Organizing a Clinical Study Across Multiple Clinical Systems: Common Data Models

This chapter details approaches for replicating the results from Electronic Health Record (EHR) studies. The first common approach includes federated learning, whereby each institution keeps their own data locally, and then that data is queried using common frameworks including Common Data Models (CDM) and standard coding approaches via GitHub. The second approach involves data aggregation, whereby each institution contributes their data to a common repository, which then members of a collaborative are able to query across all institutions data simultaneously. The federated learning approach is useful for global studies whereby EHR regulations vary widely from country-to-country, and it enables each institution to maintain ownership and control over their data and access to those data. The aggregation approach is popular within a given country and often for a particular use-case. For example, the AllofUs study utilizes a data aggregation approach, as does the USA’s National COVID Cohort Collaborative (N3C). We describe both methods and how they can enhance the replication and reproducibility of research studies using EHR data. We cover a popular Common Data Model (CDM) for federated learning systems developed through the Observational Health Data Sciences & Informatics (OHDSI) consortium and how to utilize that structure. We also discuss nuances of EHR data and replication efforts, including the usage of diagnostic coding patterns that varies from institution to institution and country to country, even when standardized terminologies such as the International Classification of Diseases (ICD) are used. After reading this chapter, you should be able to confidently answer the following questions: