Written and maintained by CASRAI Editorial Board
Last updated
The National COVID Cohort Collaborative (N3C) is a centralized, secure clinical data harmonization and analytics infrastructure stewarded by the National Center for Advancing Translational Sciences (NCATS) at the National Institutes of Health (NIH). Whereas distributed research networks execute queries locally at each hospital and return aggregate counts, N3C established a cloud-hosted secure enclave (running on the FedRAMP High-authorized Palantir Foundry platform) that ingests longitudinal, patient-level electronic health record (EHR) data from dozens of academic medical centers and community health systems across the United States and harmonizes them into a single unified OMOP Common Data Model v5.3 repository.
Originally launched in 2020 to support rapid observational research on SARS-CoV-2 acute infection and Post-Acute Sequelae of SARS-CoV-2 (PASC / Long COVID), the N3C engineering architecture serves as the blueprint for the broader NCATS National Clinical Cohort Collaborative model across renal, oncology, and rare disease domains.
How Does N3C Harmonize OMOP, PCORnet, ACT, and TriNetX into OMOP CDM?
At the start of the pandemic, U.S. clinical and translational science centers did not share a single common data model; instead, participating sites maintained their local data warehouses in one of four major federated models: OHDSI OMOP, PCORnet CDM, ACT (Accrual to Clinical Trials / i2b2), or TriNetX. Rather than forcing sites to rebuild their local ETL pipelines, N3C engineered a centralized multi-CDM ingestion and harmonization pipeline inside the enclave:
| Source Data Model | Native Vocabulary & Schema Characteristics | N3C-to-OMOP v5.3 Harmonization Transformation |
|---|---|---|
| OHDSI OMOP CDM | Relational event tables already mapped to Athena Standard Concepts (v5.2 or v5.3). | Schema version alignment to OMOP v5.3.1, vocabulary version upgrading, and unit/value canonicalization. |
| PCORnet CDM | Sentinel-derived relational tables retaining native ICD-10-CM, CPT, HCPCS, NDC, RxNorm, and LOINC codes. | Maps DIAGNOSIS/CONDITION, PRESCRIBING/DISPENSING/MED_ADMIN, and LAB_RESULT_CM into target OMOP domains via concept_relationship joins. |
| ACT / i2b2 | Entity-attribute-value OBSERVATION_FACT table indexed by ontology path prefixes. |
Pivots fact rows by ontology domain and maps source codes to target OMOP Standard Concepts and visit hierarchies. |
| TriNetX | Standardized clinical fact exports with ICD, CPT, LOINC, and RxNorm coding. | Parses flat file tables into relational OMOP entities and resolves target domain routing driven by Athena concept domains. |
Domain Routing, Unit Harmonization, and Data Quality Gates
Simply translating table names is insufficient for cross-network analytics because clinical concepts frequently cross table boundaries depending on the source model (for example, an ICD-10-CM or CPT code stored in a source PROCEDURES table may map to an OMOP Standard Concept whose official Athena domain_id is Measurement, Observation, or Device). The N3C harmonization pipeline enforces domain-driven routing, moving every transformed record into the exact OMOP table dictated by its target Standard Concept’s domain_id.
In addition, because quantitative laboratory values (such as D-dimer, ferritin, C-reactive protein, and serum creatinine) arrive from different institutions with heterogeneous unit codes (`mg/L`, `ug/mL`, `ng/mL`, or missing units), N3C applies an automated unit harmonization pipeline. For key harmonized measurement variables, the pipeline inspects the LOINC code, source unit concept, and empirical value distribution to apply validated multiplicative conversion factors into a single canonical unit and flags physiologically implausible values before releasing tables to researchers.
What Are the Three N3C Data Access Tiers (Synthetic, De-identified, Limited Data Set)?
To balance broad scientific access with patient privacy and institutional governance, the N3C Data Enclave partitions the harmonized OMOP database into three tiered datasets:
- Tier 1 — Synthetic Data: Computationally generated synthetic OMOP tables that mirror the statistical distributions and cross-variable correlations of the real cohort without corresponding to any actual human individuals. Used for code development, pipeline testing, and educational training.
- Tier 2 — De-identified Data (HIPAA Safe Harbor): Patient-level longitudinal OMOP records in which all dates of service for a given patient are shifted by a consistent site-specific random offset (up to ±180 days) to preserve exact inter-event intervals, ages over 89 are top-coded, and residential geography is truncated to 3-digit ZIP code prefixes (masked entirely if the 3-digit prefix population is ≤20,000).
- Tier 3 — HIPAA Limited Data Set (LDS): Retains true, unshifted calendar dates of clinical encounters, laboratory tests, and vaccinations alongside full 5-digit residential ZIP codes. Accessing Tier 3 requires an active Institutional DUA, a project-specific workspace proposal approved by the N3C Data Access Committee (DAC), and an institutional IRB determination for the specific research protocol.
Under N3C’s Download Policy, individual patient-level rows can never be exported from any tier of the enclave; researchers perform all cohort building, statistical modeling (in Python, R, and Spark SQL), and visualization inside the browser-based enclave workspace, and may export only aggregate results and summary figures that satisfy minimum cell-size suppression rules (suppressing counts below 20).
How Does N3C Use Privacy-Preserving Record Linkage (PPRL)?
Because patients frequently seek care across multiple health systems—or have mortality, viral sequencing, and vaccination records stored in external state or federal registries—N3C integrates Privacy-Preserving Record Linkage (PPRL):
- Participating sites generate irreversible cryptographic hashes (tokens) from local patient identifiers (name, date of birth, sex, address) behind their own firewall using secret salts before transmitting any data.
- An independent Linkage Honest Broker matches cryptographic tokens across institutions without ever seeing either the raw patient identifiers or the clinical OMOP data inside the enclave.
- This architecture enables N3C to deduplicate patients who visited multiple contributing health systems and to link EHR cohorts with external datasets including Social Security Death Master File / mortality registries, CMS Medicare/Medicaid claims, NCI SEER cancer registries, and SARS-CoV-2 viral variant sequencing repositories.
Frequently Asked Questions
Why does N3C shift dates in Tier 2 instead of removing timestamps altogether?
In longitudinal infectious disease and pharmacoepidemiology studies, researchers must calculate exact elapsed days between symptom onset, laboratory testing, drug administration, ICU admission, and post-acute sequelae. Shifting all timestamps for a given patient by the exact same random integer offset preserves every within-patient time interval (e.g., 14 days from positive PCR test to hospitalization) while preventing re-identification via exact admission dates. However, studies examining seasonal epidemiological waves, viral variant eras, or environmental exposures require unshifted Tier 3 (LDS) dates.
What are N3C Concept Sets and Knowledge Store objects?
Inside the N3C Enclave, multidisciplinary Clinical Domain Teams curate, peer-review, and version-control reusable OMOP Concept Sets (for comorbidities, medications, laboratory panels, and PASC phenotypes) along with shared analytic code templates in the enclave’s Knowledge Store, ensuring that independent research teams apply consistent, reproducible phenotype definitions.
How are contributing site identities protected in N3C analyses?
Each contributing institution is represented only by an obfuscated numeric data_partner_id (alongside macro-level site metadata such as source CDM type). The N3C Community Guiding Principles prohibit researchers from attempting to re-identify individual contributing health systems or benchmarking one hospital’s clinical outcomes against another’s.
