OSIRIS RWD — Common Data Model for Real-World Oncology Data

Version 2.0.0 · June 2026 · An INCa (Institut National du Cancer) project, coordinated with the support of Unicancer

OSIRIS RWD is a common data model for Real-World Data in oncology. It defines a minimum, standardized, event-based schema for sharing clinical oncology data extracted from hospital information systems (HIS / SIH), with the explicit goal of enabling automatic, unambiguous population of the database rather than manual data entry.

The model was specified by a multi-institution Working Group (comprehensive cancer centres, university hospitals, and partner organizations). It is freely released to the community and is expected to evolve with usage and field feedback.

---

Why OSIRIS RWD

OSIRIS RWD targets routine-care, real-world conditions, where data quality, completeness, and provenance differ substantially from curated trial data. The design rests on a few explicit principles:

• Automatic data population first. Every variable has a precisely defined population rule so it can be populated programmatically from existing HIS sources without ambiguity.
• Structured sources over unstructured ones. Coded administrative/clinical data (e.g. PMSI) is always preferred to free text (e.g. reports) when both are available.
• International terminologies are used wherever they exist.
• Referentials are chosen pragmatically — by their actual use in HIS, the availability of crosswalks, ANS (Agence du Numérique en Santé) recommendations, granularity, and maintenance status.
• FAIR principles for scientific data management.
• Provenance is traceable. A dedicated Metadata table records source, terminology, NLP system, and quality metrics for any attribute — essential when part of the data is extracted by NLP.

---

Repository contents

Artifact	File	Purpose
Minimum Data Set (MDS)	OSIRIS_RWD_Minimum_DataSet__MDS__*.xlsx	The authoritative variable list: tables, variables, definitions (FR/EN), mandatory status, expected types/formats, value sets and terminologies.
Data population guide	OSIRIS_RWD_Guide_d_alimentation_*.docx	Per-variable population rules, edge cases, examples, generic rules, and annexes. (Currently in French.)
Physical model	OSIRIS_RWD_Physical_model_*.png	Entity-relationship diagram (keys, cardinalities, foreign keys).
Changelog	OSIRIS_RWD_Changelog_*.xlsx	Version-to-version changes, each tagged with a GitHub issue number and a retro-compatibility impact flag.
REDCap data dictionary	redcap/OSIRIS_RWD_REDCap_DataDictionary_*.csv OR redcap/OSIRIS_RWD_REDCap_Project_*.xml	REDCap-importable definition of the model as an eCRF — instruments, fields, types, value sets, and branching logic.

---

Model at a glance

OSIRIS RWD is a patient-centric, event-based star schema.

• Patient is the only mandatory link. In real-world data, the patient is the sole relationship known with certainty; every other table attaches to Patient (directly or through a parent entity).
• 27 tables in the physical model; 114 documented variables, of which 73 are technically mandatory (per the MDS).
• A Patient row is unique on PatientID and belongs to exactly one Dataset.

Table catalogue (by domain)

Metadata & provenance

• Dataset — describes the data extract (origin center, name, derived update date). Unique on (OriginCenterID, DatasetName).
• Metadata — describes any attribute in the base: source, vocabulary, NLP system / lexical variant / quality measure / score, and the Parent{Concept,ID,Variable} it attaches to.

Patient & follow-up

• Patient — birth date, biological sex, update date.
• Opposition — patient information & objection to secondary data reuse (one row per patient).
• LastNews — vital status, death/last-visit/last-contact dates (one row per patient).

Risk & history

• RiskFactor / Infection — IARC Group 1 carcinogens and oncogenic infectious agents.
• FamilyCancerHistory — 1st/2nd-degree family cancer history.
• RelatedPathology — medical history, comorbidities, and adverse events (ICD-10), merged into one item.

Cancer

• PrimaryCancer — topography/morphology (ICD-O-3.2), groups, laterality, diagnosis date/method. Tumor uniqueness follows IACR rules on (TopographyGroup, MorphologyGroup, Laterality).
• TumorSize, TNMEvent, PrimaryCancerStage, PrimaryCancerGrade.
• TumorPathoEvent (tumor events: relapse/progression) and Metastasis.
• Progression.

Treatments

• Surgery — surgery & interventional radiology procedures (CCAM).
• Medication — systemic therapies (ATC), clinical-trial / early-access flags.
• Radiotherapy / RadiatedOrgan.

Specimens, biomarkers & exams

• BiologicalSpecimen, Biomarker (incl. HGNC/UniProt gene & protein references).
• Analysis, ImagingSeries (DICOM-aligned imaging/nuclear-medicine).

Generic extension tables

• Measure and Observation — generic numeric/qualitative facts attached to any table row via Parent{Concept,ID}, terminology-flexible (SNOMED CT / NCIt / LOINC depending on the concept).

The clinical core can be extended with additional modules (e.g. genomic, radiomic) where needed.

---

Generic rules (must-read before implementing)

• Dates are split into three attributes (day / month / year). When a collected date is imprecise, leave the missing parts empty — except the year, which is always required. (The eCRF imputation convention of 15/06 applies only inside the eCRF and must remain distinguishable from real data.)
• Missing values are left empty. Only explicit data is collected; there is no imputation or computation of missing information. Non-applicable vs unknown is distinguished only in the eCRF (NA / UK).
• Update semantics. All information for a patient is updated together; PatientUpdateDate carries the timestamp. DatasetUpdateDate is derived as the max over the dataset's patients.
• Pseudonymisation is the implementer's responsibility. The local IPP is directly identifying; dates of birth/care/death are quasi-identifiers. Each project using the model is responsible for regulatory compliance of secondary use (pseudonymisation, patient information, objection handling). See the EDPB pseudonymisation guidance.
• Terminologies. When a terminology is specified for an item, the expected values are codes from that terminology, formatted as [terminology]_[version] in Metadata.Vocabulary.

---

Data capture with REDCap

A REDCap data dictionary (CSV) and project template (XML) are available in this repository under redcap/. They provide the REDCap-native, importable definition of OSIRIS RWD as an electronic case report form (eCRF), intended primarily to assess the quality of automatically structured data and to support manual capture where needed.

It provides:

• One REDCap instrument per OSIRIS RWD table, with fields mapping 1:1 to MDS variables (matching variable names, types, and value sets).
• Field-level validation (date parts as day/month/year, numeric ranges, formats such as 9-digit FINESS) and terminology-backed choice lists (e.g. ICD-O-3.2, ICD-10, ATC, CCAM) exported as REDCap code lists.
• Branching logic and repeating instruments to reflect model cardinalities (e.g. 0..n primary cancers, treatments, biomarkers per patient).
• eCRF-specific conventions that differ from the database rules — notably the imprecise-date imputation (15 for day, 06 for month) used only inside the eCRF, and the explicit NA / UK distinction for non-applicable vs unknown values, both of which must remain identifiable as eCRF artifacts rather than source data.

Deployment & Installation

1. Project Creation: Create a new REDCap project using the provided XML file.
2. Data Dictionary Import: Log in with administrator privileges and import the Data Dictionary (.csv format) into the newly created project.
3. Center Configuration:
   • Update the origincenterid item value in the Patient form to match the participating center's FINESS code (9 digits).
   • Single-center setup: You can set the default FINESS code using the REDCap Action Tag on the item:

     @DEFAULT='finess_code'
     

     (replace finess_code with the center's 9-digit FINESS code).

The data dictionary will be versioned and changelogged alongside the rest of the model; breaking changes affecting it are flagged in the changelog like any other artifact.

---

Terminologies & referentials

The MDS value sets reference, among others:

ICD-O-3.2 (topography & morphology) · ICD-10 · ATC · CCAM · LOINC · SNOMED CT · NCIt · HGNC · UniProt · UCUM · DICOM · FINESS (center identification) · ENCR.

Codes intended for the French context default to HIS-aligned referentials (FINESS, CCAM, ICD-10/PMSI) with ANS recommendations and crosswalk availability as selection criteria.

---

Versioning

OSIRIS RWD follows semantic versioning (MAJOR.MINOR.PATCH). The current release is 2.0.0.

Every documented change is tracked against a GitHub issue and flagged for retro-compatibility impact in the changelog. Changes touching the physical model, MDS, data population guide, or REDCap material are listed there. When implementing, always read the changelog for the version you are migrating from — several v2.0.0 changes (e.g. composite Dataset key, mandatory Opposition, removal of "mandatory status" items, Imaging → ImagingSeries rename) are breaking.

---

Repository structure (TO BE VALIDATED)

.
├── README.md
├── CHANGELOG.md                # or the .xlsx changelog, summarized
├── mds/                        # Minimum Data Set (xlsx + exported CSV/value sets)
├── data-population-guide/      # data population guide (docx / PDF / md)
├── physical-model/             # ER diagram (png/svg) + DDL if available
├── redcap/                     # REDCap data dictionary (CSV) + code lists — forthcoming
└── modules/
    ├── genomic/
    └── radiomic/

---

Contributing

Contributions follow an open, issue-based process: propose new concepts or updates to existing ones via GitHub issues. Each accepted change is recorded in the changelog with its issue number and retro-compatibility impact. (Document your local contribution/RFC and conformance policy in CONTRIBUTING.md / GOVERNANCE.md.)

---

Citation (TO BE VALIDATED)

A reference publication for OSIRIS RWD is forthcoming. Until it is available, please cite the model directly, e.g.:

OSIRIS RWD — Common Data Model for Real-World Oncology Data, version 2.0.0 (2026). INCa project, coordinated with the support of Unicancer. Available at: https://github.com/InstitutNationalduCancer/OSIRIS_WG_OSIRIS-RWD.

This section will be updated with the full citation once the publication is released.

---

License (TO BE VALIDATED)

OSIRIS RWD is released under the Creative Commons Attribution-ShareAlike 4.0 International License (CC BY-SA 4.0).

You are free to share and adapt the material, including for commercial purposes, provided that you (1) give appropriate attribution and (2) distribute any adapted material under the same license (ShareAlike). Note that CC BY-SA also covers Sui Generis Database Rights, which is appropriate for a data model and its value sets.

See the LICENSE file for the full text, or https://creativecommons.org/licenses/by-sa/4.0/.

Attribution example: "OSIRIS RWD", an INCa project coordinated with the support of Unicancer, licensed under CC BY-SA 4.0.

---

Acknowledgements

OSIRIS RWD is a project of the Institut National du Cancer (INCa), coordinated with the support of Unicancer. The model was defined by the OSIRIS RWD Working Group, gathering members from comprehensive cancer centres, university hospitals, and partner organizations (full list in the data population guide annex).