Privacy-preserving analysis of time-to-event data under nested case-control sampling.

Area of Science:

• Biostatistics
• Epidemiology
• Data Privacy

Background:

• Distributed data network analyses for rare diseases face privacy and ethical hurdles.
• Sharing individual participant records from recruiting sites is often infeasible.
• Existing privacy-preserving methods for time-to-event data are computationally intensive or inaccessible.

Purpose of the Study:

• To develop an easy-to-implement, privacy-preserving technique for Cox proportional hazards regression in distributed networks.
• To enable accurate analysis of rare disease data while protecting participant confidentiality.
• To overcome limitations of current privacy-preserving methods in applied research.

Main Methods:

• Proposed pooling individual covariate records at recruiting sites under a nested case-control sampling framework.
• Developed a method for generating pseudo-event times for pooled nested case-control subsamples.
• Validated the approach using extensive simulations and real-world data from the National Lung Screening Trial.

Main Results:

• Pooled hazard ratio estimators are maximum likelihood estimators and provide consistent estimates of full cohort hazard ratios.
• The proposed pooling technique achieves near-optimal performance, comparable to full cohort analysis and synthetic data.
• Efficiency gains were observed in rare event settings with increased control matching.

Conclusions:

• The proposed pooling method offers a practical and effective solution for privacy-preserving Cox regression in distributed rare disease data networks.
• This approach enhances data sharing feasibility and analytical accuracy, addressing critical ethical and logistical challenges.
• The method demonstrates significant utility and efficiency, particularly for rare disease research with limited data.