Skip to main content

Welcome to MEDS

The MEDS Ecosystem, Visually

MEDS (the Medical Event Data Standard) is a shockingly simple, highly flexible, and efficient data standard for structured, longitudinal medical record data, built for reproducible, efficient Machine Learning (ML)/Artificial Intelligence (AI) research in healthcare. Building on MEDS are a vareity of open-source tools and libraries that make it easy to work with MEDS data, from pre-processing to cohort extraction to modeling and beyond!

How can MEDS help you?​

Working with MEDS is incredibly simple. Data in MEDS are stored in a simple, longitudinal format that consists of only four mandatory columns: subject_id, time, code, and numeric_value. This simplicity makes it easy to convert your data into MEDS and to use it when its in the MEDS format, and a number of existing datasets and models already support MEDS out of the box!

tip

See this list for a list of datasets and models that already support MEDS, and if you have a dataset that you'd like to convert to MEDS see this tutorial to see how!

Once your data is in MEDS, you can rely on the bevy of open-source tools built on MEDS to help you with your modeling tasks. Be it in task-extraction, data pre-processing (with MEDS-Transforms or MEDS-Reader), building baseline or neural network models, or evaluating predictions, the MEDS ecosystem has tools that can help make your research easier, more efficient, and more reproducible.

Doing Effective Science in Health AI​

MEDS is built to enable as close to frictionless reproducibility as is possible in the field of health AI -- precisely so that we can do effective empirical science in this field and generate meaningful understanding of what works and what doesn't. By using MEDS, you can ensure that your work is reproducible by default to add your insights into this communal body of evidence and to ensure that your work can build on the insights of others.

The MEDS Decentralized, Extensible, Validation (MEDS-DEV) effort facilitates exactly this kind of effective science. MEDS-DEV is a place to share task definitions (through the ACES framework), to share reproducible model training recipes and details on publicly and privately used datasets, and to share model evaluation results in a decentralized fashion over these shared tasks and datasets. This allows us to build a shared understanding of the science of health AI and to ensure that our models are robust and generalizable. Contributing your model results or your task definitions to MEDS-DEV is easy, as is reproducing prior models for comparison on your local data. Check out the MEDS-DEV GitHub repository to get started!

Where to go from here?​

  1. If you're new to MEDS and want to learn more about the nuts and bolts, check out the introductory tutorials to get started!
  2. IF you're looking for a dataset to work with, a model to compare to, or tools to help your research, check out the public research resources that support MEDS!
  3. If you're looking to help build a shared understanding of the science of health AI, check out the MEDS-DEV repository and see how you can contribute!
  4. If you're looking to read academic papers about MEDS, check out our workshop paper!
  5. Finally, if you have a question or comment not covered here, feel free to file an issue on the MEDS GitHub repository!