MIMIC-IV
MIT Laboratory for Computational Physiology and Beth Israel Deaconess Medical Center (via PhysioNet)
Real, de-identified records of 364,000 Boston hospital patients, free to anyone who completes ethics training.
A de-identified electronic health record database from Beth Israel Deaconess Medical Center, covering hospital and ICU stays from 2008 to 2022: vital signs, labs, medications, procedures, diagnoses, and notes in linked modules. It is the most widely used open clinical data set for machine learning and critical care research. Access requires PhysioNet credentialing, CITI training, and a data use agreement.
Access
- Cost
- By applicationFree, but requires a credentialed PhysioNet account, completion of CITI human subjects training, and signing the data use agreement.
- License
- Provider's own termsPhysioNet Credentialed Health Data License 1.5.0
- Sign-up
- Application required
- How you get it
- Bulk downloadCSV, BigQuery
What's in it
- One row is
- Relational tables keyed by patient, hospital admission, and ICU stay
- Covers
- One academic medical center in Boston, Massachusetts
- Years
- 2008-2022
- Updated
- Irregular
- Size
- 364,627 patients, 546,028 hospitalizations, and 94,458 ICU stays
Quality
Carefully de-identified, versioned, and documented, with a large community and shared code.
- Single hospital, so findings may not generalize
- Dates are shifted into the future for privacy
- Clinical notes are released in a separate module
Connects to
- MIT-BIH Arrhythmia Database: Another PhysioNet resource for physiologic signals
- CheXpert: Chest X-ray labels comparable with the sister MIMIC-CXR imaging set
- Epic Cosmos: Massive multi-site EHR data for checking whether single-site findings hold
Questions it could answer
- Can early lab trends predict which ICU patients develop sepsis?
- How did ICU length of stay change during the COVID-19 years?