Oncology Data Standards
Educational note: This page overviews healthcare data interoperability as it applies to oncology (2025).
TL;DR
To use data to fight cancer, that data must be computable. However, oncology data in Electronic Health Records (EHRs) is notoriously messy, locked in unstructured text (pathology reports, clinical notes), and siloed. Data standards are the syntaxes and vocabularies that allow algorithms to understand that a "Stage IIIA Breast Carcinoma" in Hospital A is the exact same clinical entity as an "ER+ ductal ca, stg 3a" in Hospital B.
1. mCODE (Minimal Common Oncology Data Elements)
mCODE is the most important standard specifically built for cancer. Historically, EHRs tracked billing codes perfectly but failed to track the most vital clinical variables in a standardized way. mCODE defines a core set of structured data elements for oncology that every EHR should be able to capture and share.
Core elements include:
- Patient info: Demographics.
- Disease: Primary cancer site, histology, stage (T, N, M), and grade.
- Treatments: Surgery, radiation, systemic therapies (chemo/targeted/immuno), and their intent (curative vs palliative).
- Outcomes: Disease status (e.g., complete response, progression) and survival.
- Genomics: Key actionable mutations and biomarkers.
By standardizing around mCODE, researchers can pool data from different hospital systems to run massive real-world evidence (RWE) studies.
2. HL7 FHIR (Fast Healthcare Interoperability Resources)
FHIR is the modern internet protocol for healthcare data. If you are building an API that pulls data from a hospital system, you are likely using FHIR.
- How it works: FHIR represents healthcare data as discrete "Resources" (e.g.,
Patient,Observation,Condition,MedicationRequest) exposed via a RESTful API using JSON or XML. - In Oncology: mCODE is actually implemented as an "Implementation Guide" (IG) on top of FHIR. It takes generic FHIR resources and applies strict constraints (e.g., "An oncology
Conditionresource MUST have an SNOMED CT code representing a malignancy").
3. OMOP CDM (Observational Medical Outcomes Partnership)
While FHIR is designed for transactional interoperability (moving data between systems in real-time), the OMOP Common Data Model is designed for analytics and research.
- How it works: OMOP forces all observational data (claims, EHRs, registries) into a single, highly normalized relational database schema.
- Vocabulary mapping: It maps hundreds of different local coding systems into standard vocabularies.
- The OHDSI network: Institutions around the world map their data to OMOP. They don't share the raw data with each other; instead, they share the SQL queries. A researcher writes an algorithm once, and it can run on OMOP databases across dozens of hospitals globally, yielding federated insights without violating data privacy.
4. The Standard Vocabularies
When standardizing "what" the data means (semantics), technologists rely on massive controlled vocabularies:
- ICD-O-3 (International Classification of Diseases for Oncology): The gold standard for tumor location (Topography) and cell type (Morphology).
- SNOMED CT: The most comprehensive, multilingual clinical healthcare terminology in the world. Essential for encoding complex pathology findings.
- LOINC: The universal standard for identifying medical laboratory observations. (e.g., the specific code for a PD-L1 expression lab test).
- HGVS / ClinVar / OncoKB: Standards for representing genetic variants (e.g.,
BRAF V600E) and their clinical actionability.
5. The Engineering Challenge
The biggest challenge in oncology tech today is abstraction: converting the unstructured text of a PDF pathology report into structured mCODE/FHIR resources. Historically, this was done manually by clinical data abstractors (tumor registrars). Today, the vanguard of health-tech involves using Large Language Models (LLMs) and specialized Natural Language Processing (NLP) pipelines to automate this abstraction, extracting TNM stages, biomarker status, and treatment responses directly from the clinical narrative and outputting clean, standardized JSON.