Oxford Summer School 2025: Real World Evidence using the OMOP Common Data Model

David Chibuike Ikechi Akwuru, Arya C Soman

We are delighted to share a key milestone in our research journey, which was made possible through the generous support of the eHealth-Hub for Cancer, funded by the Higher Education Authority's (HEA) North-South Research Programme. This support enabled us to participate in the Oxford Summer School on Real World Evidence Using the OMOP Common Data Model, held at the University of Oxford from June 23rd to 27th, 2025 . This was a remarkable opportunity that significantly enriched our knowledge and research capabilities.

Hosted at the historic St. Hilda's College, the summer school was focused on generating real-world evidence by harmonising diverse, real-world data sources. The core of the programme was the standardisation of health data using the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM), to enable large-scale federated analysis. The week brought together a diverse group of researchers, data scientists, policymakers, clinicians, and industry delegates from over five countries, including Ireland, Spain, Kenya, Portugal, Peru, Brazil, France, and the United Kingdom. The global collaboration and commitment to adopting the OMOP CDM across different countries was truly inspirational to witness.

A Photo of St. Hilda's College, University of Oxford.

Curriculum Overview

The Oxford summer school offered a well-balanced mix of morning lectures and afternoon practical sessions.

Morning Lectures

Held in the Jacqueline du Pré Music Building at St. Hilda's College, the morning sessions ran from 9:00 AM to 12:30 PM and introduced several aspects of the process of RWE generation, from data discovery to causal inference and target trial emulation.

A Photo of the morning lectures captured at the Jacqueline du Pré Music Building in St. Hildah College, with over 30 participants from more than 5 countries.

The week began with a lecture from Professor Dani Prieto Alhambra, who introduced the concepts of real-world data and real-world evidence. He highlighted the exponential growth of data from disconnected sources like primary care electronic health records, hospital data, registries, and biobanks, making a compelling case for the need to connect them. By linking these sources, we can derive evidence about what has happened to patients (clinical characterisation), what will happen to them (patient-level prediction), and the causal effects of interventions (population-level effect estimation). Professor Prieto Alhambra illustrated these points with relevant examples, such as the CPRD GOLD database, which includes over 21 million patients across the UK, and the DARWIN EU network, which connects data from over 30 partners in 16 European countries, covering approximately 160 million patients (1, 2). These examples demonstrated how harmonizing healthcare data with the OMOP CDM can generate reliable real-world evidence for regulatory decision-making.

Edward Burn further delivered a lecture on the OMOP CDM, explaining its origin, evolution, and importance in a federated data setting. He detailed the various tables within the model, including the standardised vocabularies (3). A particularly useful insight for our research on cancer data was understanding that patients can have multiple, overlapping observation periods, which reshaped our perspective on data standardisation.

Throughout the week, the morning lectures covered other key topics, including:

Study Design: Delivered by Mario Sanchez-Santhos, this session covered cohort and case-control studies (4). A key takeaway was that cohort studies select patients based on exposure and are excellent for tracking rare exposures to establish that the exposure preceded the outcome. In contrast, case-control studies select patients based on their outcome, which is efficient for rare or chronic diseases but can be subject to recall bias.

Phenotyping: Albert Prats-Uribe from the Botnar Research Centre explained the complex process of identifying patients with specific diseases, treatments, or characteristics across different datasets. An intriguing lesson was that there is no single "gold standard" for defining a phenotype, and for diseases like cancer, staging is a crucial component. The session highlighted the advantages of using standardised tools like ATLAS to build consistent phenotypes across federated networks. A key insight was that phenotypes are essentially time-bound cohorts, and the DARWIN EU process emphasizes transparency and reproducibility to ensure they accurately represent the intended patient populations (5).

Descriptive Epidemiology: Dr. D. Newby covered the fundamental concepts of incidence and prevalence, using practical examples from COVID-19 surveillance and cancer registry analysis.

Drug Utilisation Studies: Dr. Annika Jödicke and Martí Català Sabaté discussed how these studies help us understand how medicines are prescribed and used in the real world (6). They are essential for answering questions about who receives a drug, for what indication, at what dose, for how long, and in combination with other medicines.

Causal Inference: Delivered by Martí Català Sabaté, and Xintong Li from the Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences (NDORMS), the lectures provided an in-depth review of key concepts in this field. They covered methods to address confounding by indication, the use of propensity scores with inverse probability of treatment weighting (IPTW), marginal structural models for time-varying data, and negative controls with empirical calibration to handle unobserved bias (7, 8). These methods are foundational for statistical models used in comparative effectiveness research, especially when analysing non-randomised data.

The week concluded with a focus on health economics and two insightful sessions from Prof. Sonia Hernández-Díaz of Harvard University. Her talks covered:

Target Trial Emulation (TTE): A framework for designing observational studies to simulate randomised clinical trials (9).

Pregnancy Pharmacoepidemiology: An exploration of complex exposure-outcome timing relationships (6).

It concluded on synthesizing many of the statistical and design concepts emphasized throughout the week, highlighting the necessity for clarity and precision when developing policy-relevant real-world evidence.

Afternoon Sessions

The afternoon sessions, held from 1:30 PM to 4:45 PM, were dedicated to hands-on learning, allowing us to apply the concepts from the morning lectures. The practicals were split into two tracks: an epidemiology track using the EHDEN portal for study design and an analytical track for Health Data Scientists (HDS). We split into both groups to gain a comprehensive understanding of the tools available.

Caption of practical session across the Analytical and Epidemiological track.

In the analytic track, we started by setting up our RStudio environment and downloading synthetic standardised datasets. While we didn't map raw data to the OMOP CDM, we explored the types of analyses that can be run once data is standardised. We worked with a suite of open-source R packages to turn data into evidence.

We began with CdmConnector, which provides an interface for OMOP CDM databases, and OmopSketch, which characterises and visualises the data, providing a detailed snapshot that includes analysis dates and CDM versioning (10, 11). Using packages like CodelistGenerator, CohortConstructor, and PhenotypeR, we gained hands-on experience developing cohorts and phenotypes (12-14). CodelistGenerator was used to create lists of medical codes to define our study populations and then used these to build several types of cohorts including ConceptCohorts, MeasurementCohorts, DeathCohort and DemographicCohorts. A particularly useful tool was PhenotypeR, which provides a Shiny app interface to visualise the cohorts we created.

To understand how medicines are used in practice, we explored the DrugUtilisation package (15). This tool analyses prescribing patterns, treatment persistence, and dosing strategies. A key insight from this session was the importance of properly defining a "drug era" by setting a gap parameter to determine how many days can pass between prescriptions before they are considered separate treatment episodes.

The CohortCharacteristics package introduced us to an essential tool that automates the creation of baseline summary tables, allowing for quick and clear comparisons between patient groups (16).

We also explored the CohortSurvival package to study survival outcomes within our defined treatment cohorts (17). Finally, to measure the overall burden of a disease, we looked at the IncidencePrevalence package. It provides the fundamental epidemiological measures of incidence (new cases) and prevalence (existing cases), allowing us to track how common a condition is across different age groups, genders, and time periods.

Beyond the Classroom: Networking and Collaboration

The summer school also provided structured social activities, including punting on the river and formal dinners, which were excellent opportunities for networking and collaboration. These informal settings allowed us to discuss research methodologies and share experiences with our peers. Connecting with researchers at similar stages of their OMOP CDM implementation journey was particularly valuable. We shared practical insights on everything from technical tools like DBT, Docker, Apache NiFi, R, Python to common challenges and effective solutions. These conversations also introduced us to new tools, such as MedCat, which extracts information from electronic health records and links it to biomedical ontologies (18). We found this interesting and relevant to our research, as it can help us automate data structuring and develop a domain-specific ontology.

A group photo from punting and dinner section. Building connections with researchers.

We also had the opportunity to speak with the instructors and experts. A discussion with Professor Dani Prieto Alhambra and Edward Burn helped us refine our approach to mapping oncology data. We learned that there is no "one-size-fits-all" ETL (Extract, Transform, Load) script; it must be tailored to the specific needs of the data and the institution. Edward also stressed the critical importance of producing high-quality mapping, as it directly influences the validity of all subsequent analyses.

Building connection with Instructors

(David Akwuru, Professor Dani Prieto Alhambra and Arya Soman)

Application to Our Research

As PhD researchers at the eHealth Hub at the University of Limerick, our work focuses on developing frameworks to standardise and share cancer data for federated analysis. This summer school experience was transformative, significantly expanding our understanding of the analyses possible with standardised datasets.

For instance, one research area involves creating a framework for the federated sharing of Acute Myeloid Leukemia (AML) data across Ireland. The summer school provided invaluable insights into best practices for data mapping, particularly for clinical tables such as observation periods, condition occurrences, and drug exposures, and important analysis that can be performed. Understanding the need for drug eras and multiple observation periods proved especially relevant for a complex disease like AML. These insights are crucial for mapping the distinct cycles of treatment, remission, and relapse to the episode table, which helps capture the entire patient journey accurately (19).

Another research focus includes developing a statistical infrastructure to support federated analysis across clinical cancer data institutions. The Oxford Summer School provided invaluable hands-on exposure to OMOP-compatible tools, primarily cohort construction, data conversion, and study characterisation, all foundations to building scalable and reproducible analytical pipelines. This opportunity developed a comprehensive understanding on how standardisation enables insightful statistical modelling where patient-level evidence cannot be aggregated, a core issue within federated research. Practical tool exposure and design methodologies cemented key statistical concepts, study design, descriptive epidemiology, and causal inference, all vital to preserving analytical validity and reproducibility across distributed networks.

The exposure to a wide range of analytical tools will also greatly impact the federated aspect of our project. Federated studies require distributing analytical software across multiple partner institutions. The knowledge gained from these established packages means we don't have to "reinvent the wheel." Instead, we can use validated, reproducible tools, ensuring consistency, reliability, and efficiency in our research.

Conclusion

The Oxford Summer School provided us with transformational insights into real-world evidence generation and the implementation of the OMOP CDM. We return to our research with a deeper understanding of high-quality data mapping, study design, and the wide range of open-source analytical tools available to us. The hands-on experience and the connections we made with international researchers and experts will undoubtedly enhance the quality and impact of our work, streamlining our research and enriching our capabilities in clinical evidence generation.

None of this would have been possible without the generous support of the Higher Education Authority's (HEA) North-South Research Programme (NSRP). We are incredibly grateful for this funding and extend our special thanks to the eHealth-Hub for Cancer for providing us with this invaluable opportunity.

References

1. Sanchez-Santos, M.T., et al., Data Resource Profile Update: CPRD GOLD. Int J Epidemiol, 2025. 54(4).

2. Raventós, B. and D. Prieto-Alhambra, Real-world evidence for regulatory purposes: The example of DARWIN EU®. Farmacia Hospitalaria, 2025. 49(2): p. 62-64.

3. Reich, C., et al., OHDSI Standardized Vocabularies-a large-scale centralized reference ontology for international data harmonization. J Am Med Inform Assoc, 2024. 31(3): p. 583-590.

4. Prieto-Alhambra, D., et al., HIV Infection and Its Association With an Excess Risk of Clinical Fractures: A Nationwide Case–Control Study. JAIDS Journal of Acquired Immune Deficiency Syndromes, 2014. 66(1): p. 90-95.

5. Dernie, F., et al., Standardised and Reproducible Phenotyping Using Distributed Analytics and Tools in the Data Analysis and Real World Interrogation Network (DARWIN EU). Pharmacoepidemiol Drug Saf, 2024. 33(11): p. e70042.

6. Huybrechts, K.F., B.T. Bateman, and S. Hernández-Díaz, Use of real-world evidence from healthcare utilization data to evaluate drug safety during pregnancy. Pharmacoepidemiol Drug Saf, 2019. 28(7): p. 906-922.

7. Ali, M.S., et al., Propensity Score Methods in Health Technology Assessment: Principles, Extended Applications, and Recent Advances. Frontiers in Pharmacology, 2019. Volume 10 - 2019.

8. Tian, Y., M.J. Schuemie, and M.A. Suchard, Evaluating large-scale propensity score performance through real-world and synthetic data experiments. Int J Epidemiol, 2018. 47(6): p. 2005-2014.

9. Hernán, M.A., et al., Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses. J Clin Epidemiol, 2016. 79: p. 70-75.

10. Alcalde-Herraiz, M., et al. OmopSketch: Characterise Tables of an OMOP Common Data Model Instance. 2025 2025; Available from: https://OHDSI.github.io/OmopSketch/.

11. Black, A., et al. CDMConnector: Connect to an OMOP Common Data Model. Available from: https://darwin-eu.github.io/CDMConnector/.

12. Burn, E., et al. CohortConstructor: Build and Manipulate Study Cohorts Using a Common Data Model. 2025; Available from: https://ohdsi.github.io/CohortConstructor/.

13. Burn, E., X. Chen, and N. Mercade-Besora. CodelistGenerator: Identify Relevant Clinical Codes and Evaluate Their Use. 2025; Available from: https://darwin-eu.github.io/CodelistGenerator/.

14. Burn, E., et al. PhenotypeR: Assess Study Cohorts Using a Common Data Model. 2025; Available from: https://ohdsi.github.io/PhenotypeR/.

15. Wang, S.V. and A. Pottegård, Advancing Research Transparency and Reproducibility in Pharmacoepidemiology. Pharmacoepidemiology and Drug Safety, 2025. 34(2): p. e70096.

16. Catala, M., et al. CohortCharacteristics: Summarise and Visualise Characteristics of Patients in the OMOP CDM. 2025; Available from: https://darwin-eu.github.io/CohortCharacteristics/.

17. López-Güell, K., et al. CohortSurvival: Estimate Survival from Common Data Model Cohorts. 2025; Available from: https://darwin-eu-dev.github.io/CohortSurvival/.

18. Kraljevic, Z., et al., Multi-domain clinical natural language processing with MedCAT: the medical concept annotation toolkit. Artificial intelligence in medicine, 2021. 117: p. 102083.

19. Carus, J., et al., Mapping Cancer Registry Data to the Episode Domain of the Observational Medical Outcomes Partnership Model (OMOP). Applied Sciences, 2022. 12(8): p. 4010.