Multiple imputation for analysis of incomplete data in distributed health data networks

Area of Science:

• Health Informatics
• Biostatistics
• Data Science

Background:

• Distributed health data networks (DHDNs) are gaining traction for collaborative research without sharing sensitive patient data.
• Handling missing data is a significant challenge in DHDNs, as traditional methods require centralized data pooling.
• Existing approaches are not suitable for the distributed nature of DHDNs, hindering analysis.

Purpose of the Study:

• To develop novel methods for addressing missing data in horizontally partitioned distributed health data networks.
• To propose communication-efficient distributed multiple imputation techniques suitable for DHDNs.
• To enhance patient privacy and public trust in health data analysis within distributed settings.

Main Methods:

• Developed communication-efficient distributed multiple imputation methods for horizontally partitioned data.
• Ensured subject-level data remain within each site, preventing data sharing or transfer.
• Evaluated method performance through extensive simulation studies.

Main Results:

• The proposed methods effectively handle missing data in distributed environments without centralizing patient information.
• Simulations demonstrated the efficacy and communication efficiency of the developed techniques.
• The methods were successfully applied to a real-world acute stroke dataset.

Conclusions:

• The novel distributed multiple imputation methods offer a viable solution for analyzing incomplete data in DHDNs.
• These methods enhance data privacy and security, crucial for sensitive health information.
• The approach facilitates collaboration and strengthens trust in distributed health data analysis.