OpenCLIF

Open ICU Dataset Mappings for the Common Longitudinal ICU Data Format (CLIF)

OpenCLIF bridges the gap between CLIF's standardized clinical categories and publicly available ICU datasets. It provides dataset-specific identifiers that enable researchers to extract CLIF-compatible data from multiple open ICU databases using a single set of mappings.

Purpose

The Common Longitudinal ICU Data Format (CLIF) consortium has defined standard clinical categories (via the mCIDE specification) for ICU data. However, mapping these categories to specific datasets requires knowing the exact item IDs, variable names, or regex patterns for each database.

OpenCLIF solves this by:

1. Taking CLIF's mCIDE category definitions
2. Adding mappings from the ricu package's concept dictionary
3. Providing ready-to-use identifiers for 4 major open ICU datasets

Supported Datasets

Dataset	Source	Access
eICU-CRD	Philips eICU Research Institute	PhysioNet
HiRID	Bern University Hospital	PhysioNet
AmsterdamUMCdb	Amsterdam UMC	AmsterdamMedicalDataScience
SICdb	Salzburg University Hospital	PhysioNet

Mapping Structure

Each CSV file in the mappings/ directory contains:

Column	Description
*_category	CLIF category name (e.g., heart_rate, albumin, norepinephrine)
description	Clinical description from CLIF mCIDE
*_examples	Example names/strings from source systems
ricu_concept	Corresponding concept name in ricu's concept-dict.json
eicu_ids	Table/column or lab name for eICU-CRD
hirid_ids	Variable ID(s) for HiRID
aumc_ids	Item ID(s) for AmsterdamUMCdb
sic_ids	Item ID(s) for SICdb (Salzburg ICU Database)

ID Format Notes

• Numeric IDs: Direct item IDs (e.g., 711 for heart rate in SICdb)
• Multiple IDs: Separated by semicolons (e.g., 700; 703)
• Regex patterns: Prefixed with regex: (e.g., regex:^norepi)
• Column references: Prefixed with col: (e.g., col:heartrate)
• Empty values: No mapping available in ricu for that dataset

Directory Structure

OpenCLIF/
├── README.md
├── mappings/
│   ├── vitals/
│   │   └── clif_vitals_categories.csv
│   ├── labs/
│   │   └── clif_lab_categories.csv
│   ├── medications/
│   │   └── clif_medication_categories.csv
│   └── respiratory_support/
│       └── clif_respiratory_support_device_categories.csv
└── scripts/
    ├── build_openclif.py
    └── concept-dict.json

Usage Examples

Python - Extract heart rate from SICdb

import pandas as pd

# Load mappings
vitals = pd.read_csv('mappings/vitals/clif_vitals_categories.csv')

# Get SICdb item IDs for heart rate
hr_row = vitals[vitals['vital_category'] == 'heart_rate']
sic_ids = hr_row['sic_ids'].values[0]
# Returns: '711'

# Use in query
query = f"""
SELECT *
FROM data_float_h
WHERE DataID = {sic_ids}
"""

R - Use with ricu

library(ricu)
library(readr)

# Load OpenCLIF mappings
vitals <- read_csv("mappings/vitals/clif_vitals_categories.csv")

# Cross-reference with ricu concept
hr_concept <- vitals$ricu_concept[vitals$vital_category == "heart_rate"]
# Returns: "hr"

# Load using ricu
hr_data <- load_concepts("hr", "sic")

Mapping Coverage

Vitals (9 categories)

• ✅ All 9 categories mapped (100%)
• Full coverage: temp, heart_rate, sbp, dbp, spo2, respiratory_rate, map, height, weight

Labs (55 categories)

• ✅ ~35 categories mapped (~64%)
• Well covered: basic metabolic panel, liver function tests, CBC, coagulation
• Gaps: Some differential counts (absolute values), specialized markers

Medications (50 categories)

• ✅ ~8 categories mapped (~16%)
• Well covered: vasopressors (norepinephrine, epinephrine, vasopressin, dopamine, dobutamine, phenylephrine), insulin
• Gaps: Most sedatives, anticoagulants, paralytics (not tracked by rate in ricu)

Respiratory Support (9 categories)

• ✅ 2 categories mapped (~22%)
• Covered: IMV, NIPPV via mech_vent concept
• Gaps: CPAP, HFNC, other oxygen devices

Sources & Attribution

• CLIF mCIDE definitions: Common-Longitudinal-ICU-data-Format/skills
• Initial ID mappings reference: ricu concept-dict.json (eth-mds/ricu) - used as starting point, extended by OpenCLIF
• eICU-CRD: Pollard et al., PhysioNet
• HiRID: Hyland et al., PhysioNet
• AmsterdamUMCdb: Thoral et al., Amsterdam UMC
• SICdb: Salzburg ICU Database, PhysioNet

Note: OpenCLIF does not use ricu as a library dependency. The ricu concept-dict.json was used as a reference for initial mappings, but OpenCLIF maintains its own independent mapping configurations.

Contributing

Contributions are welcome! Areas that need work:

1. Adding missing mappings - Especially for sedatives, paralytics, and respiratory devices
2. Validation - Verifying mappings against actual dataset schemas
3. Unit conversions - Documenting unit differences between datasets
4. Additional datasets - MIMIC-IV, other open ICU databases

License

This project is part of the CLIF consortium. Mappings derived from ricu are subject to its license terms.

Related Projects

• CLIF - Core CLIF specification
• CLIFpy - Python tools for CLIF
• ricu - R interface for intensive care data
• CLIF-MIMIC - MIMIC to CLIF ETL

---

Built by the CLIF Consortium to accelerate ICU research across open datasets.