Synthetic data for privacy-preserving clinical risk prediction

Area of Science:

• Medical Informatics
• Bioinformatics
• Health Data Science

Background:

• Synthetic data offers a privacy-preserving alternative for healthcare data sharing.
• Unlike federated learning, analyses on synthetic data require no downstream modifications.
• The full potential of synthetic data in clinical model development is not yet fully understood.

Purpose of the Study:

• To evaluate the utility of synthetic data in building lung cancer prognostic models.
• To assess the effectiveness of synthetic data throughout the entire medical prognostic modeling pipeline.
• To explore the impact of different data release strategies on synthetic biobank data deployment.

Main Methods:

• Utilized state-of-the-art, privacy-preserving generators to create synthetic UK Biobank data.
• Focused on a cohort of ever-smokers for lung cancer risk prediction.
• Developed prognostic models under various data release scenarios.

Main Results:

• Demonstrated that synthetic data can be effectively integrated into the medical prognostic modeling pipeline.
• Showed successful model development even without subsequent access to the original real data.
• Illustrated the consequences of different data release approaches on synthetic biobank data utilization.

Conclusions:

• Synthetic data is a viable tool for developing clinical prognostic models, ensuring privacy.
• The use of synthetic data can facilitate broader data sharing and research in healthcare.
• Strategic data release policies are crucial for optimizing the deployment of synthetic biobank data.