Individual Data Protected Integrative Regression Analysis of High-Dimensional Heterogeneous Data.

Area of Science:

• Biostatistics
• Computational Biology
• Health Informatics

Background:

• Meta-analysis enhances study precision and generalizability but faces challenges with ultra-high dimensional data.
• Integrating heterogeneous studies is complex, especially under DataSHIELD constraints preventing individual data sharing.

Purpose of the Study:

• To propose a novel integrative estimation procedure, SHIR (data-Shielding High-dimensional Integrative Regression), for ultra-high dimensional settings with DataSHIELD constraints.
• To develop a method that protects individual data while accommodating between-study heterogeneity and enabling consistent variable selection.

Main Methods:

• SHIR utilizes a summary-statistics-based integrating procedure to protect individual data privacy.
• The method accommodates heterogeneity in both covariate distribution and sparse regression model parameters across studies.
• SHIR is theoretically compared to existing distributed approaches, demonstrating superior statistical efficiency.

Main Results:

• SHIR achieves consistent variable selection in ultra-high dimensional settings.
• The estimation error from aggregated summary statistics is negligible, approaching the statistical minimax rate.
• SHIR is asymptotically equivalent to ideal estimators that would result from sharing all data.

Conclusions:

• SHIR offers a statistically efficient and privacy-preserving solution for integrative analysis of high-dimensional data across multiple studies.
• The method demonstrates superior performance compared to existing distributed approaches.
• SHIR's utility is validated through application to electronic health records for coronary artery disease phenotyping.