Privacy-by-Design Approach to Generate Two Virtual Clinical Trials for Multiple Sclerosis and Release Them as Open...

Area of Science:

• Health Informatics
• Artificial Intelligence
• Clinical Trials

Background:

• Sharing individual patient data is restricted by privacy regulations.
• Generative AI offers a solution by creating virtual patient populations.
• Explicit privacy demonstration is crucial for synthetic data.

Purpose of the Study:

• Evaluate the
• avatars
• privacy-by-design technique for synthetic clinical trial data.
• Assess if synthetic datasets can replicate information from randomized clinical trials (RCTs).

Main Methods:

• Generated 2160 synthetic datasets from two Phase 3 multiple sclerosis RCTs.
• Optimized configurations for privacy (membership inference attacks) and utility (endpoint replication).
• Assessed fidelity via variable distributions and utility via endpoint effect directions and statistical significance.

Main Results:

• Achieved robust privacy and replicated primary endpoints in a significant portion of datasets.
• Utility varied, with some endpoints not fully captured.
• Selected datasets replicated all efficacy endpoints for placebo and treatment arms with satisfactory privacy.

Conclusions:

• Synthetic RCT datasets can replicate efficacy endpoints with explicit privacy.
• This method has potential to unlock health data sharing.
• Placebo arms from two RCTs were released as open datasets.