Secure distributed multiple imputation enables missing data inference for private data proprietors.

Area of Science:

• Health Informatics
• Computational Security
• Biostatistics

Background:

• Electronic Health Records (EHR) are vital for medical research but are often fragmented across institutions.
• Privacy concerns and data incompleteness hinder collaborative research using distributed EHRs.
• Current methods struggle to pool private data for comprehensive analysis and imputation.

Purpose of the Study:

• To develop a secure, privacy-preserving solution for collaborative analysis of distributed EHRs.
• To enable accurate statistical imputation of missing data in incomplete EHR datasets.
• To improve the classification of high-risk patient outcomes in intensive care settings.

Main Methods:

• Implementation of a provably secure solution using secure multiparty computation (SMC).
• Enabling distributed datasets to be utilized as a whole for imputation and collective studies.
• Testing the solution on synthetic and real-world datasets.

Main Results:

• The SMC-based solution achieves practical runtimes and accuracy comparable to non-secure methods.
• The approach effectively imputes missing data in distributed EHRs.
• Demonstrated significant improvement in classifying high-risk patient outcomes during ICU admission.

Conclusions:

• Secure multiparty computation offers a viable solution for privacy-preserving collaborative EHR research.
• The developed method overcomes limitations of data privacy and incompleteness.
• This facilitates more comprehensive studies and enhances clinical decision-making for patient outcomes.