Project Victoria: A pragmatic data model to automate RWE generation from the national French claims database

Area of Science:

• Health Informatics
• Epidemiology
• Data Science

Background:

• The French National Health Insurance Fund for Employed Persons (SNDS) database contains extensive health data.
• Existing data pipelines may lack scalability and adaptability for large datasets and evolving research needs.
• Efficient and compliant data access is crucial for epidemiological research.

Purpose of the Study:

• To develop and describe Victoria, an automated and scalable data pipeline for the SNDS.
• To facilitate scientific and epidemiological research by providing clear processes and documented data.
• To ensure compliant and eased access to SNDS data.

Main Methods:

• Victoria employs a two-step process: initial data cleaning and creation of two linearised data models.
• Data cleaning involves formatting, removing errors/duplicates, and standardizing variables.
• Two models are generated: an epidemiological model for phenotyping and a medico-economic model for cost analysis.

Main Results:

• The pipeline was successfully executed on datasets of approximately 85,000 and 870,000 beneficiaries.
• Total execution times were 25 hours and 96 hours, respectively, with data cleaning being efficient (4 hours).
• The epidemiological and medico-economic models were generated, with processing times varying based on dataset size.

Conclusions:

• The Victoria pipeline is a successful implementation for processing SNDS data.
• Its design emphasizes reviewability, with integrated unit tests and quality assessments.
• The pipeline has supported published studies and is adaptable for future SNDS format changes and platform integration.