MIMIC-III
MIMIC-III (Medical Information Mart for Intensive Care) is a reference hospital dataset containing detailed clinical data on patients admitted to intensive care. Developed by MIT, it is widely used for medical research, the analysis of care trajectories, and the development of predictive health tools.
Over 40,000 patient records, CSV formats
Restricted access, reserved for academic research, subject to authentication and acceptance of the PhysioNet confidentiality agreement
Description
The dataset contains:
- Data from more than 40,000 hospitalized patients in intensive care (ICU)
- Demographic Information, Diagnostics (ICD-9), Prescriptions, Lab Results
- Temporal data: vital constants, curves, interventions, length of hospital stay
- Files available in CSV for direct integration into analysis environments
MIMIC-III covers a period from 2001 to 2012 and guarantees the complete anonymization of patients.
What is this dataset for?
MIMIC-III is commonly used for:
- Analysis of care pathways in intensive care
- Training predictive models of mortality, relapse, or length of stay
- Research in precision medicine and hospital management
- The development of clinical decision support systems (CDSS)
- The study of the relationships between treatments, diagnoses and results
Can it be enriched or improved?
Yes, with several axes:
- Integration of unstructured data (clinical notes, imagery)
- Addition of variables from connected devices or physiological curves
- Crossing with other databases (MIMIC-CXR for imaging, eICU for multisite extension)
- Development of benchmarks on specific tasks (prediction, clustering, medical NLP)
🔗 Source: MIMIC-III Dataset
Frequently Asked Questions
Does the dataset contain time data?
Yes, much of the data is temporal (vital signals, interventions, prescriptions) and can be used for sequential models.
What is the difference between MIMIC-III and MIMIC-IV?
MIMIC-IV is a more recent, enriched and restructured version of MIMIC-III, including data from after 2012 with better table organization.
Do you need ethical training to access it?
Yes. Access requires training in good research practices (CITI Program) as well as a signed agreement on confidentiality and terms of use.
