the MEDS ecosystem on binary classification tasks, and writes the results to JSON.

    ⭐ 8 stars

    📦 0.0.6

    GitHubPyPI

Benchmark and share

  • MEDS-DEV

    A decentralized benchmark in which data owners run shared ACES tasks and model training recipes on their own MEDS datasets.

    ⭐ 44 stars

    📦 0.2.0

    GitHubDocsPyPI

Across the workflow

MEDS Core

  • MEDS Schema

    Defines the MEDS schemas for events, dataset and code metadata, subject splits and labels, with matching Python types.

    ⭐ 143 stars

    📦 0.4.1

    GitHubPyPI

Testing

  • MEDS Testing Helpers

    Provides pytest fixtures, sample MEDS datasets and generated larger datasets for testing and benchmarking MEDS tools and pipelines.

    ⭐ 4 stars

    📦 0.3.1

    GitHubDocsPyPI

Semantic Web

  • MEDS Ontology

    An OWL ontology and SHACL shapes that formally encode the MEDS schemas, from subjects and measurements to splits and labels.

    ⭐ 1 stars

    📦 v1.0.2

    GitHubDocs
  • MEDS2RDF

    Converts MEDS datasets into RDF knowledge graphs based on the MEDS Ontology, streaming the triples to memory or to files.

    ⭐ 4 stars

    📦 v2.0.0

    GitHub