Written and maintained by CASRAI Editorial Board
Last updated
OHDSI (Observational Health Data Sciences and Informatics)—pronounced ‘Odyssey’—is an international, multi-stakeholder, open-science collaborative founded in 2014 and coordinated at Columbia University’s Department of Biomedical Informatics to generate reproducible real-world evidence (RWE) from large-scale observational health databases. By combining the OMOP Common Data Model (CDM) with an open-source software stack—including the ATLAS cohort design platform, Achilles and the DataQualityDashboard (DQD), and the HADES (Health Analytics Data-to-Evidence Suite) R package library—OHDSI enables federated network studies across hundreds of electronic health record (EHR) and administrative claims repositories worldwide without transferring patient-level Protected Health Information (PHI).
What Is OHDSI and How Is It Used in Clinical Research and Healthcare?
Established as the open-community successor to the public-private Observational Medical Outcomes Partnership (OMOP), OHDSI organizes observational research around three standardized methodological pillars when investigators design a cohort study or retrospective cohort study:
- Clinical Characterization (What happened to them?): Descriptive epidemiology summarizing disease natural history, treatment utilization pathways, baseline comorbidity distributions, and population-level incidence rates across multi-national health systems.
- Population-Level Effect Estimation (What are the causal effects?): Comparative effectiveness and active medical product safety surveillance comparing new-user exposure cohorts (evaluating hazard ratios or incidence rate ratios while controlling for measured and unmeasured confounding).
- Patient-Level Prediction (What will happen to me?): Development and external transportability validation of individualized clinical risk prediction models estimating a patient’s probability of experiencing a defined outcome within a target time-at-risk window.
How Federated OHDSI Network Studies Execute Without Sharing PHI
Unlike centralized data warehouses that pool patient rows in a single cloud enclave, an OHDSI network study operates on a federated execution model:
- A study coordinator designs a phenotype and statistical analysis protocol in ATLAS or R (using
CaprandStrategus) and publishes the self-contained study package to a public GitHub repository. - Participating data partners download the open-source R study package and execute it locally behind their institutional firewall against their OMOP CDM v5.4 database.
- Only de-identified, aggregate summary statistics (such as propensity-score-adjusted log hazard ratios, likelihood profiles, diagnostics, and cell-count-suppressed tables meeting minimum count thresholds, typically
n ≥ 5) are returned to the study coordinating center for meta-analysis.
The OHDSI Open-Source Software Stack: ETL, ATLAS, and HADES
How does OHDSI integrate with EHR and health informatics workflows? Health systems transform raw EHR tables (such as Epic Clarity/Caboodle or Cerner Millennium) and HL7 FHIR R4 resources into an analytical OHDSI node using a six-part open-source toolchain:
| OHDSI Tool / Suite | Primary Role in the Research Pipeline | Key Technical Components & Functions |
|---|---|---|
| Athena & Standardized Vocabularies | Semantic harmonization engine | Maps local source codes (ICD-10-CM, CPT-4, NDC, HCPCS) to standard concepts in SNOMED CT (conditions), RxNorm / RxNorm Extension (drugs), and LOINC (measurements). |
| WhiteRabbit & Rabbit-in-a-Hat | ETL profiling and schema mapping | WhiteRabbit scans source EHR/claims tables to profile column distributions and value frequencies; Rabbit-in-a-Hat provides an interactive GUI to document table-to-table and field-to-field OMOP CDM v5.4 ETL specifications. |
| Achilles & DataQualityDashboard (DQD) | Database characterization and quality verification | Achilles computes pre-aggregated descriptive summaries; DataQualityDashboard runs over 3,300 Kahn-framework checks verifying Plausibility, Conformance, and Completeness. |
| ATLAS & WebAPI | Browser-based cohort and study design | Defines computable phenotypes via Cohort Entry Events (index date), Inclusion Criteria, Cohort Exit (end of continuous observation, fixed offset, or drug era persistence), and Cohort Eras. |
| SqlRender & DatabaseConnector | Cross-platform SQL dialect translation | Translates a single parameterized OHDSI SQL dialect into PostgreSQL, Microsoft SQL Server, Oracle, Amazon Redshift, Snowflake, Apache Spark/Databricks, or Google BigQuery. |
| HADES (Health Analytics Data-to-Evidence Suite) | Validated R statistical packages | Includes CohortGenerator, CohortDiagnostics, CohortMethod, SelfControlledCaseSeries (SCCS), PatientLevelPrediction (PLP), and EmpiricalCalibration. |
What Standards and Regulatory Requirements Govern OHDSI?
Because OHDSI network studies play an expanding role in regulatory pharmacovigilance and real-world evidence submissions alongside randomized controlled trials vs. observational studies, OHDSI tools are engineered to satisfy international regulatory and reporting frameworks:
- FDA Sentinel and EMA DARWIN EU Alignment: The European Medicines Agency’s DARWIN EU (Data Analysis and Real World Interrogation Network) uses the OMOP CDM and OHDSI analytical tools as its foundational federated data network standard, while FDA real-world evidence guidance and the Sentinel Initiative collaborate closely with OHDSI on computable phenotyping and methodological validation.
- Methodological Bias Control via LEGEND Principles: Following OHDSI’s Large-Scale Evidence Generation and Evaluation across a Network of Databases (LEGEND) principles,
CohortMethodfits large-scale regularized propensity score models (via theCyclopshigh-performance solver using L1 Lasso penalties across tens of thousands of baseline covariates), verifies covariate balance (requiring standardized mean differenceSMD < 0.1) and clinical equipoise, and appliesEmpiricalCalibrationusing dozens of negative control outcomes to calibrate p-values and 95% confidence intervals against residual systematic error. - Observational Reporting & Privacy Governance: Study protocols are pre-registered (e.g., on the EU PAS / HMA-EMA Catalogues or OSF), reported in alignment with the STROBE checklist and RECORD-PE guidelines, and governed locally by each site’s Institutional Review Board (IRB) and HIPAA de-identification or Limited Data Set controls.
Frequently Asked Questions
What is OHDSI and how is it used in clinical research and healthcare?
OHDSI (Observational Health Data Sciences and Informatics) is an open-science global research collaborative that maintains the OMOP Common Data Model (CDM), Standardized Vocabularies, the ATLAS web platform, and the HADES R package suite to execute reproducible, federated observational studies across EHR and claims databases without sharing patient-level PHI.
What standards or regulatory requirements govern OHDSI?
OHDSI studies rely on the OMOP CDM v5.4 and standard biomedical terminologies (SNOMED CT, RxNorm, LOINC), Kahn-framework data quality verification via the DataQualityDashboard, LEGEND methodological standards (large-scale propensity scores and negative-control empirical calibration), STROBE/RECORD reporting guidelines, and regulatory RWE frameworks including EMA DARWIN EU and FDA Real-World Evidence guidance.
How does OHDSI integrate with EHR and health informatics workflows?
Health informatics teams profile source EHR or HL7 FHIR databases with WhiteRabbit, design ETL mappings in Rabbit-in-a-Hat, map local ICD-10/CPT/NDC codes to Athena Standardized Vocabularies, validate the resulting OMOP CDM instance with Achilles and DataQualityDashboard, and deploy ATLAS and HADES R packages via SqlRender across cloud or on-premise SQL warehouses.
Sources
- Observational Health Data Sciences and Informatics (OHDSI): The Book of OHDSI: Observational Health Data Sciences and Informatics (
https://ohdsi.github.io/TheBookOfOhdsi/). - OHDSI HADES Documentation: Health Analytics Data-to-Evidence Suite (HADES) R Packages — CohortMethod, PatientLevelPrediction, and DataQualityDashboard (
https://ohdsi.github.io/Hades/). - European Medicines Agency (EMA): Data Analysis and Real World Interrogation Network (DARWIN EU) Common Data Model Specifications.
