OverviewOncology

Oncology Real-World Evidence using the OMOP CDM

Oncology is a part of medicine, and cancers are diseases. The OMOP CDM includes patient healthcare experiences for various diseases, but cancer presents specific challenges. It is not a single disease but differs by histology, location, stage, biomarkers, and other factors. Cancer evolves over time, influencing prognosis and complications, and is managed through complex treatment regimens.

The scientific questions or use cases, while potentially applicable to other areas, are particularly specific to cancer and often include timing between disease dynamics and interventions:

• Overall survival or mortality
• Time to metastatic disease
• Progression-free survival, event-free survival or time to local recurrence
• Time to treatment discontinuation
• Time to next line of treatment or regimen
• Predictive and prognostic value of biomarkers

Consequently, obtaining Real-World Evidence (RWE) in oncology presents significant challenges. There are three primary reasons for these difficulties:

1. Insufficient data in structured form: Information related to tumor histology, biomarkers, and stage is frequently recorded only in free text or unstructured formats within pathology or radiology reports. Treatment regimens are seldom documented explicitly and often need to be reconstructed based on drug exposures. For instance, identifying patients with metastatic adenocarcinoma of the lung and exon 14 skipping mutations as indicated for first-line capmatinib necessitates access to detailed information on histology, topography, biomarkers, regimen names, and treatment sequence.

2. Cancer is a rare condition: The complexity of cancer means each subgroup is considered a rare disease. For instance, lung cancer includes biologically distinct non-small cell lung cancer (NSCLC) and small cell lung cancer (SCLC), with NSCLC further divided into adeno, squamous cell, large cell and other rare carcinomas. These subtypes can occur in various lung regions and stages, with unique mutations or expression aberrations like EGFR, ALK fusions, KRAS G12C, and MET exon 14 skipping mutations. Each combination defines a clinically relevant subtype, making research questions like "What is the effectiveness of first-line capmatinib in metastatic lung adenocarcinoma with MET exon 14 skipping mutation?" relevant for only a few patients, even in large datasets.

3. Inadequate Cancer Terminologies: No single terminology fully captures all aspects of cancer. ICD-O-3 lacks details on stage, grade, and other attributes. SNOMED covers more but lacks precision. Biomarkers may be recorded in various systems like LOINC, NAACCR, proprietary coding systems or free text. Tumor attributes such as stage, grade, metastasis, and lymph node involvement are inconsistently represented. Treatment regimens lack a standardized vocabulary. Disease dynamics like remission or progression also lack standard representation. This results in fragmented vocabularies that fail to cover oncology's complexity.

Figure 1 shows the representation of metastatic NSCLC in various terminologies. Some vocabularies do not differentiate between NSCLC and SCLC, others lack hierarchical relationships, and most cannot combine all the required attributes to define clinically relevant cohorts. Table 1 highlights the inconsistent and scattered representation of lymph node involvement in CAP and NAACCR.

Figure 1. Inconsistent representation of metastatic NSCLC across terminologies

Category	Terms
Lymph node involvement	Nodal Status: para-aortic, mediastinal, pelvic, femora inguinal and distant (mediastinal, scalene) LN Status: femoral-inguinal, para-aortic, pelvic Clinical status of lymph node mets Clinical status of lymph nodes LN distant: mediastinal, scalene Adenopathy Nodal stations involved
Lymph node size	LN size Size of lymph nodes
Laterality	Laterality LN laterality Regional lymph nodes - laterality

Table 1. Inconsistent representation of lymph node involvement in CAP

OHDSI's Solution

The OHDSI Oncology Workgroup is collecting analytical use cases from the community to identify needed data elements and guide priorities for vocabulary maintenance and convention development. As a result, the Workgroup created a comprehensive data modeling convention within the OMOP CDM (1), consisting of three core components:

• Cancer disease model
• Cancer treatment model
• Cancer episode model

Cancer disease model

Cancers are represented in OMOP slightly differently from other conditions: Normally, conditions are fully pre-coordinated concepts, representing a diagnosis, sign or symptom into a single entity, which is stored in a record in the CONDITION_OCCURRENCE table. However, applying this approach to cancer would require encoding every possible combination of tumor attributes, resulting in a permutational explosion of concepts.

To avoid this, the representation of cancer in OMOP is distributed across three domains, requiring you to write records into three separate tables:

1. Cancer conditions in OMOP are confined only to pre-coordinated combinations of type of cancer (histology) and the location of the cancer (topography), to be represented through SNOMED or ICD-O-3.

2. Attributes such as metastases, lymph nodes, biomarkers, margins, stages, and grades are included in the Cancer Modifiers vocabulary. Document these attributes in the MEASUREMENT table and post-coordinate them with the corresponding CONDITION_OCCURRENCE record, provided that such a link is supported by the source data. Similarly, document somatic genomic variants in tumor tissue in the MEASUREMENT table using concepts from the OMOP Genomic vocabulary [cross link to chapter 7. Genomics Subchapter].

Figure 2. Cancer Disease model

3. Episodes: See below.

Cancer treatment model

Cancer treatment usually involves surgery or radiotherapy, or systemic therapies like chemotherapy, targeted therapy, and immunotherapy (2). These treatments often have complex dosing schedules, multiple agents, and repeated cycles (2). The cancer treatment model captures this complexity by tracking both composite Treatment Episodes and individual Drug Exposures or Procedures. This method allows the detailed representation of regimens, cycles, doses, and their sequence relative to other therapies. The HemOnc vocabulary in the OHDSI Standardized Vocabularies serves as the reference for regimens (2).

Cancer Episode model

A cancer patient's journey typically follows a sequence of disease states and treatments such as diagnosis, remission, progression, and changes in therapy. We called these Episodes, defined as time intervals grouping together conditions, treatments, modifiers, visits, and other clinical events. This is supported by two tables predominantly used in oncology data:

• The EPISODE table combines high level abstraction of the disease or treatment state with a specific cancer diagnosis or composite or individual treatment.

• The EPISODE_EVENT table links individual clinical events such as records from the CONDITION_OCCURENCE, DRUG_EXPOUSRE, or PROCEDURE_OCCUREENCE tables to their corresponding episode, preserving both detail and structure.

It is possible to nest Episodes. For instance, a disease episode can comprise various dynamic states like stable disease and subsequent progression. Similarly, a treatment episode may consist of multiple chemotherapy cycles.

There are three types of Disease Episodes:

• Episode of Care: Encompasses all events related to a primary cancer diagnosis and the full course of the disease.
• Dynamic Episode: Covers a period where the cancer remains in a consistent clinical state (remission, stable, or progression).
• Extent Episode: Describes the spread of cancer (confined, locally invasive, or metastatic).

Treatment Episode captures structured therapeutic interventions, including chemotherapy regimens, surgery, or radiotherapy.

Disease and Treatment Episodes are crucial for research but are often unstructured, except in tumor registries. Cancer episode derivation methods are improving, using rule-based and probabilistic approaches (3, 4). ARTEMIS detects treatment regimens from real-world data by comparing sequences of antineoplastic agents to standardized regimens in HemOnc. This method ensures scalable, accurate, and consistent identification of cancer treatments across various datasets (5).

References

1. Belenkaya R, Gurley MJ, Golozar A, Dymshyts D, Miller RT, Williams AE, et al. Extending the OMOP Common Data Model and Standardized Vocabularies to Support Observational Cancer Research. JCO Clin Cancer Inform [Internet]. 2021 Jan 7 [cited 2025 Jun 19];5:CCI.20.00079. Available from: <https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8140810/>

2. Warner JL, Dymshyts D, Reich CG, Gurley MJ, Hochheiser H, Moldwin ZH, et al. HemOnc: A new standard vocabulary for chemotherapy regimen representation in the OMOP common data model. J Biomed Inform. 2019 Aug;96:103239.

3. Verschueren MV, Abedian Kalkhoran H, Deenen M, van den Borne BEEM, Zwaveling J, Visser LE, et al. Development and Portability of a Text Mining Algorithm for Capturing Disease Progression in Electronic Health Records of Patients With Stage IV Non-Small Cell Lung Cancer. JCO Clin Cancer Inform. 2024 Oct;8:e2400053.

4. Carroll NM, Burniece KM, Holzman J, McQuillan DB, Plata A, Ritzwoller DP. Algorithm to Identify Systemic Cancer Therapy Treatment Using Structured Electronic Data. JCO Clin Cancer Inform. 2017 Nov;1:1-9.

5. Golozar A, Lawrence-Archer L, Zack T, Reich C, Warner JL. Introducing ARTEMIS: Advanced Regimen Detection Using an Adapted Smith-Waterman Algorithm - OHDSI [Internet]. 2023 [cited 2025 Jun 19]. Available from: https://www.ohdsi.org/2023showcase-416/