A data-driven approach to choosing privacy parameters for clinical trial data sharing under differential privacy.

Area of Science:

• Medical research
• Data privacy
• Statistical analysis

Background:

• Clinical trial data sharing is vital for research transparency and collaboration.
• Differential privacy (DP) is a state-of-the-art anonymization technique balancing data privacy and accuracy using a privacy budget (ε).
• DP is currently underutilized in clinical trial data sharing.

Purpose of the Study:

• To identify appropriate privacy budget (ε) values for sharing clinical trial data using differential privacy.
• To evaluate the accuracy of DP-protected data compared to original clinical trial datasets.

Main Methods:

• Analyzed two clinical trial datasets using DP with privacy budgets (ε) from 0.01 to 10.
• Compared key statistical metrics (rates, odds ratios, means) between original and DP-estimated data.
• Investigated the impact of increasing ε on data accuracy and privacy.

Main Results:

• DP-estimated rates closely matched original rates (6.5%) when ε > 1.
• DP-estimated odds ratios aligned with original ratios (0.689) when ε ≥ 3.
• DP-estimated means approximated original means (164.64) when ε ≥ 1, with convergence observed as ε increased to 5.

Conclusions:

• Differential privacy demonstrates significant promise for secure and accurate clinical trial data sharing.
• The study provides insights into selecting privacy budget (ε) values for DP application in clinical research.
• Further research is needed to establish consensus on choosing ε for optimal privacy-utility balance.