Sherbrooke University
Sharing health data publicly to facilitate scientific research
Challenges
- Identify patients at the end of life to refer them to discussions about goals of care with their clinicians.
- Develop a solution that uses longitudinal data patients for predicting their mortality risk within one year after they were admitted to hospital.
- Sharing data publicly to allow the reproduction of the experiments carried out.
- Insure the patient confidentiality : sharing data without compromising privacy.
- Preserve the data quality : maintaining the predictive capacity of our solution.
- Comply with strict regulations regarding the use of health data.
Results
- Contribute to the Open Science :
- Publication of a research article: https://doi.org/10.1101/2024.06.21.24309191
- Sharing Data in confidence and without compromising on confidentiality: https://doi.org/10.5281/zenodo.12954672
- Publication of a short report comparing synthetic and real data: https://doi.org/10.21203/rs.3.rs-5363467/v1
- Code sharing to replicate research experiments: https://github.com/MEDomics-UdeS/POYM
- Trade-off between the usefulness of predictive modeling and maintaining the confidentiality of original data
Maintaining statistical quality and utility
- AUROC on all the last patient visits
.png)
“Thanks to Octopize's synthetic data creation methods, we were able to obtain the approval of the ethics committee of the CIUSSS de l'Estrie — CHUS to share this data online and promote open science.” - Martin Vallières, Professor, Department of Computer Science @Université de Sherbrooke
“The avatar solution made it possible to generate a synthetic data set with characteristics very similar to the original data. The metrics on different patient groups were similar to those in the original data. In addition, the privacy metrics had exceeded the required thresholds, confirming the anonymity of the data. We are very happy with the data generated and thank the team for the iterations leading up to this final product!” - Hakima Laribi, Ph.D. Informatics @Université from Sherbrooke
~ 250,000 publicly shared patient visit data
Our customers
All our customers
.png){kind=logo}
