APHP / Echopen

Challenge

The APHP is collecting clinical data on patients with suspected Covid, which includes severity scores on eight lung fields after an ultrasound scan.
This data is valuable for assessing the severity level of patients with Covid, personalizing their care pathways, and relieving institutions facing a critical public health problem.

The aim? Create a Machine Learning algorithm to predict the severity level of patients and improve the quality of care while respecting patient privacy.

Solution

• Octopize's personal data anonymization solution allowed the exploitation of APHP data for secondary uses: feeding the Machine Learning algorithm, data sharing...
• Octopize sent a new set of synthetic and anonymous data (avatar data) and subsequently also made it possible to carry out a hackathon on this same data.
• This new data set allowed performances similar to the initial data while protecting the confidentiality of the patients behind this data.

Setting up

Benefit of anonymization service : Octopize moved to the APHP premises to anonymize personal data directly on its servers.

Duration : 2 days.

Maintaining statistical quality & utility

The prediction model trained on synthetic avatar data is on average 7% more efficient than the prediction model trained on the original data.

ROI

• Save lives in a case of public health crisis (Covid-19 pandemic)
• Optimization of flow management Emergencies
• Improvement and acceleration of researching