MIMIC-IV to eICU Mapping

This skill maps equivalent tables, columns, and concepts between MIMIC-IV and eICU databases to enable cross-database research and external validation.

When to Use This Skill

• External validation of MIMIC-IV models on eICU
• Multi-center studies combining both databases
• Understanding structural differences
• Translating queries between databases

Database Overview

Characteristic	MIMIC-IV	eICU
Institution	Beth Israel Deaconess (single center)	208 hospitals (multi-center)
Patients	~300,000 admissions	~200,000 patients
Time Period	2008-2019	2014-2015
ICU Types	All ICU types	All ICU types
Geography	Boston, MA	United States (nationwide)

Identifier Mapping

Concept	MIMIC-IV	eICU
Patient ID	subject_id	uniquepid
Hospital Admission	hadm_id	patienthealthsystemstayid
ICU Stay	stay_id	patientunitstayid
Hospital ID	N/A (single center)	hospitalid
Unit Visit	icustay_seq	unitvisitnumber

Core Table Mapping

Patient Demographics

MIMIC-IV	eICU	Notes
mimiciv_hosp.patients	patient	Base demographics
mimiciv_hosp.admissions	patient	Admission info combined
mimiciv_icu.icustays	patient	ICU stay info in patient table
mimiciv_derived.icustay_detail	icustay_detail (concept)	Derived convenience table

Vital Signs

MIMIC-IV	eICU	Notes
mimiciv_derived.vitalsign	vitalperiodic, vitalaperiodic	eICU splits periodic/aperiodic
mimiciv_icu.chartevents	nursecharting	Raw charted values

Laboratory Values

MIMIC-IV	eICU	Notes
mimiciv_hosp.labevents	lab	Different labname conventions
mimiciv_derived.chemistry	pivoted_lab (concept)	Derived/pivoted
mimiciv_derived.complete_blood_count	pivoted_lab (concept)

Medications

MIMIC-IV	eICU	Notes
mimiciv_hosp.prescriptions	medication	Hospital medications
mimiciv_icu.inputevents	infusiondrug	IV infusions
mimiciv_derived.antibiotic	-	Concept needs creation for eICU

Diagnoses

MIMIC-IV	eICU	Notes
mimiciv_hosp.diagnoses_icd	diagnosis	eICU uses text descriptions
-	admissiondx	eICU has admission diagnosis
-	apacheapsvar	APACHE diagnosis categories

Concept Availability

Available in Both (May Require Recalculation)

Concept	MIMIC-IV	eICU
SOFA	mimiciv_derived.sofa	Requires custom calculation
APACHE IV	Not pre-computed	apachepatientresult
OASIS	mimiciv_derived.oasis	pivoted_oasis (concept)
GCS	mimiciv_derived.gcs	pivoted_score (concept)
Urine Output	mimiciv_derived.urine_output	pivoted_uo (concept)

MIMIC-IV Only

• Waveform data (vital sign waveforms)
• Radiology reports
• Detailed microbiology (organism/sensitivity)
• ED data

eICU Only

• APACHE IV scores (pre-computed)
• Multi-center hospital data
• Respiratory care plan documentation
• Nurse care plan

Key Structural Differences

1. Time Representation

MIMIC-IV: Absolute timestamps (DATETIME)
eICU: Offset in minutes from unit admission (INTEGER)

-- MIMIC-IV
WHERE charttime BETWEEN ie.intime AND ie.outtime

-- eICU (convert offset to time)
WHERE chartoffset >= 0 AND chartoffset <= unitdischargeoffset

2. Hospital Structure

MIMIC-IV: Single hospital, no hospital identifier
eICU: hospitalid links to hospital table with region, bed count

3. Diagnosis Coding

MIMIC-IV: ICD-9 and ICD-10 codes
eICU: Free-text diagnosis strings + APACHE categories

4. Lab Value Names

MIMIC-IV: itemid (numeric codes) with d_labitems lookup
eICU: labname (text strings), less standardized

Example: Equivalent Queries

First ICU Stay Selection

MIMIC-IV:

SELECT *
FROM mimiciv_derived.icustay_detail
WHERE first_icu_stay = TRUE;

eICU:

SELECT *
FROM patient
WHERE unitvisitnumber = 1;

Mortality Outcome

MIMIC-IV:

SELECT stay_id, hospital_expire_flag
FROM mimiciv_hosp.admissions a
INNER JOIN mimiciv_icu.icustays ie ON a.hadm_id = ie.hadm_id;

eICU:

SELECT patientunitstayid,
       CASE WHEN hospitaldischargestatus = 'Expired' THEN 1 ELSE 0 END AS hosp_mort
FROM patient;

Age

MIMIC-IV:

SELECT stay_id, admission_age
FROM mimiciv_derived.icustay_detail;
-- Note: Ages > 89 are shifted

eICU:

SELECT patientunitstayid,
       CASE WHEN age = '> 89' THEN 90 ELSE CAST(age AS INT) END AS age
FROM patient;
-- Note: age is stored as string, "> 89" for elderly

Validation Considerations

1. Population Differences: eICU is multi-center with different case-mix
2. Time Period: Different years may have different practices
3. Documentation Patterns: Single vs multi-center charting variability
4. Missing Data: Different missingness patterns
5. Outcome Definitions: Verify mortality/LOS definitions match

References

• Johnson AEW et al. "MIMIC-IV, a freely accessible electronic health record dataset." Scientific Data. 2023.
• Pollard TJ et al. "The eICU Collaborative Research Database." Scientific Data. 2018.