Bridging the ICU Data Gap: Extending OMOP CDM with SCCM’s Critical Care Data Dictionary (C2D2)

Highlight

– Nearly half of SCCM’s C2D2 elements map directly to OMOP concepts; most remaining elements need modification or extension.
– Ventilator parameters, composite scores, and temporally precise measurements are the main sources of incompatibility.
– Large language model semantic matching showed high recall but modest precision, indicating a helpful but imperfect tool for mapping.
– Communities must choose between extending OMOP, adapting C2D2 for OMOP compatibility, pursuing vocabulary inclusion via OHDSI, or partnering with EHR vendors for native standards support.

Background

High-fidelity, interoperable critical care data are essential for observational research, clinical decision support, quality improvement, and device/regulatory evaluation. Intensive care units (ICUs) generate dense, temporally granular data (physiologic waveforms, ventilator settings and modes, advanced organ support parameters, and composite severity scores). Standardizing these elements is challenging but necessary to enable multi-center observational studies and scalable analytics.

The Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) has become a widely used framework for harmonizing routine clinical data to support reproducible analyses across institutions. The Society of Critical Care Medicine (SCCM) developed the Critical Care Data Dictionary (C2D2) via expert consensus to curate a clinically prioritized set of ICU variables. The study by Adams et al. (2025) evaluated compatibility between C2D2 and OMOP CDM and explored steps needed to extend OMOP for critical care use cases.

Study design

Adams et al. conducted a systematic mapping and semantic analysis of 226 variables from SCCM’s C2D2 to the OMOP CDM. A three-tiered semantic matching approach classified each variable as a full match (direct OMOP equivalent), partial match (requires modification or extension), or no match (no OMOP representation). The study also evaluated an automated, large language model (LLM)-based semantic matching tool and assessed its performance at different similarity thresholds.

Key findings

Overall mapping results

Of 226 C2D2 elements, 49.6% were classified as full matches to existing OMOP concepts, 46.4% required modification (partial matches), and 4.0% had no suitable OMOP mapping. These percentages translate to roughly half of curated ICU concepts being directly representable in OMOP, with nearly half needing adaptation.

Major incompatibility domains

The principal areas of OMOP incompatibility were:

• Ventilator parameters and modes. Many C2D2 concepts encode mode-specific settings (e.g., pressure support parameters, waveforms, trigger settings) and device-reported states that lack 1:1 OMOP concept equivalents.
• Composite scores and derived clinical instruments. Elements such as aggregated organ support indices or composite severity scores (including customized or age-specific variants) posed challenges because OMOP typically represents atomic observations rather than curated composite constructs.
• Temporal precision. ICU data often require sub-minute timestamping and event-level granularity (e.g., ventilator changes, alarms, and waveform-associated events), whereas OMOP’s standard use has emphasized encounter-level and daily-level observations.
• Specialized constructs from the SCCM Delphi process. Some concepts were unique to critical care practice or defined by age-specific criteria (pediatric thresholds), which were absent from OMOP vocabularies.

Automated semantic matching performance

The study deployed an LLM-based semantic matching system to suggest OMOP equivalents and optimize mapping throughput. At an optimized similarity threshold of 0.90, system performance was measured as precision 59.5%, recall 87.0%, and F1 score 70.7%. These results indicate that the LLM prioritized sensitivity (finding most potential matches) while producing a significant number of false positives that required manual curation.

Interpretation of mapping categories

Full matches encompassed vitals, common laboratory tests, medication administrations, and many demographic and administrative elements already well represented in OMOP. Partial matches often required either (a) additional OMOP standard concepts/vocabularies to express device-specific parameters, (b) new attributes or table extensions (for temporally dense observations), or (c) standardized ways to represent stacked concepts — multiple clinical pieces encoded within a single C2D2 variable (for example, ‘ventilator mode + settings + alarm state’). No-match items were uncommon but included highly specialized ICU constructs curated specifically for critical care practice without broad adoption in standard healthcare vocabularies.

Expert commentary and critical appraisal

Adams et al. provide valuable empirical data quantifying the gap between a specialty-curated ICU dictionary and a widely used general-purpose CDM. The balanced mapping outcomes — roughly half full, half partial/no match — confirm OMOP’s baseline utility while highlighting substantive extension needs for critical care work.

Strengths of the analysis include systematic classification, transparent metrics, and exploration of an AI-assisted mapping workflow. The LLM assistance demonstrated that modern NLP tools can accelerate mapping but cannot replace domain expert curation due to modest precision and the clinical risk of incorrect mappings.

Limitations include analysis constrained to one curated dictionary (C2D2) and potential subjectivity in classification (full vs partial match). Additionally, mapping feasibility does not equate to clinical readiness: implementing ICU-specific OMOP extensions will require governance, consensus on representation (atomic vs composite), and maintenance processes.

Implications for the community

Several strategic pathways emerge, each with trade-offs:

• Extend OMOP CDM for critical care: Add new tables or vocabulary extensions to capture ventilator settings, waveform-driven events, device states, and sub-minute timestamps. Pros: leverages OMOP analysis ecosystem and multi-center comparability. Cons: requires OHDSI governance, versioning, and long-term maintenance.
• Adapt C2D2 for OMOP v2 compatibility: Recast C2D2 elements into atomic observations and map to existing OMOP concepts where possible. Pros: fewer changes to OMOP core. Cons: may lose clinically meaningful composite constructs and require local transformation logic.
• Pursue vocabulary inclusion via OHDSI processes: Propose new standard concepts to the OMOP vocabulary repository for acceptance and distribution. Pros: formalizes critical care concepts within OMOP standards. Cons: requires time, community review, and acceptance criteria compliance.
• Partner with EHR vendors to embed native critical care standards: Work with vendors to provide structured ICU data exports that align with OMOP or C2D2. Pros: potentially higher fidelity and less downstream transformation. Cons: vendor heterogeneity and variable adoption across sites.

Practical recommendations and prioritized next steps

Based on the study’s findings and stakeholder considerations, a practical, phased approach is advisable:

1. Convene a cross-disciplinary working group (SCCM, OHDSI, ICU informaticians, EHR vendors) to prioritize a minimal viable critical care extension for OMOP, focusing on high-impact items: ventilator modes/parameters, composite score representation, and temporal precision attributes.
2. Define representation patterns: agree whether to represent composite scores as precomputed derived concepts (preserving clinical meaning) plus their atomic components, or to strictly decompose them into atomic observations with standard derivation metadata.
3. Create an OMOP extension package and pilot implementation in one or more centers with mature ICU data capture; document transformation rules, versioning strategy, and validation checks (including waveform-to-OMOP linkage where applicable).
4. Use hybrid mapping workflows: LLM-assisted candidate generation followed by domain expert adjudication to maximize throughput while maintaining accuracy.
5. Publish mapping conventions and tooling, and submit vetted concepts to OHDSI vocabulary processes to enable broader adoption and maintenance.

Operational and governance considerations

Extending OMOP for critical care will require investments in:

• Vocabulary governance: clear stewardship of newly defined critical care concepts and their lifecycle.
• Technical tooling: ETL templates, validation suites, and tooling to preserve temporal resolution and device-linkage semantics.
• Training and documentation: guidance for sites implementing ICU OMOP extensions to ensure consistent semantics across centers.
• Maintenance and community support: processes to update concepts, manage backward compatibility, and address new ICU technologies (e.g., novel devices, monitoring modalities).

Conclusions

Adams et al. demonstrate that mapping a specialty-curated ICU dictionary to OMOP is technically feasible but not trivial. Nearly half of C2D2 concepts map directly to OMOP, while many critical care–specific elements require deliberate extensions, representation conventions, and governance. The high recall but moderate precision of LLM-based mapping tools indicates such approaches can accelerate work but cannot obviate expert curation. The broader critical care and informatics communities must decide among competing strategies — OMOP extension, C2D2 adaptation, OHDSI vocabulary inclusion, or vendor collaboration — weighing technical feasibility, long-term sustainability, and the need for multi-center interoperability.

Funding and clinicaltrials.gov

Funding: Not specified in the source summary provided. ClinicalTrials.gov: Not applicable.

References

1. Adams MCB, Hurley RW, Bartels K, Perkins ML, Hudson C, Topaloglu U, Cobb JP, Reuter-Rice K, Stocking JC, Khanna AK. Extending the Observational Medical Outcomes Partnership (OMOP) Common Data Model for Critical Care Medicine: A Framework for Standardizing Complex ICU Data Using the Society of Critical Care Medicine’s Critical Care Data Dictionary (C2D2). Crit Care Med. 2025 Nov 21. doi: 10.1097/CCM.0000000000006969 . Epub ahead of print. PMID: 41269063 .

2. Observational Health Data Sciences and Informatics (OHDSI). OMOP Common Data Model. https://ohdsi.org/data-standardization/the-common-data-model/ (accessed 2025).

3. Society of Critical Care Medicine (SCCM). Critical Care Data Dictionary (C2D2). SCCM resources and working group publications (accessed 2025).