MIMIC-IV schema overview

Table of contents

  1. Modular design
  2. Core identifier system
    1. Patient level
    2. Hospital admission level
    3. Unit stay level
  3. Data flow through the hospital
  4. Module characteristics
    1. Hospital (hosp) module
    2. ICU module
    3. Emergency department (ED) module
    4. Note module
    5. CXR module
    6. ECG module
  5. Data relationships
    1. One-to-many relationships
    2. Cross-module linking
  6. Temporal considerations
    1. Time precision
  7. Common analysis patterns
    1. Patient cohort selection
    2. Longitudinal analysis
    3. Outcome assessment
  8. Best practices
    1. Query design
    2. Data validation

Understanding the overall structure and organization of the MIMIC-IV database is crucial for effective analysis.

Modular design

MIMIC-IV uses a modular design. The hospital (hosp) module contains data acquired from the hospital wide electronic health record. The ICU (icu) module contains data from the clinical information system used within the ICU. Additional modules (ED, CXR, ECG, Note) extend MIMIC-IV with data from other systems.

Core identifier system

The database uses a hierarchical identifier system:

Patient level

subject_id is a unique identifier which specifies an individual patient. Any rows associated with a single subject_id pertain to the same individual.

Hospital admission level

hadm_id is an integer identifier which is unique for each patient hospitalization.

Unit stay level

stay_id is an integer which uniquely identifies a stay in an specific department within the hospital. The identifier is used to delineate contiguous ICU or ED stays. As stay_id is generated using the earliest transfer_id existing for the department (e.g. the first transfer_id for an ICU), stay_id and transfer_id are often identical for the same ward stay.

Data flow through the hospital

Understanding how patients move through the hospital helps in understanding the data:

  1. Patient arrives → subject_id assigned
  2. Hospital admission → hadm_id assigned
  3. Unit transfer → stay_id assigned (ICU, ED, etc.)
  4. Data collection → Events recorded with appropriate IDs

Module characteristics

Hospital (hosp) module

  • Purpose: Hospital-wide EHR data
  • Key tables: patients, admissions, transfers, labevents, prescriptions, diagnoses_icd
  • Coverage: All hospital patients
  • Granularity: Order/event-level

ICU module

  • Purpose: Intensive care monitoring
  • Key tables: chartevents, inputevents, outputevents, procedureevents
  • Coverage: ICU patients only
  • Granularity: Hour-to-hour or more frequent

Emergency department (ED) module

  • Purpose: Emergency department care
  • Key tables: edstays, triage, vitalsign, medrecon, pyxis, diagnosis
  • Coverage: ED patients only
  • Granularity: Visit-level and event-level

Note module

  • Purpose: De-identified free-text clinical notes
  • Key tables: discharge, radiology (and their detail tables)
  • Coverage: Subset of hospitalized patients
  • Granularity: Note-level

CXR module

  • Purpose: Chest x-ray images and reports linked to MIMIC-IV
  • Key tables: lookup tables linking subject_id to study_id and dicom_id
  • Coverage: ED patients with chest radiographs
  • Granularity: Study- and image-level

ECG module

  • Purpose: Diagnostic 12-lead ECG waveforms and machine measurements
  • Key tables: record_list, machine_measurements, waveform_note_links
  • Coverage: Subset of patients with ECG recordings
  • Granularity: Study-level

Data relationships

One-to-many relationships

  • One patient → Many admissions
  • One admission → Many diagnoses
  • One admission → Many lab results
  • One ICU stay → Many vital sign measurements

Cross-module linking

Patients can be followed across modules using identifiers:

-- Link patient demographics to ICU data
SELECT p.gender, c.valuenum AS heart_rate
FROM `physionet-data.mimiciv_hosp.patients` p
JOIN `physionet-data.mimiciv_hosp.admissions` a ON p.subject_id = a.subject_id
JOIN `physionet-data.mimiciv_icu.icustays` i ON a.hadm_id = i.hadm_id
JOIN `physionet-data.mimiciv_icu.chartevents` c ON i.stay_id = c.stay_id
WHERE c.itemid = 220045  -- Heart rate

Temporal considerations

Time precision

  • Hosp: Usually day or hour precision
  • ICU: Minute-level precision common
  • ED: Varies by event type

For more on how time is represented in MIMIC-IV, including charttime vs. storetime and date shifting, see the Core concepts page.

Common analysis patterns

Patient cohort selection

  1. Start with the patients table for demographics
  2. Join to admissions for admission criteria
  3. Add module-specific criteria as needed

Longitudinal analysis

  1. Identify patient population
  2. Extract events from relevant modules
  3. Align timestamps for temporal analysis

Outcome assessment

  1. Define outcome from appropriate module
  2. Link back to patient characteristics
  3. Account for censoring and follow-up

Best practices

Query design

  • Always include appropriate time filters
  • Be mindful of data volume in the ICU module
  • Use indexed columns for joins when possible

Data validation

  • Check for reasonable value ranges
  • Validate identifier linkages
  • Account for missing data patterns

This schema overview provides the foundation for understanding MIMIC-IV. Each module has its own detailed documentation with table-specific information.