Limoges University Hospital
Accelerate research and publish sensitive health data in a compliant manner
Challenges
- Reproduce the results of pharmacogenetic analyses on small data sets while maintaining patient privacy.
- Maintain the fidelity of statistical models (selection of the same key variables and estimation of effects) with synthetic data.
- Improve the stability of results by generating multiple sets of synthetic data and aggregating performances.
Maintaining statistical quality and usefulness
Key variable preservation (haplotype)
ROI
- Scientific publication in the journal CPT PSP (IF 3) “To be or not to be, when synthetic data meets clinical pharmacology: A focused study on pharmacogenetics”
- Academic validation of the Avatar method: demonstration of its ability to combine strong data fidelity and privacy protection.
- Data sharing and collaboration made possible between centers in the context of multicenter studies.
- Avoid the legal complexities, costs, and lengthy procedures associated with traditional data sharing
“The creation of synthetic data promotes open science by providing data at the same time as the results. It facilitates the sharing of data between centers as an alternative to federated learning. This helps avoid the legal complexities, costs, and time-consuming procedures often associated with traditional data sharing.” - Jean-Baptiste Woillard - Professor/Hospital Practitioner Pharmacology @CHU de Limoges
“The variability within a bootstrap dataset is similar to that obtained by launching avatar several times.” - Jean-Baptiste Woillard - Professor/Hospital Practitioner Pharmacology @CHU from Limoges
Our customers
All our customers.png){kind=logo}
