Multi-site health research integrating complementary data sources: A scoping review of statistical inference methods...

Area of Science:

• Health Informatics
• Statistical Methodology
• Data Privacy

Background:

• Health research increasingly requires integrating diverse data sources for comprehensive analysis.
• Vertical methods allow statistical inference on distributed datasets without sharing individual-level data, crucial for sensitive health information.

Purpose of the Study:

• To identify existing vertical methods for statistical inference (confidence intervals, hypothesis testing).
• To characterize these methods' properties and their application in health data analysis.

Main Methods:

• A scoping review using PRISMA-ScR guidelines across four databases.
• Systematic extraction of method characteristics (comparability, communication efficiency, confidentiality).
• Screening of cited articles for real-world health data applications.

Main Results:

• 30 articles were included, mostly concerning health analytics.
• Linear and logistic regression inference were common; equivalence to pooled analysis was not consistently addressed.
• Most methods claimed privacy preservation, but few offered formal assessments; real-world health applications were scarce.

Conclusions:

• The range of vertical methods for statistical inference on partitioned data is limited.
• Existing methods often compromise between achieving pooled-analysis equivalence, communication efficiency, and data privacy.