BioMérieux

Challenge

• A health care institution has established a clinical trial on the past hospitalizations of 17843 patients with acute kidney injuries. bioMérieux, the pharmaceutical industry, wanted develop a new treatment protocol using this data.
• Problems : it is a retrospective cohort (dating back 10 years) which complicates the collection of new consent. In addition, this data had to be anonymized to be shared to a service provider outside the EU (invalidation of Privacy Shield).
• The aim? Quickly access the informative value of data to develop a new care protocol.

Solution

• Octopize's personal data anonymization solution made it possible to unleash data sharing for secondary uses.
• Here the personal data of bioMérieux (European pharmaceutical industry) are blocked by the regulatory context and their anonymization allows them to be shared and used by a provider located in the USA.
• To allow this sharing, Octopize has transmitted a new game by synthetic and anonymous data (avatar data) that helped protect the privacy the individuals behind this data.
• In addition, the informative quality of the initial data was maintained in order to allow research with similar performances to the initial data without risking privacy.

Setting up

Anonymization service provided by Octopize remotely.

Maintaining statistical quality & utility

• Original data on the left, avatar data on the right. Patients are represented according to their reason for inclusion (5 subgroups)
• The characteristics of the patients by cluster are respected → data can be used at the multifactorial scale for advanced analyses

The ML model trained on avatar data has the same prediction performance as the model trained on the original data.

How do I read this graph? 100 iterations for each model. Each point represents the prediction score of a model (0.76 = 76% good prediction) 50% of the values of the models are included in the box (see blue and gray rectangle) and the median of the performance of the models is represented by the vertical bar in the box. So original median = 0.76 and avatar median = 0.76

‍

ROI

• Share data to a subsidiary in the USA while respecting the regulatory context
• Accelerate research deadlines and develop a new treatment protocol
• Access the informative value of the data without compromising patient privacy

Additional resource

Find the Replay of our webinar with bioMérieux on data transfers using avatar anonymization technology with experts:

• Julien Textoris, Vice President & EME Medical Affairs at @BioMerieux, shares the challenges they faced in the data sharing between the EU and the US.
• Delphine Ganoote-Mary, Associate Lawyer at @Clairmont Novus Avocats, presents the legal & regulatory context of these data transfers.
• Olivier Breillacq, CEO of Octopize, explains the advantages of the solution ofAvatar anonymization - summary data - in these use cases.
• Finally, to open the subject to other perspectives, Thomas Wentz, Responsible AI and AI Act Consulting at @Speredia, introduces tensions between the AI Act and the GDPR.