The French branch of the Swiss laboratory Roche and the University Hospital Center (CHU) of Brest are joining forces with Rennes-based startup Octopize (formerly WeData) to reuse medical data. This collaboration is part of the Coalition Next initiative, which brings together manufacturers, healthcare providers and patient associations.
Use data without exposing it
As part of the design competition, Octopize presented its Anonymization Without Compromise project, designed to address the challenges of digitalization of clinical trials. He was seduced by the University Hospital of Brest and Roche. The experiment aims to demonstrate that “avatarization technology allows the generation of anonymous summary data that retains the statistical quality of the data while maintaining the privacy of individuals,” the partners explain.
“We are often asked to communicate patient cohort data in the context of clinical data. But, given the laws, it is often impossible to do this without contacting the patients and without obtaining their written consent, ”explains Alain Sarauks. -President for scientific work of the Brest University Hospital.
As a reminder, health data is personal data within the meaning of the General Data Protection Regulation (GDPR). Given their very high level of sensitivity, their re-use is subject to very strict regulations that can become an obstacle in the field of medical research. That’s why Octopize came up with the idea to develop a technology for anonymizing health data – based on a machine learning system – without losing its information potential.
An important clarification: Octopize technology allows you to anonymize data, and not just pseudonymize it. Anonymization makes it impossible to identify a person from a data set, while pseudonymization simply consists in replacing the directly identifying data (last name, first name, etc.) of a data set with indirect identifiers. This is a reversible process.
“Having available data similar enough to our cohorts of Breton patients, but without any possibility of identifying them (…), would be the ideal solution,” said Alain Sarah. times gives the same result every time.