Best practices to design, plan, and execute large-scale federated analyses - key learnings and suggestions from a study comprising 52 databases.

Background and significance: Federated network studies allow data to remain locally while the research is conducted through sharing of analytical code and aggregated results across different healthcare settings and countries. A large number of databases have been mapped to the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM), boosting the use of analytical pipelines for standardized observational research within this open science framework. Transparency, reproducibility, and robustness of results have positioned federated analyses using the OMOP CDM within the European Health Data and Evidence Network (EHDEN) as an essential tool for generating large-scale evidence.

Objectives: We conducted large-scale federated analyses involving 52 databases from 19 countries using the OMOP CDM. In this State of the Art / Best practice article, we aimed to share key lessons and strategies for conducting such complex, large multi-database analyses. Learnings and suggestions: Meticulous planning, establishing a strong community of collaborators, efficient communication channels, standardized analytics, and strategic division of responsibilities are essential. We highlight the benefits of network engagement, cross-fertilization of ideas, and shared learning. Further key elements contributing to the study's success included an inclusive, incremental implementation of the analytical code, timely engagement of data partners, and community webinars to discuss and interpret study findings. We received predominantly positive feedback from data custodians about their participation, and included input for further improvements for future large-scale federated network studies from this shared learning experience.

Conclusion: Our learnings and suggestions aim to help other teams conduct large-scale multinational federated network studies efficiently and within a timely manner. The successful execution of such analyses, as demonstrated here, fostered positive experiences for data partners and stakeholders, encouraging future participation and contributing to sustainable large-scale evidence generation.