Identifying patient-level data risks in trusted research environments: Worked examples with synthetic data

Area of Science:

• Health Informatics
• Data Security
• Clinical Research

Background:

• Trusted Research Environments (TREs) and Secure Data Environments (SDEs) are critical for accessing sensitive patient data.
• Unintended data egress poses a significant risk to patient privacy and data integrity.
• The PIONEER Hub in Acute Care provides a real-world context for studying these risks.

Purpose of the Study:

• To examine and illustrate real-world risks of unintended patient-level data egress from TREs and SDEs.
• To use synthetic data to recreate and analyze potential data egress scenarios.
• To inform best practices for data security within research environments.

Main Methods:

• Synthetic datasets with demographics and vital signs were generated using SciPy and NumPy.
• Data was transformed for machine learning and embedded into various formats to simulate egress.
• Three scenarios were explored: binary serialization of data, complex objects, and plain text reports.

Main Results:

• File size can be an initial indicator of potential patient-level data egress, prompting manual review.
• Standard data analysis tools (Python, R) do not inherently warn of hidden sensitive data.
• Manual inspection remains essential, even with automated checks, to identify and remove embedded data.

Conclusions:

• Identifying and preventing identifiable data egress from TREs is complex.
• Clear guidance for researchers and robust egress policies are necessary.
• Plain-text data exports facilitate easier verification compared to binary formats, which have security vulnerabilities.