Synthetic data as external control arms in scarce single-arm clinical trials

Area of Science:

• Health Informatics
• Data Science
• Pharmacology

Background:

• Single-arm drug development studies often lack comparators.
• Health registry data offers an alternative but has access barriers.
• Clinical trial data is vulnerable to re-identification due to small sample sizes and high-cardinality features.

Purpose of the Study:

• To evaluate synthetic data as a privacy-preserving proxy for empirical external control arm data.
• To assess the utility and privacy risks of synthetic data in drug development.
• To introduce a reversible data generalization procedure for enhancing synthetic data generation.

Main Methods:

• A reversible data generalization procedure was proposed and applied.
• State-of-the-art synthetic data generators were tested.
• Synthetic data utility and privacy metrics were evaluated using Norwegian health registry data.

Main Results:

• Data generalization improved both utility and privacy for small, high-cardinality datasets.
• Generators produced high-utility synthetic data without compromising empirical data confidentiality.
• Synthetic external control arms demonstrated viability as an alternative to observational data.

Conclusions:

• Synthetic data, particularly with data generalization, can effectively serve as external control arms in drug development.
• This method enhances patient privacy and potentially accelerates drug approval processes.
• Synthetic data offers a secure and efficient alternative for comparative studies in pharmaceutical research.