OBJECTIVES: The importance of real-world data to assess the performance of health technology is growing. Real-world data can indeed provide generalizable evidence, reflective of clinical practice. The IBM Marketscan Research Databases in the United States, the Clinical Practice Research Datalink (CPRD) database in the United Kingdom, and the national health insurance database (SNDS) in France are widely used to conduct observational studies. The objective of this study was to compare the three databases to help researchers identify the best source(s) of evidence depending on the study question and the challenges associated with the database.

METHODS: A list of criteria to be compared between the databases was established based on a targeted literature review on their strengths and limitations. The overview was completed based on interviews of users of at least one of these databases. A tabular summary of the findings was prepared and a synthesis was developed to support recommendations for future studies.

RESULTS: Major differences were observed in terms of data collection process, type of data collected, number of patients, coding systems used, and process and conditions to access the data. In particular, the Read codes are used in the CPRD GOLD database, whereas the IBM MarketScan Commercial Database and the SNDS database include diagnosis codes from the International Classification of Diseases (ICD), restricted to hospitalisations and patients with a chronic disease in the SNDS database. The main challenges identified included the identification of the study population, the understanding of the coding systems, and the non-availability of important data.

CONCLUSIONS: To our knowledge, this is the first study that directly compared three large real-world databases from three different countries to provide some guidance on how to deal with their specific challenges.